• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * Copyright © 2014 Connor Abbott
3  *
4  * Permission is hereby granted, free of charge, to any person obtaining a
5  * copy of this software and associated documentation files (the "Software"),
6  * to deal in the Software without restriction, including without limitation
7  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8  * and/or sell copies of the Software, and to permit persons to whom the
9  * Software is furnished to do so, subject to the following conditions:
10  *
11  * The above copyright notice and this permission notice (including the next
12  * paragraph) shall be included in all copies or substantial portions of the
13  * Software.
14  *
15  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
18  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
21  * IN THE SOFTWARE.
22  *
23  * Authors:
24  *    Connor Abbott (cwabbott0@gmail.com)
25  *
26  */
27 
28 #ifndef NIR_H
29 #define NIR_H
30 
31 #include "util/hash_table.h"
32 #include "compiler/glsl/list.h"
33 #include "util/list.h"
34 #include "util/log.h"
35 #include "util/ralloc.h"
36 #include "util/set.h"
37 #include "util/bitscan.h"
38 #include "util/bitset.h"
39 #include "util/compiler.h"
40 #include "util/enum_operators.h"
41 #include "util/macros.h"
42 #include "util/format/u_format.h"
43 #include "compiler/nir_types.h"
44 #include "compiler/shader_enums.h"
45 #include "compiler/shader_info.h"
46 #define XXH_INLINE_ALL
47 #include "util/xxhash.h"
48 #include <stdio.h>
49 
50 #ifndef NDEBUG
51 #include "util/debug.h"
52 #endif /* NDEBUG */
53 
54 #include "nir_opcodes.h"
55 
56 #if defined(_WIN32) && !defined(snprintf)
57 #define snprintf _snprintf
58 #endif
59 
60 #ifdef __cplusplus
61 extern "C" {
62 #endif
63 
64 extern uint32_t nir_debug;
65 extern bool nir_debug_print_shader[MESA_SHADER_KERNEL + 1];
66 
67 #ifndef NDEBUG
68 #define NIR_DEBUG(flag) unlikely(nir_debug & (NIR_DEBUG_ ## flag))
69 #else
70 #define NIR_DEBUG(flag) false
71 #endif
72 
73 #define NIR_DEBUG_CLONE                  (1u << 0)
74 #define NIR_DEBUG_SERIALIZE              (1u << 1)
75 #define NIR_DEBUG_NOVALIDATE             (1u << 2)
76 #define NIR_DEBUG_VALIDATE_SSA_DOMINANCE (1u << 3)
77 #define NIR_DEBUG_TGSI                   (1u << 4)
78 #define NIR_DEBUG_PRINT_VS               (1u << 5)
79 #define NIR_DEBUG_PRINT_TCS              (1u << 6)
80 #define NIR_DEBUG_PRINT_TES              (1u << 7)
81 #define NIR_DEBUG_PRINT_GS               (1u << 8)
82 #define NIR_DEBUG_PRINT_FS               (1u << 9)
83 #define NIR_DEBUG_PRINT_CS               (1u << 10)
84 #define NIR_DEBUG_PRINT_TS               (1u << 11)
85 #define NIR_DEBUG_PRINT_MS               (1u << 12)
86 #define NIR_DEBUG_PRINT_RGS              (1u << 13)
87 #define NIR_DEBUG_PRINT_AHS              (1u << 14)
88 #define NIR_DEBUG_PRINT_CHS              (1u << 15)
89 #define NIR_DEBUG_PRINT_MHS              (1u << 16)
90 #define NIR_DEBUG_PRINT_IS               (1u << 17)
91 #define NIR_DEBUG_PRINT_CBS              (1u << 18)
92 #define NIR_DEBUG_PRINT_KS               (1u << 19)
93 #define NIR_DEBUG_PRINT_CONSTS           (1u << 20)
94 #define NIR_DEBUG_VALIDATE_GC_LIST       (1u << 21)
95 
96 #define NIR_DEBUG_PRINT (NIR_DEBUG_PRINT_VS  | \
97                          NIR_DEBUG_PRINT_TCS | \
98                          NIR_DEBUG_PRINT_TES | \
99                          NIR_DEBUG_PRINT_GS  | \
100                          NIR_DEBUG_PRINT_FS  | \
101                          NIR_DEBUG_PRINT_CS  | \
102                          NIR_DEBUG_PRINT_TS  | \
103                          NIR_DEBUG_PRINT_MS  | \
104                          NIR_DEBUG_PRINT_RGS | \
105                          NIR_DEBUG_PRINT_AHS | \
106                          NIR_DEBUG_PRINT_CHS | \
107                          NIR_DEBUG_PRINT_MHS | \
108                          NIR_DEBUG_PRINT_IS  | \
109                          NIR_DEBUG_PRINT_CBS | \
110                          NIR_DEBUG_PRINT_KS)
111 
112 #define NIR_FALSE 0u
113 #define NIR_TRUE (~0u)
114 #define NIR_MAX_VEC_COMPONENTS 16
115 #define NIR_MAX_MATRIX_COLUMNS 4
116 #define NIR_STREAM_PACKED (1 << 8)
117 typedef uint16_t nir_component_mask_t;
118 
119 static inline bool
nir_num_components_valid(unsigned num_components)120 nir_num_components_valid(unsigned num_components)
121 {
122    return (num_components >= 1  &&
123            num_components <= 5) ||
124            num_components == 8  ||
125            num_components == 16;
126 }
127 
128 static inline nir_component_mask_t
nir_component_mask(unsigned num_components)129 nir_component_mask(unsigned num_components)
130 {
131    assert(nir_num_components_valid(num_components));
132    return (1u << num_components) - 1;
133 }
134 
135 void
136 nir_process_debug_variable(void);
137 
138 bool nir_component_mask_can_reinterpret(nir_component_mask_t mask,
139                                         unsigned old_bit_size,
140                                         unsigned new_bit_size);
141 nir_component_mask_t
142 nir_component_mask_reinterpret(nir_component_mask_t mask,
143                                unsigned old_bit_size,
144                                unsigned new_bit_size);
145 
146 /** Defines a cast function
147  *
148  * This macro defines a cast function from in_type to out_type where
149  * out_type is some structure type that contains a field of type out_type.
150  *
151  * Note that you have to be a bit careful as the generated cast function
152  * destroys constness.
153  */
154 #define NIR_DEFINE_CAST(name, in_type, out_type, field, \
155                         type_field, type_value)         \
156 static inline out_type *                                \
157 name(const in_type *parent)                             \
158 {                                                       \
159    assert(parent && parent->type_field == type_value);  \
160    return exec_node_data(out_type, parent, field);      \
161 }
162 
163 struct nir_function;
164 struct nir_shader;
165 struct nir_instr;
166 struct nir_builder;
167 struct nir_xfb_info;
168 
169 
170 /**
171  * Description of built-in state associated with a uniform
172  *
173  * \sa nir_variable::state_slots
174  */
175 typedef struct {
176    gl_state_index16 tokens[STATE_LENGTH];
177    uint16_t swizzle;
178 } nir_state_slot;
179 
180 typedef enum {
181    nir_var_system_value    = (1 << 0),
182    nir_var_uniform         = (1 << 1),
183    nir_var_shader_in       = (1 << 2),
184    nir_var_shader_out      = (1 << 3),
185    nir_var_image           = (1 << 4),
186    /** Incoming call or ray payload data for ray-tracing shaders */
187    nir_var_shader_call_data = (1 << 5),
188    /** Ray hit attributes */
189    nir_var_ray_hit_attrib  = (1 << 6),
190 
191    /* Modes named nir_var_mem_* have explicit data layout */
192    nir_var_mem_ubo          = (1 << 7),
193    nir_var_mem_push_const   = (1 << 8),
194    nir_var_mem_ssbo         = (1 << 9),
195    nir_var_mem_constant     = (1 << 10),
196    nir_var_mem_task_payload = (1 << 11),
197 
198    /* Generic modes intentionally come last. See encode_dref_modes() in
199     * nir_serialize.c for more details.
200     */
201    nir_var_shader_temp     = (1 << 12),
202    nir_var_function_temp   = (1 << 13),
203    nir_var_mem_shared      = (1 << 14),
204    nir_var_mem_global      = (1 << 15),
205 
206    nir_var_mem_generic     = (nir_var_shader_temp |
207                               nir_var_function_temp |
208                               nir_var_mem_shared |
209                               nir_var_mem_global),
210 
211    nir_var_read_only_modes = nir_var_shader_in | nir_var_uniform |
212                              nir_var_system_value | nir_var_mem_constant |
213                              nir_var_mem_ubo,
214    /** Modes where vector derefs can be indexed as arrays */
215    nir_var_vec_indexable_modes = nir_var_mem_ubo | nir_var_mem_ssbo |
216                                  nir_var_mem_shared | nir_var_mem_global |
217                                  nir_var_mem_push_const,
218    nir_num_variable_modes  = 16,
219    nir_var_all             = (1 << nir_num_variable_modes) - 1,
220 } nir_variable_mode;
221 MESA_DEFINE_CPP_ENUM_BITFIELD_OPERATORS(nir_variable_mode)
222 
223 /**
224  * Rounding modes.
225  */
226 typedef enum {
227    nir_rounding_mode_undef = 0,
228    nir_rounding_mode_rtne  = 1, /* round to nearest even */
229    nir_rounding_mode_ru    = 2, /* round up */
230    nir_rounding_mode_rd    = 3, /* round down */
231    nir_rounding_mode_rtz   = 4, /* round towards zero */
232 } nir_rounding_mode;
233 
234 typedef union {
235    bool b;
236    float f32;
237    double f64;
238    int8_t i8;
239    uint8_t u8;
240    int16_t i16;
241    uint16_t u16;
242    int32_t i32;
243    uint32_t u32;
244    int64_t i64;
245    uint64_t u64;
246 } nir_const_value;
247 
248 #define nir_const_value_to_array(arr, c, components, m) \
249 { \
250    for (unsigned i = 0; i < components; ++i) \
251       arr[i] = c[i].m; \
252 } while (false)
253 
254 static inline nir_const_value
nir_const_value_for_raw_uint(uint64_t x,unsigned bit_size)255 nir_const_value_for_raw_uint(uint64_t x, unsigned bit_size)
256 {
257    nir_const_value v;
258    memset(&v, 0, sizeof(v));
259 
260    switch (bit_size) {
261    case 1:  v.b   = x;  break;
262    case 8:  v.u8  = x;  break;
263    case 16: v.u16 = x;  break;
264    case 32: v.u32 = x;  break;
265    case 64: v.u64 = x;  break;
266    default:
267       unreachable("Invalid bit size");
268    }
269 
270    return v;
271 }
272 
273 static inline nir_const_value
nir_const_value_for_int(int64_t i,unsigned bit_size)274 nir_const_value_for_int(int64_t i, unsigned bit_size)
275 {
276    nir_const_value v;
277    memset(&v, 0, sizeof(v));
278 
279    assert(bit_size <= 64);
280    if (bit_size < 64) {
281       assert(i >= (-(1ll << (bit_size - 1))));
282       assert(i < (1ll << (bit_size - 1)));
283    }
284 
285    return nir_const_value_for_raw_uint(i, bit_size);
286 }
287 
288 static inline nir_const_value
nir_const_value_for_uint(uint64_t u,unsigned bit_size)289 nir_const_value_for_uint(uint64_t u, unsigned bit_size)
290 {
291    nir_const_value v;
292    memset(&v, 0, sizeof(v));
293 
294    assert(bit_size <= 64);
295    if (bit_size < 64)
296       assert(u < (1ull << bit_size));
297 
298    return nir_const_value_for_raw_uint(u, bit_size);
299 }
300 
301 static inline nir_const_value
nir_const_value_for_bool(bool b,unsigned bit_size)302 nir_const_value_for_bool(bool b, unsigned bit_size)
303 {
304    /* Booleans use a 0/-1 convention */
305    return nir_const_value_for_int(-(int)b, bit_size);
306 }
307 
308 /* This one isn't inline because it requires half-float conversion */
309 nir_const_value nir_const_value_for_float(double b, unsigned bit_size);
310 
311 static inline int64_t
nir_const_value_as_int(nir_const_value value,unsigned bit_size)312 nir_const_value_as_int(nir_const_value value, unsigned bit_size)
313 {
314    switch (bit_size) {
315    /* int1_t uses 0/-1 convention */
316    case 1:  return -(int)value.b;
317    case 8:  return value.i8;
318    case 16: return value.i16;
319    case 32: return value.i32;
320    case 64: return value.i64;
321    default:
322       unreachable("Invalid bit size");
323    }
324 }
325 
326 static inline uint64_t
nir_const_value_as_uint(nir_const_value value,unsigned bit_size)327 nir_const_value_as_uint(nir_const_value value, unsigned bit_size)
328 {
329    switch (bit_size) {
330    case 1:  return value.b;
331    case 8:  return value.u8;
332    case 16: return value.u16;
333    case 32: return value.u32;
334    case 64: return value.u64;
335    default:
336       unreachable("Invalid bit size");
337    }
338 }
339 
340 static inline bool
nir_const_value_as_bool(nir_const_value value,unsigned bit_size)341 nir_const_value_as_bool(nir_const_value value, unsigned bit_size)
342 {
343    int64_t i = nir_const_value_as_int(value, bit_size);
344 
345    /* Booleans of any size use 0/-1 convention */
346    assert(i == 0 || i == -1);
347 
348    return i;
349 }
350 
351 /* This one isn't inline because it requires half-float conversion */
352 double nir_const_value_as_float(nir_const_value value, unsigned bit_size);
353 
354 typedef struct nir_constant {
355    /**
356     * Value of the constant.
357     *
358     * The field used to back the values supplied by the constant is determined
359     * by the type associated with the \c nir_variable.  Constants may be
360     * scalars, vectors, or matrices.
361     */
362    nir_const_value values[NIR_MAX_VEC_COMPONENTS];
363 
364    /* we could get this from the var->type but makes clone *much* easier to
365     * not have to care about the type.
366     */
367    unsigned num_elements;
368 
369    /* Array elements / Structure Fields */
370    struct nir_constant **elements;
371 } nir_constant;
372 
373 /**
374  * \brief Layout qualifiers for gl_FragDepth.
375  *
376  * The AMD/ARB_conservative_depth extensions allow gl_FragDepth to be redeclared
377  * with a layout qualifier.
378  */
379 typedef enum {
380     nir_depth_layout_none, /**< No depth layout is specified. */
381     nir_depth_layout_any,
382     nir_depth_layout_greater,
383     nir_depth_layout_less,
384     nir_depth_layout_unchanged
385 } nir_depth_layout;
386 
387 /**
388  * Enum keeping track of how a variable was declared.
389  */
390 typedef enum {
391    /**
392     * Normal declaration.
393     */
394    nir_var_declared_normally = 0,
395 
396    /**
397     * Variable is implicitly generated by the compiler and should not be
398     * visible via the API.
399     */
400    nir_var_hidden,
401 } nir_var_declaration_type;
402 
403 /**
404  * Either a uniform, global variable, shader input, or shader output. Based on
405  * ir_variable - it should be easy to translate between the two.
406  */
407 
408 typedef struct nir_variable {
409    struct exec_node node;
410 
411    /**
412     * Declared type of the variable
413     */
414    const struct glsl_type *type;
415 
416    /**
417     * Declared name of the variable
418     */
419    char *name;
420 
421    struct nir_variable_data {
422       /**
423        * Storage class of the variable.
424        *
425        * \sa nir_variable_mode
426        */
427       unsigned mode:16;
428 
429       /**
430        * Is the variable read-only?
431        *
432        * This is set for variables declared as \c const, shader inputs,
433        * and uniforms.
434        */
435       unsigned read_only:1;
436       unsigned centroid:1;
437       unsigned sample:1;
438       unsigned patch:1;
439       unsigned invariant:1;
440 
441       /**
442        * Is the variable a ray query?
443        */
444       unsigned ray_query:1;
445 
446      /**
447        * Precision qualifier.
448        *
449        * In desktop GLSL we do not care about precision qualifiers at all, in
450        * fact, the spec says that precision qualifiers are ignored.
451        *
452        * To make things easy, we make it so that this field is always
453        * GLSL_PRECISION_NONE on desktop shaders. This way all the variables
454        * have the same precision value and the checks we add in the compiler
455        * for this field will never break a desktop shader compile.
456        */
457       unsigned precision:2;
458 
459       /**
460        * Has this variable been statically assigned?
461        *
462        * This answers whether the variable was assigned in any path of
463        * the shader during ast_to_hir.  This doesn't answer whether it is
464        * still written after dead code removal, nor is it maintained in
465        * non-ast_to_hir.cpp (GLSL parsing) paths.
466        */
467       unsigned assigned:1;
468 
469       /**
470        * Can this variable be coalesced with another?
471        *
472        * This is set by nir_lower_io_to_temporaries to say that any
473        * copies involving this variable should stay put. Propagating it can
474        * duplicate the resulting load/store, which is not wanted, and may
475        * result in a load/store of the variable with an indirect offset which
476        * the backend may not be able to handle.
477        */
478       unsigned cannot_coalesce:1;
479 
480       /**
481        * When separate shader programs are enabled, only input/outputs between
482        * the stages of a multi-stage separate program can be safely removed
483        * from the shader interface. Other input/outputs must remains active.
484        *
485        * This is also used to make sure xfb varyings that are unused by the
486        * fragment shader are not removed.
487        */
488       unsigned always_active_io:1;
489 
490       /**
491        * Interpolation mode for shader inputs / outputs
492        *
493        * \sa glsl_interp_mode
494        */
495       unsigned interpolation:3;
496 
497       /**
498        * If non-zero, then this variable may be packed along with other variables
499        * into a single varying slot, so this offset should be applied when
500        * accessing components.  For example, an offset of 1 means that the x
501        * component of this variable is actually stored in component y of the
502        * location specified by \c location.
503        */
504       unsigned location_frac:2;
505 
506       /**
507        * If true, this variable represents an array of scalars that should
508        * be tightly packed.  In other words, consecutive array elements
509        * should be stored one component apart, rather than one slot apart.
510        */
511       unsigned compact:1;
512 
513       /**
514        * Whether this is a fragment shader output implicitly initialized with
515        * the previous contents of the specified render target at the
516        * framebuffer location corresponding to this shader invocation.
517        */
518       unsigned fb_fetch_output:1;
519 
520       /**
521        * Non-zero if this variable is considered bindless as defined by
522        * ARB_bindless_texture.
523        */
524       unsigned bindless:1;
525 
526       /**
527        * Was an explicit binding set in the shader?
528        */
529       unsigned explicit_binding:1;
530 
531       /**
532        * Was the location explicitly set in the shader?
533        *
534        * If the location is explicitly set in the shader, it \b cannot be changed
535        * by the linker or by the API (e.g., calls to \c glBindAttribLocation have
536        * no effect).
537        */
538       unsigned explicit_location:1;
539 
540       /**
541        * Is this varying used by transform feedback?
542        *
543        * This is used by the linker to decide if it's safe to pack the varying.
544        */
545       unsigned is_xfb:1;
546 
547       /**
548        * Is this varying used only by transform feedback?
549        *
550        * This is used by the linker to decide if its safe to pack the varying.
551        */
552       unsigned is_xfb_only:1;
553 
554       /**
555        * Was a transfer feedback buffer set in the shader?
556        */
557       unsigned explicit_xfb_buffer:1;
558 
559       /**
560        * Was a transfer feedback stride set in the shader?
561        */
562       unsigned explicit_xfb_stride:1;
563 
564       /**
565        * Was an explicit offset set in the shader?
566        */
567       unsigned explicit_offset:1;
568 
569       /**
570        * Layout of the matrix.  Uses glsl_matrix_layout values.
571        */
572       unsigned matrix_layout:2;
573 
574       /**
575        * Non-zero if this variable was created by lowering a named interface
576        * block.
577        */
578       unsigned from_named_ifc_block:1;
579 
580       /**
581        * Non-zero if the variable must be a shader input. This is useful for
582        * constraints on function parameters.
583        */
584       unsigned must_be_shader_input:1;
585 
586       /**
587        * How the variable was declared.  See nir_var_declaration_type.
588        *
589        * This is used to detect variables generated by the compiler, so should
590        * not be visible via the API.
591        */
592       unsigned how_declared:2;
593 
594       /**
595        * Is this variable per-view?  If so, we know it must be an array with
596        * size corresponding to the number of views.
597        */
598       unsigned per_view:1;
599 
600       /**
601        * Whether the variable is per-primitive.
602        * Can be use by Mesh Shader outputs and corresponding Fragment Shader inputs.
603        */
604       unsigned per_primitive:1;
605 
606       /**
607        * \brief Layout qualifier for gl_FragDepth. See nir_depth_layout.
608        *
609        * This is not equal to \c ir_depth_layout_none if and only if this
610        * variable is \c gl_FragDepth and a layout qualifier is specified.
611        */
612       unsigned depth_layout:3;
613 
614       /**
615        * Vertex stream output identifier.
616        *
617        * For packed outputs, NIR_STREAM_PACKED is set and bits [2*i+1,2*i]
618        * indicate the stream of the i-th component.
619        */
620       unsigned stream:9;
621 
622       /**
623        * See gl_access_qualifier.
624        *
625        * Access flags for memory variables (SSBO/global), image uniforms, and
626        * bindless images in uniforms/inputs/outputs.
627        */
628       unsigned access:9;
629 
630       /**
631        * Descriptor set binding for sampler or UBO.
632        */
633       unsigned descriptor_set:5;
634 
635       /**
636        * output index for dual source blending.
637        */
638       unsigned index;
639 
640       /**
641        * Initial binding point for a sampler or UBO.
642        *
643        * For array types, this represents the binding point for the first element.
644        */
645       unsigned binding;
646 
647       /**
648        * Storage location of the base of this variable
649        *
650        * The precise meaning of this field depends on the nature of the variable.
651        *
652        *   - Vertex shader input: one of the values from \c gl_vert_attrib.
653        *   - Vertex shader output: one of the values from \c gl_varying_slot.
654        *   - Geometry shader input: one of the values from \c gl_varying_slot.
655        *   - Geometry shader output: one of the values from \c gl_varying_slot.
656        *   - Fragment shader input: one of the values from \c gl_varying_slot.
657        *   - Fragment shader output: one of the values from \c gl_frag_result.
658        *   - Task shader output: one of the values from \c gl_varying_slot.
659        *   - Mesh shader input: one of the values from \c gl_varying_slot.
660        *   - Mesh shader output: one of the values from \c gl_varying_slot.
661        *   - Uniforms: Per-stage uniform slot number for default uniform block.
662        *   - Uniforms: Index within the uniform block definition for UBO members.
663        *   - Non-UBO Uniforms: uniform slot number.
664        *   - Other: This field is not currently used.
665        *
666        * If the variable is a uniform, shader input, or shader output, and the
667        * slot has not been assigned, the value will be -1.
668        */
669       int location;
670 
671       /**
672        * The actual location of the variable in the IR. Only valid for inputs,
673        * outputs, uniforms (including samplers and images), and for UBO and SSBO
674        * variables in GLSL.
675        */
676       unsigned driver_location;
677 
678       /**
679        * Location an atomic counter or transform feedback is stored at.
680        */
681       unsigned offset;
682 
683       union {
684          struct {
685             /** Image internal format if specified explicitly, otherwise PIPE_FORMAT_NONE. */
686             enum pipe_format format;
687          } image;
688 
689          struct {
690             /**
691              * For OpenCL inline samplers. See cl_sampler_addressing_mode and cl_sampler_filter_mode
692              */
693             unsigned is_inline_sampler : 1;
694             unsigned addressing_mode : 3;
695             unsigned normalized_coordinates : 1;
696             unsigned filter_mode : 1;
697          } sampler;
698 
699          struct {
700             /**
701              * Transform feedback buffer.
702              */
703             uint16_t buffer:2;
704 
705             /**
706              * Transform feedback stride.
707              */
708             uint16_t stride;
709          } xfb;
710       };
711    } data;
712 
713    /**
714     * Identifier for this variable generated by nir_index_vars() that is unique
715     * among other variables in the same exec_list.
716     */
717    unsigned index;
718 
719    /* Number of nir_variable_data members */
720    uint16_t num_members;
721 
722    /**
723     * Built-in state that backs this uniform
724     *
725     * Once set at variable creation, \c state_slots must remain invariant.
726     * This is because, ideally, this array would be shared by all clones of
727     * this variable in the IR tree.  In other words, we'd really like for it
728     * to be a fly-weight.
729     *
730     * If the variable is not a uniform, \c num_state_slots will be zero and
731     * \c state_slots will be \c NULL.
732     */
733    /*@{*/
734    uint16_t num_state_slots;    /**< Number of state slots used */
735    nir_state_slot *state_slots;  /**< State descriptors. */
736    /*@}*/
737 
738    /**
739     * Constant expression assigned in the initializer of the variable
740     *
741     * This field should only be used temporarily by creators of NIR shaders
742     * and then nir_lower_variable_initializers can be used to get rid of them.
743     * Most of the rest of NIR ignores this field or asserts that it's NULL.
744     */
745    nir_constant *constant_initializer;
746 
747    /**
748     * Global variable assigned in the initializer of the variable
749     * This field should only be used temporarily by creators of NIR shaders
750     * and then nir_lower_variable_initializers can be used to get rid of them.
751     * Most of the rest of NIR ignores this field or asserts that it's NULL.
752     */
753    struct nir_variable *pointer_initializer;
754 
755    /**
756     * For variables that are in an interface block or are an instance of an
757     * interface block, this is the \c GLSL_TYPE_INTERFACE type for that block.
758     *
759     * \sa ir_variable::location
760     */
761    const struct glsl_type *interface_type;
762 
763    /**
764     * Description of per-member data for per-member struct variables
765     *
766     * This is used for variables which are actually an amalgamation of
767     * multiple entities such as a struct of built-in values or a struct of
768     * inputs each with their own layout specifier.  This is only allowed on
769     * variables with a struct or array of array of struct type.
770     */
771    struct nir_variable_data *members;
772 } nir_variable;
773 
774 static inline bool
_nir_shader_variable_has_mode(nir_variable * var,unsigned modes)775 _nir_shader_variable_has_mode(nir_variable *var, unsigned modes)
776 {
777    /* This isn't a shader variable */
778    assert(!(modes & nir_var_function_temp));
779    return var->data.mode & modes;
780 }
781 
782 #define nir_foreach_variable_in_list(var, var_list) \
783    foreach_list_typed(nir_variable, var, node, var_list)
784 
785 #define nir_foreach_variable_in_list_safe(var, var_list) \
786    foreach_list_typed_safe(nir_variable, var, node, var_list)
787 
788 #define nir_foreach_variable_in_shader(var, shader) \
789    nir_foreach_variable_in_list(var, &(shader)->variables)
790 
791 #define nir_foreach_variable_in_shader_safe(var, shader) \
792    nir_foreach_variable_in_list_safe(var, &(shader)->variables)
793 
794 #define nir_foreach_variable_with_modes(var, shader, modes) \
795    nir_foreach_variable_in_shader(var, shader) \
796       if (_nir_shader_variable_has_mode(var, modes))
797 
798 #define nir_foreach_variable_with_modes_safe(var, shader, modes) \
799    nir_foreach_variable_in_shader_safe(var, shader) \
800       if (_nir_shader_variable_has_mode(var, modes))
801 
802 #define nir_foreach_shader_in_variable(var, shader) \
803    nir_foreach_variable_with_modes(var, shader, nir_var_shader_in)
804 
805 #define nir_foreach_shader_in_variable_safe(var, shader) \
806    nir_foreach_variable_with_modes_safe(var, shader, nir_var_shader_in)
807 
808 #define nir_foreach_shader_out_variable(var, shader) \
809    nir_foreach_variable_with_modes(var, shader, nir_var_shader_out)
810 
811 #define nir_foreach_shader_out_variable_safe(var, shader) \
812    nir_foreach_variable_with_modes_safe(var, shader, nir_var_shader_out)
813 
814 #define nir_foreach_uniform_variable(var, shader) \
815    nir_foreach_variable_with_modes(var, shader, nir_var_uniform)
816 
817 #define nir_foreach_uniform_variable_safe(var, shader) \
818    nir_foreach_variable_with_modes_safe(var, shader, nir_var_uniform)
819 
820 #define nir_foreach_image_variable(var, shader) \
821    nir_foreach_variable_with_modes(var, shader, nir_var_image)
822 
823 #define nir_foreach_image_variable_safe(var, shader) \
824    nir_foreach_variable_with_modes_safe(var, shader, nir_var_image)
825 
826 static inline bool
nir_variable_is_global(const nir_variable * var)827 nir_variable_is_global(const nir_variable *var)
828 {
829    return var->data.mode != nir_var_function_temp;
830 }
831 
832 typedef struct nir_register {
833    struct exec_node node;
834 
835    unsigned num_components; /** < number of vector components */
836    unsigned num_array_elems; /** < size of array (0 for no array) */
837 
838    /* The bit-size of each channel; must be one of 8, 16, 32, or 64 */
839    uint8_t bit_size;
840 
841    /**
842     * True if this register may have different values in different SIMD
843     * invocations of the shader.
844     */
845    bool divergent;
846 
847    /** generic register index. */
848    unsigned index;
849 
850    /** set of nir_srcs where this register is used (read from) */
851    struct list_head uses;
852 
853    /** set of nir_dests where this register is defined (written to) */
854    struct list_head defs;
855 
856    /** set of nir_ifs where this register is used as a condition */
857    struct list_head if_uses;
858 } nir_register;
859 
860 #define nir_foreach_register(reg, reg_list) \
861    foreach_list_typed(nir_register, reg, node, reg_list)
862 #define nir_foreach_register_safe(reg, reg_list) \
863    foreach_list_typed_safe(nir_register, reg, node, reg_list)
864 
865 typedef enum PACKED {
866    nir_instr_type_alu,
867    nir_instr_type_deref,
868    nir_instr_type_call,
869    nir_instr_type_tex,
870    nir_instr_type_intrinsic,
871    nir_instr_type_load_const,
872    nir_instr_type_jump,
873    nir_instr_type_ssa_undef,
874    nir_instr_type_phi,
875    nir_instr_type_parallel_copy,
876 } nir_instr_type;
877 
878 typedef struct nir_instr {
879    struct exec_node node;
880    struct list_head gc_node;
881    struct nir_block *block;
882    nir_instr_type type;
883 
884    /* A temporary for optimization and analysis passes to use for storing
885     * flags.  For instance, DCE uses this to store the "dead/live" info.
886     */
887    uint8_t pass_flags;
888 
889    /** generic instruction index. */
890    uint32_t index;
891 } nir_instr;
892 
893 static inline nir_instr *
nir_instr_next(nir_instr * instr)894 nir_instr_next(nir_instr *instr)
895 {
896    struct exec_node *next = exec_node_get_next(&instr->node);
897    if (exec_node_is_tail_sentinel(next))
898       return NULL;
899    else
900       return exec_node_data(nir_instr, next, node);
901 }
902 
903 static inline nir_instr *
nir_instr_prev(nir_instr * instr)904 nir_instr_prev(nir_instr *instr)
905 {
906    struct exec_node *prev = exec_node_get_prev(&instr->node);
907    if (exec_node_is_head_sentinel(prev))
908       return NULL;
909    else
910       return exec_node_data(nir_instr, prev, node);
911 }
912 
913 static inline bool
nir_instr_is_first(const nir_instr * instr)914 nir_instr_is_first(const nir_instr *instr)
915 {
916    return exec_node_is_head_sentinel(exec_node_get_prev_const(&instr->node));
917 }
918 
919 static inline bool
nir_instr_is_last(const nir_instr * instr)920 nir_instr_is_last(const nir_instr *instr)
921 {
922    return exec_node_is_tail_sentinel(exec_node_get_next_const(&instr->node));
923 }
924 
925 typedef struct nir_ssa_def {
926    /** Instruction which produces this SSA value. */
927    nir_instr *parent_instr;
928 
929    /** set of nir_instrs where this register is used (read from) */
930    struct list_head uses;
931 
932    /** set of nir_ifs where this register is used as a condition */
933    struct list_head if_uses;
934 
935    /** generic SSA definition index. */
936    unsigned index;
937 
938    uint8_t num_components;
939 
940    /* The bit-size of each channel; must be one of 8, 16, 32, or 64 */
941    uint8_t bit_size;
942 
943    /**
944     * True if this SSA value may have different values in different SIMD
945     * invocations of the shader.  This is set by nir_divergence_analysis.
946     */
947    bool divergent;
948 } nir_ssa_def;
949 
950 struct nir_src;
951 
952 typedef struct {
953    nir_register *reg;
954    struct nir_src *indirect; /** < NULL for no indirect offset */
955    unsigned base_offset;
956 
957    /* TODO use-def chain goes here */
958 } nir_reg_src;
959 
960 typedef struct {
961    nir_instr *parent_instr;
962    struct list_head def_link;
963 
964    nir_register *reg;
965    struct nir_src *indirect; /** < NULL for no indirect offset */
966    unsigned base_offset;
967 
968    /* TODO def-use chain goes here */
969 } nir_reg_dest;
970 
971 struct nir_if;
972 
973 typedef struct nir_src {
974    union {
975       /** Instruction that consumes this value as a source. */
976       nir_instr *parent_instr;
977       struct nir_if *parent_if;
978    };
979 
980    struct list_head use_link;
981 
982    union {
983       nir_reg_src reg;
984       nir_ssa_def *ssa;
985    };
986 
987    bool is_ssa;
988 } nir_src;
989 
990 static inline nir_src
nir_src_init(void)991 nir_src_init(void)
992 {
993    nir_src src = { { NULL } };
994    return src;
995 }
996 
997 #define NIR_SRC_INIT nir_src_init()
998 
999 #define nir_foreach_use(src, reg_or_ssa_def) \
1000    list_for_each_entry(nir_src, src, &(reg_or_ssa_def)->uses, use_link)
1001 
1002 #define nir_foreach_use_safe(src, reg_or_ssa_def) \
1003    list_for_each_entry_safe(nir_src, src, &(reg_or_ssa_def)->uses, use_link)
1004 
1005 #define nir_foreach_if_use(src, reg_or_ssa_def) \
1006    list_for_each_entry(nir_src, src, &(reg_or_ssa_def)->if_uses, use_link)
1007 
1008 #define nir_foreach_if_use_safe(src, reg_or_ssa_def) \
1009    list_for_each_entry_safe(nir_src, src, &(reg_or_ssa_def)->if_uses, use_link)
1010 
1011 typedef struct {
1012    union {
1013       nir_reg_dest reg;
1014       nir_ssa_def ssa;
1015    };
1016 
1017    bool is_ssa;
1018 } nir_dest;
1019 
1020 static inline nir_dest
nir_dest_init(void)1021 nir_dest_init(void)
1022 {
1023    nir_dest dest = { { { NULL } } };
1024    return dest;
1025 }
1026 
1027 #define NIR_DEST_INIT nir_dest_init()
1028 
1029 #define nir_foreach_def(dest, reg) \
1030    list_for_each_entry(nir_dest, dest, &(reg)->defs, reg.def_link)
1031 
1032 #define nir_foreach_def_safe(dest, reg) \
1033    list_for_each_entry_safe(nir_dest, dest, &(reg)->defs, reg.def_link)
1034 
1035 static inline nir_src
nir_src_for_ssa(nir_ssa_def * def)1036 nir_src_for_ssa(nir_ssa_def *def)
1037 {
1038    nir_src src = NIR_SRC_INIT;
1039 
1040    src.is_ssa = true;
1041    src.ssa = def;
1042 
1043    return src;
1044 }
1045 
1046 static inline nir_src
nir_src_for_reg(nir_register * reg)1047 nir_src_for_reg(nir_register *reg)
1048 {
1049    nir_src src = NIR_SRC_INIT;
1050 
1051    src.is_ssa = false;
1052    src.reg.reg = reg;
1053    src.reg.indirect = NULL;
1054    src.reg.base_offset = 0;
1055 
1056    return src;
1057 }
1058 
1059 static inline nir_dest
nir_dest_for_reg(nir_register * reg)1060 nir_dest_for_reg(nir_register *reg)
1061 {
1062    nir_dest dest = NIR_DEST_INIT;
1063 
1064    dest.reg.reg = reg;
1065 
1066    return dest;
1067 }
1068 
1069 static inline unsigned
nir_src_bit_size(nir_src src)1070 nir_src_bit_size(nir_src src)
1071 {
1072    return src.is_ssa ? src.ssa->bit_size : src.reg.reg->bit_size;
1073 }
1074 
1075 static inline unsigned
nir_src_num_components(nir_src src)1076 nir_src_num_components(nir_src src)
1077 {
1078    return src.is_ssa ? src.ssa->num_components : src.reg.reg->num_components;
1079 }
1080 
1081 static inline bool
nir_src_is_const(nir_src src)1082 nir_src_is_const(nir_src src)
1083 {
1084    return src.is_ssa &&
1085           src.ssa->parent_instr->type == nir_instr_type_load_const;
1086 }
1087 
1088 static inline bool
nir_src_is_undef(nir_src src)1089 nir_src_is_undef(nir_src src)
1090 {
1091    return src.is_ssa &&
1092           src.ssa->parent_instr->type == nir_instr_type_ssa_undef;
1093 }
1094 
1095 static inline bool
nir_src_is_divergent(nir_src src)1096 nir_src_is_divergent(nir_src src)
1097 {
1098    return src.is_ssa ? src.ssa->divergent : src.reg.reg->divergent;
1099 }
1100 
1101 static inline unsigned
nir_dest_bit_size(nir_dest dest)1102 nir_dest_bit_size(nir_dest dest)
1103 {
1104    return dest.is_ssa ? dest.ssa.bit_size : dest.reg.reg->bit_size;
1105 }
1106 
1107 static inline unsigned
nir_dest_num_components(nir_dest dest)1108 nir_dest_num_components(nir_dest dest)
1109 {
1110    return dest.is_ssa ? dest.ssa.num_components : dest.reg.reg->num_components;
1111 }
1112 
1113 static inline bool
nir_dest_is_divergent(nir_dest dest)1114 nir_dest_is_divergent(nir_dest dest)
1115 {
1116    return dest.is_ssa ? dest.ssa.divergent : dest.reg.reg->divergent;
1117 }
1118 
1119 /* Are all components the same, ie. .xxxx */
1120 static inline bool
nir_is_same_comp_swizzle(uint8_t * swiz,unsigned nr_comp)1121 nir_is_same_comp_swizzle(uint8_t *swiz, unsigned nr_comp)
1122 {
1123    for (unsigned i = 1; i < nr_comp; i++)
1124       if (swiz[i] != swiz[0])
1125          return false;
1126    return true;
1127 }
1128 
1129 /* Are all components sequential, ie. .yzw */
1130 static inline bool
nir_is_sequential_comp_swizzle(uint8_t * swiz,unsigned nr_comp)1131 nir_is_sequential_comp_swizzle(uint8_t *swiz, unsigned nr_comp)
1132 {
1133    for (unsigned i = 1; i < nr_comp; i++)
1134       if (swiz[i] != (swiz[0] + i))
1135          return false;
1136    return true;
1137 }
1138 
1139 void nir_src_copy(nir_src *dest, const nir_src *src);
1140 void nir_dest_copy(nir_dest *dest, const nir_dest *src);
1141 
1142 typedef struct {
1143    /** Base source */
1144    nir_src src;
1145 
1146    /**
1147     * \name input modifiers
1148     */
1149    /*@{*/
1150    /**
1151     * For inputs interpreted as floating point, flips the sign bit. For
1152     * inputs interpreted as integers, performs the two's complement negation.
1153     */
1154    bool negate;
1155 
1156    /**
1157     * Clears the sign bit for floating point values, and computes the integer
1158     * absolute value for integers. Note that the negate modifier acts after
1159     * the absolute value modifier, therefore if both are set then all inputs
1160     * will become negative.
1161     */
1162    bool abs;
1163    /*@}*/
1164 
1165    /**
1166     * For each input component, says which component of the register it is
1167     * chosen from.
1168     *
1169     * Note that which elements of the swizzle are used and which are ignored
1170     * are based on the write mask for most opcodes - for example, a statement
1171     * like "foo.xzw = bar.zyx" would have a writemask of 1101b and a swizzle
1172     * of {2, 1, x, 0} where x means "don't care."
1173     */
1174    uint8_t swizzle[NIR_MAX_VEC_COMPONENTS];
1175 } nir_alu_src;
1176 
1177 typedef struct {
1178    /** Base destination */
1179    nir_dest dest;
1180 
1181    /**
1182     * Saturate output modifier
1183     *
1184     * Only valid for opcodes that output floating-point numbers. Clamps the
1185     * output to between 0.0 and 1.0 inclusive.
1186     */
1187    bool saturate;
1188 
1189    /**
1190     * Write-mask
1191     *
1192     * Ignored if dest.is_ssa is true
1193     */
1194    unsigned write_mask : NIR_MAX_VEC_COMPONENTS;
1195 } nir_alu_dest;
1196 
1197 /** NIR sized and unsized types
1198  *
1199  * The values in this enum are carefully chosen so that the sized type is
1200  * just the unsized type OR the number of bits.
1201  */
1202 typedef enum PACKED {
1203    nir_type_invalid = 0, /* Not a valid type */
1204    nir_type_int =       2,
1205    nir_type_uint =      4,
1206    nir_type_bool =      6,
1207    nir_type_float =     128,
1208    nir_type_bool1 =     1  | nir_type_bool,
1209    nir_type_bool8 =     8  | nir_type_bool,
1210    nir_type_bool16 =    16 | nir_type_bool,
1211    nir_type_bool32 =    32 | nir_type_bool,
1212    nir_type_int1 =      1  | nir_type_int,
1213    nir_type_int8 =      8  | nir_type_int,
1214    nir_type_int16 =     16 | nir_type_int,
1215    nir_type_int32 =     32 | nir_type_int,
1216    nir_type_int64 =     64 | nir_type_int,
1217    nir_type_uint1 =     1  | nir_type_uint,
1218    nir_type_uint8 =     8  | nir_type_uint,
1219    nir_type_uint16 =    16 | nir_type_uint,
1220    nir_type_uint32 =    32 | nir_type_uint,
1221    nir_type_uint64 =    64 | nir_type_uint,
1222    nir_type_float16 =   16 | nir_type_float,
1223    nir_type_float32 =   32 | nir_type_float,
1224    nir_type_float64 =   64 | nir_type_float,
1225 } nir_alu_type;
1226 
1227 #define NIR_ALU_TYPE_SIZE_MASK 0x79
1228 #define NIR_ALU_TYPE_BASE_TYPE_MASK 0x86
1229 
1230 static inline unsigned
nir_alu_type_get_type_size(nir_alu_type type)1231 nir_alu_type_get_type_size(nir_alu_type type)
1232 {
1233    return type & NIR_ALU_TYPE_SIZE_MASK;
1234 }
1235 
1236 static inline nir_alu_type
nir_alu_type_get_base_type(nir_alu_type type)1237 nir_alu_type_get_base_type(nir_alu_type type)
1238 {
1239    return (nir_alu_type)(type & NIR_ALU_TYPE_BASE_TYPE_MASK);
1240 }
1241 
1242 nir_alu_type
1243 nir_get_nir_type_for_glsl_base_type(enum glsl_base_type base_type);
1244 
1245 static inline nir_alu_type
nir_get_nir_type_for_glsl_type(const struct glsl_type * type)1246 nir_get_nir_type_for_glsl_type(const struct glsl_type *type)
1247 {
1248    return nir_get_nir_type_for_glsl_base_type(glsl_get_base_type(type));
1249 }
1250 
1251 enum glsl_base_type
1252 nir_get_glsl_base_type_for_nir_type(nir_alu_type base_type);
1253 
1254 nir_op nir_type_conversion_op(nir_alu_type src, nir_alu_type dst,
1255                               nir_rounding_mode rnd);
1256 
1257 nir_op
1258 nir_op_vec(unsigned components);
1259 
1260 bool
1261 nir_op_is_vec(nir_op op);
1262 
1263 static inline bool
nir_is_float_control_signed_zero_inf_nan_preserve(unsigned execution_mode,unsigned bit_size)1264 nir_is_float_control_signed_zero_inf_nan_preserve(unsigned execution_mode, unsigned bit_size)
1265 {
1266     return (16 == bit_size && execution_mode & FLOAT_CONTROLS_SIGNED_ZERO_INF_NAN_PRESERVE_FP16) ||
1267         (32 == bit_size && execution_mode & FLOAT_CONTROLS_SIGNED_ZERO_INF_NAN_PRESERVE_FP32) ||
1268         (64 == bit_size && execution_mode & FLOAT_CONTROLS_SIGNED_ZERO_INF_NAN_PRESERVE_FP64);
1269 }
1270 
1271 static inline bool
nir_is_denorm_flush_to_zero(unsigned execution_mode,unsigned bit_size)1272 nir_is_denorm_flush_to_zero(unsigned execution_mode, unsigned bit_size)
1273 {
1274     return (16 == bit_size && execution_mode & FLOAT_CONTROLS_DENORM_FLUSH_TO_ZERO_FP16) ||
1275         (32 == bit_size && execution_mode & FLOAT_CONTROLS_DENORM_FLUSH_TO_ZERO_FP32) ||
1276         (64 == bit_size && execution_mode & FLOAT_CONTROLS_DENORM_FLUSH_TO_ZERO_FP64);
1277 }
1278 
1279 static inline bool
nir_is_denorm_preserve(unsigned execution_mode,unsigned bit_size)1280 nir_is_denorm_preserve(unsigned execution_mode, unsigned bit_size)
1281 {
1282     return (16 == bit_size && execution_mode & FLOAT_CONTROLS_DENORM_PRESERVE_FP16) ||
1283         (32 == bit_size && execution_mode & FLOAT_CONTROLS_DENORM_PRESERVE_FP32) ||
1284         (64 == bit_size && execution_mode & FLOAT_CONTROLS_DENORM_PRESERVE_FP64);
1285 }
1286 
1287 static inline bool
nir_is_rounding_mode_rtne(unsigned execution_mode,unsigned bit_size)1288 nir_is_rounding_mode_rtne(unsigned execution_mode, unsigned bit_size)
1289 {
1290     return (16 == bit_size && execution_mode & FLOAT_CONTROLS_ROUNDING_MODE_RTE_FP16) ||
1291         (32 == bit_size && execution_mode & FLOAT_CONTROLS_ROUNDING_MODE_RTE_FP32) ||
1292         (64 == bit_size && execution_mode & FLOAT_CONTROLS_ROUNDING_MODE_RTE_FP64);
1293 }
1294 
1295 static inline bool
nir_is_rounding_mode_rtz(unsigned execution_mode,unsigned bit_size)1296 nir_is_rounding_mode_rtz(unsigned execution_mode, unsigned bit_size)
1297 {
1298     return (16 == bit_size && execution_mode & FLOAT_CONTROLS_ROUNDING_MODE_RTZ_FP16) ||
1299         (32 == bit_size && execution_mode & FLOAT_CONTROLS_ROUNDING_MODE_RTZ_FP32) ||
1300         (64 == bit_size && execution_mode & FLOAT_CONTROLS_ROUNDING_MODE_RTZ_FP64);
1301 }
1302 
1303 static inline bool
nir_has_any_rounding_mode_rtz(unsigned execution_mode)1304 nir_has_any_rounding_mode_rtz(unsigned execution_mode)
1305 {
1306     return (execution_mode & FLOAT_CONTROLS_ROUNDING_MODE_RTZ_FP16) ||
1307         (execution_mode & FLOAT_CONTROLS_ROUNDING_MODE_RTZ_FP32) ||
1308         (execution_mode & FLOAT_CONTROLS_ROUNDING_MODE_RTZ_FP64);
1309 }
1310 
1311 static inline bool
nir_has_any_rounding_mode_rtne(unsigned execution_mode)1312 nir_has_any_rounding_mode_rtne(unsigned execution_mode)
1313 {
1314     return (execution_mode & FLOAT_CONTROLS_ROUNDING_MODE_RTE_FP16) ||
1315         (execution_mode & FLOAT_CONTROLS_ROUNDING_MODE_RTE_FP32) ||
1316         (execution_mode & FLOAT_CONTROLS_ROUNDING_MODE_RTE_FP64);
1317 }
1318 
1319 static inline nir_rounding_mode
nir_get_rounding_mode_from_float_controls(unsigned execution_mode,nir_alu_type type)1320 nir_get_rounding_mode_from_float_controls(unsigned execution_mode,
1321                                           nir_alu_type type)
1322 {
1323    if (nir_alu_type_get_base_type(type) != nir_type_float)
1324       return nir_rounding_mode_undef;
1325 
1326    unsigned bit_size = nir_alu_type_get_type_size(type);
1327 
1328    if (nir_is_rounding_mode_rtz(execution_mode, bit_size))
1329       return nir_rounding_mode_rtz;
1330    if (nir_is_rounding_mode_rtne(execution_mode, bit_size))
1331       return nir_rounding_mode_rtne;
1332    return nir_rounding_mode_undef;
1333 }
1334 
1335 static inline bool
nir_has_any_rounding_mode_enabled(unsigned execution_mode)1336 nir_has_any_rounding_mode_enabled(unsigned execution_mode)
1337 {
1338    bool result =
1339       nir_has_any_rounding_mode_rtne(execution_mode) ||
1340       nir_has_any_rounding_mode_rtz(execution_mode);
1341    return result;
1342 }
1343 
1344 typedef enum {
1345    /**
1346     * Operation where the first two sources are commutative.
1347     *
1348     * For 2-source operations, this just mathematical commutativity.  Some
1349     * 3-source operations, like ffma, are only commutative in the first two
1350     * sources.
1351     */
1352    NIR_OP_IS_2SRC_COMMUTATIVE = (1 << 0),
1353 
1354    /**
1355     * Operation is associative
1356     */
1357    NIR_OP_IS_ASSOCIATIVE = (1 << 1),
1358 
1359    /**
1360     * Operation where src[0] is used to select src[1] on true or src[2] false.
1361     * src[0] may be Boolean, or it may be another type used in an implicit
1362     * comparison.
1363     */
1364    NIR_OP_IS_SELECTION = (1 << 2),
1365 } nir_op_algebraic_property;
1366 
1367 /* vec16 is the widest ALU op in NIR, making the max number of input of ALU
1368  * instructions to be the same as NIR_MAX_VEC_COMPONENTS.
1369  */
1370 #define NIR_ALU_MAX_INPUTS NIR_MAX_VEC_COMPONENTS
1371 
1372 typedef struct nir_op_info {
1373    /** Name of the NIR ALU opcode */
1374    const char *name;
1375 
1376    /** Number of inputs (sources) */
1377    uint8_t num_inputs;
1378 
1379    /**
1380     * The number of components in the output
1381     *
1382     * If non-zero, this is the size of the output and input sizes are
1383     * explicitly given; swizzle and writemask are still in effect, but if
1384     * the output component is masked out, then the input component may
1385     * still be in use.
1386     *
1387     * If zero, the opcode acts in the standard, per-component manner; the
1388     * operation is performed on each component (except the ones that are
1389     * masked out) with the input being taken from the input swizzle for
1390     * that component.
1391     *
1392     * The size of some of the inputs may be given (i.e. non-zero) even
1393     * though output_size is zero; in that case, the inputs with a zero
1394     * size act per-component, while the inputs with non-zero size don't.
1395     */
1396    uint8_t output_size;
1397 
1398    /**
1399     * The type of vector that the instruction outputs. Note that the
1400     * staurate modifier is only allowed on outputs with the float type.
1401     */
1402    nir_alu_type output_type;
1403 
1404    /**
1405     * The number of components in each input
1406     *
1407     * See nir_op_infos::output_size for more detail about the relationship
1408     * between input and output sizes.
1409     */
1410    uint8_t input_sizes[NIR_ALU_MAX_INPUTS];
1411 
1412    /**
1413     * The type of vector that each input takes. Note that negate and
1414     * absolute value are only allowed on inputs with int or float type and
1415     * behave differently on the two.
1416     */
1417    nir_alu_type input_types[NIR_ALU_MAX_INPUTS];
1418 
1419    /** Algebraic properties of this opcode */
1420    nir_op_algebraic_property algebraic_properties;
1421 
1422    /** Whether this represents a numeric conversion opcode */
1423    bool is_conversion;
1424 } nir_op_info;
1425 
1426 /** Metadata for each nir_op, indexed by opcode */
1427 extern const nir_op_info nir_op_infos[nir_num_opcodes];
1428 
1429 static inline bool
nir_op_is_selection(nir_op op)1430 nir_op_is_selection(nir_op op)
1431 {
1432    return (nir_op_infos[op].algebraic_properties & NIR_OP_IS_SELECTION) != 0;
1433 }
1434 
1435 typedef struct nir_alu_instr {
1436    /** Base instruction */
1437    nir_instr instr;
1438 
1439    /** Opcode */
1440    nir_op op;
1441 
1442    /** Indicates that this ALU instruction generates an exact value
1443     *
1444     * This is kind of a mixture of GLSL "precise" and "invariant" and not
1445     * really equivalent to either.  This indicates that the value generated by
1446     * this operation is high-precision and any code transformations that touch
1447     * it must ensure that the resulting value is bit-for-bit identical to the
1448     * original.
1449     */
1450    bool exact:1;
1451 
1452    /**
1453     * Indicates that this instruction doese not cause signed integer wrapping
1454     * to occur, in the form of overflow or underflow.
1455     */
1456    bool no_signed_wrap:1;
1457 
1458    /**
1459     * Indicates that this instruction does not cause unsigned integer wrapping
1460     * to occur, in the form of overflow or underflow.
1461     */
1462    bool no_unsigned_wrap:1;
1463 
1464    /** Destination */
1465    nir_alu_dest dest;
1466 
1467    /** Sources
1468     *
1469     * The size of the array is given by nir_op_info::num_inputs.
1470     */
1471    nir_alu_src src[];
1472 } nir_alu_instr;
1473 
1474 void nir_alu_src_copy(nir_alu_src *dest, const nir_alu_src *src);
1475 void nir_alu_dest_copy(nir_alu_dest *dest, const nir_alu_dest *src);
1476 
1477 bool nir_alu_instr_is_copy(nir_alu_instr *instr);
1478 
1479 /* is this source channel used? */
1480 bool
1481 nir_alu_instr_channel_used(const nir_alu_instr *instr, unsigned src,
1482                            unsigned channel);
1483 nir_component_mask_t
1484 nir_alu_instr_src_read_mask(const nir_alu_instr *instr, unsigned src);
1485 /**
1486  * Get the number of channels used for a source
1487  */
1488 unsigned
1489 nir_ssa_alu_instr_src_components(const nir_alu_instr *instr, unsigned src);
1490 
1491 bool
1492 nir_alu_instr_is_comparison(const nir_alu_instr *instr);
1493 
1494 bool nir_const_value_negative_equal(nir_const_value c1, nir_const_value c2,
1495                                     nir_alu_type full_type);
1496 
1497 bool nir_alu_srcs_equal(const nir_alu_instr *alu1, const nir_alu_instr *alu2,
1498                         unsigned src1, unsigned src2);
1499 
1500 bool nir_alu_srcs_negative_equal(const nir_alu_instr *alu1,
1501                                  const nir_alu_instr *alu2,
1502                                  unsigned src1, unsigned src2);
1503 
1504 bool nir_alu_src_is_trivial_ssa(const nir_alu_instr *alu, unsigned srcn);
1505 
1506 typedef enum {
1507    nir_deref_type_var,
1508    nir_deref_type_array,
1509    nir_deref_type_array_wildcard,
1510    nir_deref_type_ptr_as_array,
1511    nir_deref_type_struct,
1512    nir_deref_type_cast,
1513 } nir_deref_type;
1514 
1515 typedef struct {
1516    nir_instr instr;
1517 
1518    /** The type of this deref instruction */
1519    nir_deref_type deref_type;
1520 
1521    /** Bitmask what modes the underlying variable might be
1522     *
1523     * For OpenCL-style generic pointers, we may not know exactly what mode it
1524     * is at any given point in time in the compile process.  This bitfield
1525     * contains the set of modes which it MAY be.
1526     *
1527     * Generally, this field should not be accessed directly.  Use one of the
1528     * nir_deref_mode_ helpers instead.
1529     */
1530    nir_variable_mode modes;
1531 
1532    /** The dereferenced type of the resulting pointer value */
1533    const struct glsl_type *type;
1534 
1535    union {
1536       /** Variable being dereferenced if deref_type is a deref_var */
1537       nir_variable *var;
1538 
1539       /** Parent deref if deref_type is not deref_var */
1540       nir_src parent;
1541    };
1542 
1543    /** Additional deref parameters */
1544    union {
1545       struct {
1546          nir_src index;
1547          bool in_bounds;
1548       } arr;
1549 
1550       struct {
1551          unsigned index;
1552       } strct;
1553 
1554       struct {
1555          unsigned ptr_stride;
1556          unsigned align_mul;
1557          unsigned align_offset;
1558       } cast;
1559    };
1560 
1561    /** Destination to store the resulting "pointer" */
1562    nir_dest dest;
1563 } nir_deref_instr;
1564 
1565 /** Returns true if deref might have one of the given modes
1566  *
1567  * For multi-mode derefs, this returns true if any of the possible modes for
1568  * the deref to have any of the specified modes.  This function returning true
1569  * does NOT mean that the deref definitely has one of those modes.  It simply
1570  * means that, with the best information we have at the time, it might.
1571  */
1572 static inline bool
nir_deref_mode_may_be(const nir_deref_instr * deref,nir_variable_mode modes)1573 nir_deref_mode_may_be(const nir_deref_instr *deref, nir_variable_mode modes)
1574 {
1575    assert(!(modes & ~nir_var_all));
1576    assert(deref->modes != 0);
1577    return deref->modes & modes;
1578 }
1579 
1580 /** Returns true if deref must have one of the given modes
1581  *
1582  * For multi-mode derefs, this returns true if NIR can prove that the given
1583  * deref has one of the specified modes.  This function returning false does
1584  * NOT mean that deref doesn't have one of the given mode.  It very well may
1585  * have one of those modes, we just don't have enough information to prove
1586  * that it does for sure.
1587  */
1588 static inline bool
nir_deref_mode_must_be(const nir_deref_instr * deref,nir_variable_mode modes)1589 nir_deref_mode_must_be(const nir_deref_instr *deref, nir_variable_mode modes)
1590 {
1591    assert(!(modes & ~nir_var_all));
1592    assert(deref->modes != 0);
1593    return !(deref->modes & ~modes);
1594 }
1595 
1596 /** Returns true if deref has the given mode
1597  *
1598  * This returns true if the deref has exactly the mode specified.  If the
1599  * deref may have that mode but may also have a different mode (i.e. modes has
1600  * multiple bits set), this will assert-fail.
1601  *
1602  * If you're confused about which nir_deref_mode_ helper to use, use this one
1603  * or nir_deref_mode_is_one_of below.
1604  */
1605 static inline bool
nir_deref_mode_is(const nir_deref_instr * deref,nir_variable_mode mode)1606 nir_deref_mode_is(const nir_deref_instr *deref, nir_variable_mode mode)
1607 {
1608    assert(util_bitcount(mode) == 1 && (mode & nir_var_all));
1609    assert(deref->modes != 0);
1610 
1611    /* This is only for "simple" cases so, if modes might interact with this
1612     * deref then the deref has to have a single mode.
1613     */
1614    if (nir_deref_mode_may_be(deref, mode)) {
1615       assert(util_bitcount(deref->modes) == 1);
1616       assert(deref->modes == mode);
1617    }
1618 
1619    return deref->modes == mode;
1620 }
1621 
1622 /** Returns true if deref has one of the given modes
1623  *
1624  * This returns true if the deref has exactly one possible mode and that mode
1625  * is one of the modes specified.  If the deref may have one of those modes
1626  * but may also have a different mode (i.e. modes has multiple bits set), this
1627  * will assert-fail.
1628  */
1629 static inline bool
nir_deref_mode_is_one_of(const nir_deref_instr * deref,nir_variable_mode modes)1630 nir_deref_mode_is_one_of(const nir_deref_instr *deref, nir_variable_mode modes)
1631 {
1632    /* This is only for "simple" cases so, if modes might interact with this
1633     * deref then the deref has to have a single mode.
1634     */
1635    if (nir_deref_mode_may_be(deref, modes)) {
1636       assert(util_bitcount(deref->modes) == 1);
1637       assert(nir_deref_mode_must_be(deref, modes));
1638    }
1639 
1640    return nir_deref_mode_may_be(deref, modes);
1641 }
1642 
1643 /** Returns true if deref's possible modes lie in the given set of modes
1644  *
1645  * This returns true if the deref's modes lie in the given set of modes.  If
1646  * the deref's modes overlap with the specified modes but aren't entirely
1647  * contained in the specified set of modes, this will assert-fail.  In
1648  * particular, if this is used in a generic pointers scenario, the specified
1649  * modes has to contain all or none of the possible generic pointer modes.
1650  *
1651  * This is intended mostly for mass-lowering of derefs which might have
1652  * generic pointers.
1653  */
1654 static inline bool
nir_deref_mode_is_in_set(const nir_deref_instr * deref,nir_variable_mode modes)1655 nir_deref_mode_is_in_set(const nir_deref_instr *deref, nir_variable_mode modes)
1656 {
1657    if (nir_deref_mode_may_be(deref, modes))
1658       assert(nir_deref_mode_must_be(deref, modes));
1659 
1660    return nir_deref_mode_may_be(deref, modes);
1661 }
1662 
1663 static inline nir_deref_instr *nir_src_as_deref(nir_src src);
1664 
1665 static inline nir_deref_instr *
nir_deref_instr_parent(const nir_deref_instr * instr)1666 nir_deref_instr_parent(const nir_deref_instr *instr)
1667 {
1668    if (instr->deref_type == nir_deref_type_var)
1669       return NULL;
1670    else
1671       return nir_src_as_deref(instr->parent);
1672 }
1673 
1674 static inline nir_variable *
nir_deref_instr_get_variable(const nir_deref_instr * instr)1675 nir_deref_instr_get_variable(const nir_deref_instr *instr)
1676 {
1677    while (instr->deref_type != nir_deref_type_var) {
1678       if (instr->deref_type == nir_deref_type_cast)
1679          return NULL;
1680 
1681       instr = nir_deref_instr_parent(instr);
1682    }
1683 
1684    return instr->var;
1685 }
1686 
1687 bool nir_deref_instr_has_indirect(nir_deref_instr *instr);
1688 bool nir_deref_instr_is_known_out_of_bounds(nir_deref_instr *instr);
1689 
1690 typedef enum {
1691    nir_deref_instr_has_complex_use_allow_memcpy_src = (1 << 0),
1692    nir_deref_instr_has_complex_use_allow_memcpy_dst = (1 << 1),
1693 } nir_deref_instr_has_complex_use_options;
1694 
1695 bool nir_deref_instr_has_complex_use(nir_deref_instr *instr,
1696                                      nir_deref_instr_has_complex_use_options opts);
1697 
1698 bool nir_deref_instr_remove_if_unused(nir_deref_instr *instr);
1699 
1700 unsigned nir_deref_instr_array_stride(nir_deref_instr *instr);
1701 
1702 typedef struct {
1703    nir_instr instr;
1704 
1705    struct nir_function *callee;
1706 
1707    unsigned num_params;
1708    nir_src params[];
1709 } nir_call_instr;
1710 
1711 #include "nir_intrinsics.h"
1712 
1713 #define NIR_INTRINSIC_MAX_CONST_INDEX 7
1714 
1715 /** Represents an intrinsic
1716  *
1717  * An intrinsic is an instruction type for handling things that are
1718  * more-or-less regular operations but don't just consume and produce SSA
1719  * values like ALU operations do.  Intrinsics are not for things that have
1720  * special semantic meaning such as phi nodes and parallel copies.
1721  * Examples of intrinsics include variable load/store operations, system
1722  * value loads, and the like.  Even though texturing more-or-less falls
1723  * under this category, texturing is its own instruction type because
1724  * trying to represent texturing with intrinsics would lead to a
1725  * combinatorial explosion of intrinsic opcodes.
1726  *
1727  * By having a single instruction type for handling a lot of different
1728  * cases, optimization passes can look for intrinsics and, for the most
1729  * part, completely ignore them.  Each intrinsic type also has a few
1730  * possible flags that govern whether or not they can be reordered or
1731  * eliminated.  That way passes like dead code elimination can still work
1732  * on intrisics without understanding the meaning of each.
1733  *
1734  * Each intrinsic has some number of constant indices, some number of
1735  * variables, and some number of sources.  What these sources, variables,
1736  * and indices mean depends on the intrinsic and is documented with the
1737  * intrinsic declaration in nir_intrinsics.h.  Intrinsics and texture
1738  * instructions are the only types of instruction that can operate on
1739  * variables.
1740  */
1741 typedef struct {
1742    nir_instr instr;
1743 
1744    nir_intrinsic_op intrinsic;
1745 
1746    nir_dest dest;
1747 
1748    /** number of components if this is a vectorized intrinsic
1749     *
1750     * Similarly to ALU operations, some intrinsics are vectorized.
1751     * An intrinsic is vectorized if nir_intrinsic_infos.dest_components == 0.
1752     * For vectorized intrinsics, the num_components field specifies the
1753     * number of destination components and the number of source components
1754     * for all sources with nir_intrinsic_infos.src_components[i] == 0.
1755     */
1756    uint8_t num_components;
1757 
1758    int const_index[NIR_INTRINSIC_MAX_CONST_INDEX];
1759 
1760    nir_src src[];
1761 } nir_intrinsic_instr;
1762 
1763 static inline nir_variable *
nir_intrinsic_get_var(nir_intrinsic_instr * intrin,unsigned i)1764 nir_intrinsic_get_var(nir_intrinsic_instr *intrin, unsigned i)
1765 {
1766    return nir_deref_instr_get_variable(nir_src_as_deref(intrin->src[i]));
1767 }
1768 
1769 typedef enum {
1770    /* Memory ordering. */
1771    NIR_MEMORY_ACQUIRE        = 1 << 0,
1772    NIR_MEMORY_RELEASE        = 1 << 1,
1773    NIR_MEMORY_ACQ_REL        = NIR_MEMORY_ACQUIRE | NIR_MEMORY_RELEASE,
1774 
1775    /* Memory visibility operations. */
1776    NIR_MEMORY_MAKE_AVAILABLE = 1 << 2,
1777    NIR_MEMORY_MAKE_VISIBLE   = 1 << 3,
1778 } nir_memory_semantics;
1779 
1780 typedef enum {
1781    NIR_SCOPE_NONE,
1782    NIR_SCOPE_INVOCATION,
1783    NIR_SCOPE_SUBGROUP,
1784    NIR_SCOPE_SHADER_CALL,
1785    NIR_SCOPE_WORKGROUP,
1786    NIR_SCOPE_QUEUE_FAMILY,
1787    NIR_SCOPE_DEVICE,
1788 } nir_scope;
1789 
1790 /**
1791  * \name NIR intrinsics semantic flags
1792  *
1793  * information about what the compiler can do with the intrinsics.
1794  *
1795  * \sa nir_intrinsic_info::flags
1796  */
1797 typedef enum {
1798    /**
1799     * whether the intrinsic can be safely eliminated if none of its output
1800     * value is not being used.
1801     */
1802    NIR_INTRINSIC_CAN_ELIMINATE = (1 << 0),
1803 
1804    /**
1805     * Whether the intrinsic can be reordered with respect to any other
1806     * intrinsic, i.e. whether the only reordering dependencies of the
1807     * intrinsic are due to the register reads/writes.
1808     */
1809    NIR_INTRINSIC_CAN_REORDER = (1 << 1),
1810 } nir_intrinsic_semantic_flag;
1811 
1812 /**
1813  * Maximum valid value for a nir align_mul value (in intrinsics or derefs).
1814  *
1815  * Offsets can be signed, so this is the largest power of two in int32_t.
1816  */
1817 #define NIR_ALIGN_MUL_MAX 0x40000000
1818 
1819 typedef struct nir_io_semantics {
1820    unsigned location:7; /* gl_vert_attrib, gl_varying_slot, or gl_frag_result */
1821    unsigned num_slots:6;  /* max 32, may be pessimistic with const indexing */
1822    unsigned dual_source_blend_index:1;
1823    unsigned fb_fetch_output:1; /* for GL_KHR_blend_equation_advanced */
1824    unsigned gs_streams:8; /* xxyyzzww: 2-bit stream index for each component */
1825    unsigned medium_precision:1; /* GLSL mediump qualifier */
1826    unsigned per_view:1;
1827    unsigned high_16bits:1; /* whether accessing low or high half of the slot */
1828    unsigned invariant:1; /* The variable has the invariant flag set */
1829    /* CLIP_DISTn, LAYER, VIEWPORT, and TESS_LEVEL_* have up to 3 uses:
1830     * - an output consumed by the next stage
1831     * - a system value output affecting fixed-func hardware, e.g. the clipper
1832     * - a transform feedback output written to memory
1833     * The following fields disable the first two. Transform feedback is disabled
1834     * by transform feedback info.
1835     */
1836    unsigned no_varying:1; /* whether this output isn't consumed by the next stage */
1837    unsigned no_sysval_output:1; /* whether this system value output has no
1838                                    effect due to current pipeline states */
1839    unsigned _pad:3;
1840 } nir_io_semantics;
1841 
1842 /* Transform feedback info for 2 outputs. nir_intrinsic_store_output contains
1843  * this structure twice to support up to 4 outputs. The structure is limited
1844  * to 32 bits because it's stored in nir_intrinsic_instr::const_index[].
1845  */
1846 typedef struct nir_io_xfb {
1847    struct {
1848       /* start_component is equal to the index of out[]; add 2 for io_xfb2 */
1849       /* start_component is not relative to nir_intrinsic_component */
1850       /* get the stream index from nir_io_semantics */
1851       uint8_t num_components:4;  /* max 4; if this is 0, xfb is disabled */
1852       uint8_t buffer:4;  /* buffer index, max 3 */
1853       uint8_t offset;    /* transform feedback buffer offset in dwords,
1854                             max (1K - 4) bytes */
1855    } out[2];
1856 } nir_io_xfb;
1857 
1858 unsigned
1859 nir_instr_xfb_write_mask(nir_intrinsic_instr *instr);
1860 
1861 #define NIR_INTRINSIC_MAX_INPUTS 11
1862 
1863 typedef struct {
1864    const char *name;
1865 
1866    uint8_t num_srcs; /** < number of register/SSA inputs */
1867 
1868    /** number of components of each input register
1869     *
1870     * If this value is 0, the number of components is given by the
1871     * num_components field of nir_intrinsic_instr.  If this value is -1, the
1872     * intrinsic consumes however many components are provided and it is not
1873     * validated at all.
1874     */
1875    int8_t src_components[NIR_INTRINSIC_MAX_INPUTS];
1876 
1877    bool has_dest;
1878 
1879    /** number of components of the output register
1880     *
1881     * If this value is 0, the number of components is given by the
1882     * num_components field of nir_intrinsic_instr.
1883     */
1884    uint8_t dest_components;
1885 
1886    /** bitfield of legal bit sizes */
1887    uint8_t dest_bit_sizes;
1888 
1889    /** source which the destination bit size must match
1890     *
1891     * Some intrinsics, such as subgroup intrinsics, are data manipulation
1892     * intrinsics and they have similar bit-size rules to ALU ops. This enables
1893     * validation to validate a bit more and enables auto-generated builder code
1894     * to properly determine destination bit sizes automatically.
1895     */
1896    int8_t bit_size_src;
1897 
1898    /** the number of constant indices used by the intrinsic */
1899    uint8_t num_indices;
1900 
1901    /** list of indices */
1902    uint8_t indices[NIR_INTRINSIC_MAX_CONST_INDEX];
1903 
1904    /** indicates the usage of intr->const_index[n] */
1905    uint8_t index_map[NIR_INTRINSIC_NUM_INDEX_FLAGS];
1906 
1907    /** semantic flags for calls to this intrinsic */
1908    nir_intrinsic_semantic_flag flags;
1909 } nir_intrinsic_info;
1910 
1911 extern const nir_intrinsic_info nir_intrinsic_infos[nir_num_intrinsics];
1912 
1913 unsigned
1914 nir_intrinsic_src_components(const nir_intrinsic_instr *intr, unsigned srcn);
1915 
1916 unsigned
1917 nir_intrinsic_dest_components(nir_intrinsic_instr *intr);
1918 
1919 /**
1920  * Helper to copy const_index[] from src to dst, without assuming they
1921  * match in order.
1922  */
1923 void nir_intrinsic_copy_const_indices(nir_intrinsic_instr *dst, nir_intrinsic_instr *src);
1924 
1925 #include "nir_intrinsics_indices.h"
1926 
1927 static inline void
nir_intrinsic_set_align(nir_intrinsic_instr * intrin,unsigned align_mul,unsigned align_offset)1928 nir_intrinsic_set_align(nir_intrinsic_instr *intrin,
1929                         unsigned align_mul, unsigned align_offset)
1930 {
1931    assert(util_is_power_of_two_nonzero(align_mul));
1932    assert(align_offset < align_mul);
1933    nir_intrinsic_set_align_mul(intrin, align_mul);
1934    nir_intrinsic_set_align_offset(intrin, align_offset);
1935 }
1936 
1937 /** Returns a simple alignment for a load/store intrinsic offset
1938  *
1939  * Instead of the full mul+offset alignment scheme provided by the ALIGN_MUL
1940  * and ALIGN_OFFSET parameters, this helper takes both into account and
1941  * provides a single simple alignment parameter.  The offset X is guaranteed
1942  * to satisfy X % align == 0.
1943  */
1944 static inline unsigned
nir_intrinsic_align(const nir_intrinsic_instr * intrin)1945 nir_intrinsic_align(const nir_intrinsic_instr *intrin)
1946 {
1947    const unsigned align_mul = nir_intrinsic_align_mul(intrin);
1948    const unsigned align_offset = nir_intrinsic_align_offset(intrin);
1949    assert(align_offset < align_mul);
1950    return align_offset ? 1 << (ffs(align_offset) - 1) : align_mul;
1951 }
1952 
1953 static inline bool
nir_intrinsic_has_align(const nir_intrinsic_instr * intrin)1954 nir_intrinsic_has_align(const nir_intrinsic_instr *intrin)
1955 {
1956    return nir_intrinsic_has_align_mul(intrin) &&
1957           nir_intrinsic_has_align_offset(intrin);
1958 }
1959 
1960 unsigned
1961 nir_image_intrinsic_coord_components(const nir_intrinsic_instr *instr);
1962 
1963 /* Converts a image_deref_* intrinsic into a image_* one */
1964 void nir_rewrite_image_intrinsic(nir_intrinsic_instr *instr,
1965                                  nir_ssa_def *handle, bool bindless);
1966 
1967 /* Determine if an intrinsic can be arbitrarily reordered and eliminated. */
1968 static inline bool
nir_intrinsic_can_reorder(nir_intrinsic_instr * instr)1969 nir_intrinsic_can_reorder(nir_intrinsic_instr *instr)
1970 {
1971    if (nir_intrinsic_has_access(instr) &&
1972        nir_intrinsic_access(instr) & ACCESS_VOLATILE)
1973       return false;
1974    if (instr->intrinsic == nir_intrinsic_load_deref) {
1975       nir_deref_instr *deref = nir_src_as_deref(instr->src[0]);
1976       return nir_deref_mode_is_in_set(deref, nir_var_read_only_modes) ||
1977              (nir_intrinsic_access(instr) & ACCESS_CAN_REORDER);
1978    } else if (instr->intrinsic == nir_intrinsic_load_ssbo ||
1979               instr->intrinsic == nir_intrinsic_bindless_image_load ||
1980               instr->intrinsic == nir_intrinsic_image_deref_load ||
1981               instr->intrinsic == nir_intrinsic_image_load) {
1982       return nir_intrinsic_access(instr) & ACCESS_CAN_REORDER;
1983    } else {
1984       const nir_intrinsic_info *info =
1985          &nir_intrinsic_infos[instr->intrinsic];
1986       return (info->flags & NIR_INTRINSIC_CAN_ELIMINATE) &&
1987              (info->flags & NIR_INTRINSIC_CAN_REORDER);
1988    }
1989 }
1990 
1991 bool nir_intrinsic_writes_external_memory(const nir_intrinsic_instr *instr);
1992 
1993 /** Texture instruction source type */
1994 typedef enum {
1995    /** Texture coordinate
1996     *
1997     * Must have nir_tex_instr::coord_components components.
1998     */
1999    nir_tex_src_coord,
2000 
2001    /** Projector
2002     *
2003     * The texture coordinate (except for the array component, if any) is
2004     * divided by this value before LOD computation and sampling.
2005     *
2006     * Must be a float scalar.
2007     */
2008    nir_tex_src_projector,
2009 
2010    /** Shadow comparator
2011     *
2012     * For shadow sampling, the fetched texel values are compared against the
2013     * shadow comparator using the compare op specified by the sampler object
2014     * and converted to 1.0 if the comparison succeeds and 0.0 if it fails.
2015     * Interpolation happens after this conversion so the actual result may be
2016     * anywhere in the range [0.0, 1.0].
2017     *
2018     * Only valid if nir_tex_instr::is_shadow and must be a float scalar.
2019     */
2020    nir_tex_src_comparator,
2021 
2022    /** Coordinate offset
2023     *
2024     * An integer value that is added to the texel address before sampling.
2025     * This is only allowed with operations that take an explicit LOD as it is
2026     * applied in integer texel space after LOD selection and not normalized
2027     * coordinate space.
2028     */
2029    nir_tex_src_offset,
2030 
2031    /** LOD bias
2032     *
2033     * This value is added to the computed LOD before mip-mapping.
2034     */
2035    nir_tex_src_bias,
2036 
2037    /** Explicit LOD */
2038    nir_tex_src_lod,
2039 
2040    /** Min LOD
2041     *
2042     * The computed LOD is clamped to be at least as large as min_lod before
2043     * mip-mapping.
2044     */
2045    nir_tex_src_min_lod,
2046 
2047    /** MSAA sample index */
2048    nir_tex_src_ms_index,
2049 
2050    /** Intel-specific MSAA compression data */
2051    nir_tex_src_ms_mcs_intel,
2052 
2053    /** Explicit horizontal (X-major) coordinate derivative */
2054    nir_tex_src_ddx,
2055 
2056    /** Explicit vertical (Y-major) coordinate derivative */
2057    nir_tex_src_ddy,
2058 
2059    /** Texture variable dereference */
2060    nir_tex_src_texture_deref,
2061 
2062    /** Sampler variable dereference */
2063    nir_tex_src_sampler_deref,
2064 
2065    /** Texture index offset
2066     *
2067     * This is added to nir_tex_instr::texture_index.  Unless
2068     * nir_tex_instr::texture_non_uniform is set, this is guaranteed to be
2069     * dynamically uniform.
2070     */
2071    nir_tex_src_texture_offset,
2072 
2073    /** Dynamically uniform sampler index offset
2074     *
2075     * This is added to nir_tex_instr::sampler_index.  Unless
2076     * nir_tex_instr::sampler_non_uniform is set, this is guaranteed to be
2077     * dynamically uniform.  This should not be present until GLSL ES 3.20, GLSL
2078     * 4.00, or ARB_gpu_shader5, because in ES 3.10 and GL 3.30 samplers said
2079     * "When aggregated into arrays within a shader, samplers can only be indexed
2080     * with a constant integral expression."
2081     */
2082    nir_tex_src_sampler_offset,
2083 
2084    /** Bindless texture handle
2085     *
2086     * This is, unfortunately, a bit overloaded at the moment.  There are
2087     * generally two types of bindless handles:
2088     *
2089     *  1. For GL_ARB_bindless bindless handles. These are part of the
2090     *     GL/Gallium-level API and are always a 64-bit integer.
2091     *
2092     *  2. HW-specific handles.  GL_ARB_bindless handles may be lowered to
2093     *     these.  Also, these are used by many Vulkan drivers to implement
2094     *     descriptor sets, especially for UPDATE_AFTER_BIND descriptors.
2095     *     The details of hardware handles (bit size, format, etc.) is
2096     *     HW-specific.
2097     *
2098     * Because of this overloading and the resulting ambiguity, we currently
2099     * don't validate anything for these.
2100     */
2101    nir_tex_src_texture_handle,
2102 
2103    /** Bindless sampler handle
2104     *
2105     * See nir_tex_src_texture_handle,
2106     */
2107    nir_tex_src_sampler_handle,
2108 
2109    /** Plane index for multi-plane YCbCr textures */
2110    nir_tex_src_plane,
2111 
2112    /**
2113     * Backend-specific vec4 tex src argument.
2114     *
2115     * Can be used to have NIR optimization (copy propagation, lower_vec_to_movs)
2116     * apply to the packing of the tex srcs.  This lowering must only happen
2117     * after nir_lower_tex().
2118     *
2119     * The nir_tex_instr_src_type() of this argument is float, so no lowering
2120     * will happen if nir_lower_int_to_float is used.
2121     */
2122    nir_tex_src_backend1,
2123 
2124    /** Second backend-specific vec4 tex src argument, see nir_tex_src_backend1. */
2125    nir_tex_src_backend2,
2126 
2127    nir_num_tex_src_types
2128 } nir_tex_src_type;
2129 
2130 /** A texture instruction source */
2131 typedef struct {
2132    /** Base source */
2133    nir_src src;
2134 
2135    /** Type of this source */
2136    nir_tex_src_type src_type;
2137 } nir_tex_src;
2138 
2139 /** Texture instruction opcode */
2140 typedef enum {
2141    nir_texop_tex,                /**< Regular texture look-up */
2142    nir_texop_txb,                /**< Texture look-up with LOD bias */
2143    nir_texop_txl,                /**< Texture look-up with explicit LOD */
2144    nir_texop_txd,                /**< Texture look-up with partial derivatives */
2145    nir_texop_txf,                /**< Texel fetch with explicit LOD */
2146    nir_texop_txf_ms,             /**< Multisample texture fetch */
2147    nir_texop_txf_ms_fb,          /**< Multisample texture fetch from framebuffer */
2148    nir_texop_txf_ms_mcs_intel,   /**< Multisample compression value fetch */
2149    nir_texop_txs,                /**< Texture size */
2150    nir_texop_lod,                /**< Texture lod query */
2151    nir_texop_tg4,                /**< Texture gather */
2152    nir_texop_query_levels,       /**< Texture levels query */
2153    nir_texop_texture_samples,    /**< Texture samples query */
2154    nir_texop_samples_identical,  /**< Query whether all samples are definitely
2155                                   * identical.
2156                                   */
2157    nir_texop_tex_prefetch,       /**< Regular texture look-up, eligible for pre-dispatch */
2158    nir_texop_fragment_fetch_amd,      /**< Multisample fragment color texture fetch */
2159    nir_texop_fragment_mask_fetch_amd, /**< Multisample fragment mask texture fetch */
2160 } nir_texop;
2161 
2162 /** Represents a texture instruction */
2163 typedef struct {
2164    /** Base instruction */
2165    nir_instr instr;
2166 
2167    /** Dimensionality of the texture operation
2168     *
2169     * This will typically match the dimensionality of the texture deref type
2170     * if a nir_tex_src_texture_deref is present.  However, it may not if
2171     * texture lowering has occurred.
2172     */
2173    enum glsl_sampler_dim sampler_dim;
2174 
2175    /** ALU type of the destination
2176     *
2177     * This is the canonical sampled type for this texture operation and may
2178     * not exactly match the sampled type of the deref type when a
2179     * nir_tex_src_texture_deref is present.  For OpenCL, the sampled type of
2180     * the texture deref will be GLSL_TYPE_VOID and this is allowed to be
2181     * anything.  With SPIR-V, the signedness of integer types is allowed to
2182     * differ.  For all APIs, the bit size may differ if the driver has done
2183     * any sort of mediump or similar lowering since texture types always have
2184     * 32-bit sampled types.
2185     */
2186    nir_alu_type dest_type;
2187 
2188    /** Texture opcode */
2189    nir_texop op;
2190 
2191    /** Destination */
2192    nir_dest dest;
2193 
2194    /** Array of sources
2195     *
2196     * This array has nir_tex_instr::num_srcs elements
2197     */
2198    nir_tex_src *src;
2199 
2200    /** Number of sources */
2201    unsigned num_srcs;
2202 
2203    /** Number of components in the coordinate, if any */
2204    unsigned coord_components;
2205 
2206    /** True if the texture instruction acts on an array texture */
2207    bool is_array;
2208 
2209    /** True if the texture instruction performs a shadow comparison
2210     *
2211     * If this is true, the texture instruction must have a
2212     * nir_tex_src_comparator.
2213     */
2214    bool is_shadow;
2215 
2216    /**
2217     * If is_shadow is true, whether this is the old-style shadow that outputs
2218     * 4 components or the new-style shadow that outputs 1 component.
2219     */
2220    bool is_new_style_shadow;
2221 
2222    /**
2223     * True if this texture instruction should return a sparse residency code.
2224     * The code is in the last component of the result.
2225     */
2226    bool is_sparse;
2227 
2228    /** nir_texop_tg4 component selector
2229     *
2230     * This determines which RGBA component is gathered.
2231     */
2232    unsigned component : 2;
2233 
2234    /** Validation needs to know this for gradient component count */
2235    unsigned array_is_lowered_cube : 1;
2236 
2237    /** Gather offsets */
2238    int8_t tg4_offsets[4][2];
2239 
2240    /** True if the texture index or handle is not dynamically uniform */
2241    bool texture_non_uniform;
2242 
2243    /** True if the sampler index or handle is not dynamically uniform.
2244     *
2245     * This may be set when VK_EXT_descriptor_indexing is supported and the
2246     * appropriate capability is enabled.
2247     *
2248     * This should always be false in GLSL (GLSL ES 3.20 says "When aggregated
2249     * into arrays within a shader, opaque types can only be indexed with a
2250     * dynamically uniform integral expression", and GLSL 4.60 says "When
2251     * aggregated into arrays within a shader, [texture, sampler, and
2252     * samplerShadow] types can only be indexed with a dynamically uniform
2253     * expression, or texture lookup will result in undefined values.").
2254     */
2255    bool sampler_non_uniform;
2256 
2257    /** The texture index
2258     *
2259     * If this texture instruction has a nir_tex_src_texture_offset source,
2260     * then the texture index is given by texture_index + texture_offset.
2261     */
2262    unsigned texture_index;
2263 
2264    /** The sampler index
2265     *
2266     * The following operations do not require a sampler and, as such, this
2267     * field should be ignored:
2268     *    - nir_texop_txf
2269     *    - nir_texop_txf_ms
2270     *    - nir_texop_txs
2271     *    - nir_texop_query_levels
2272     *    - nir_texop_texture_samples
2273     *    - nir_texop_samples_identical
2274     *
2275     * If this texture instruction has a nir_tex_src_sampler_offset source,
2276     * then the sampler index is given by sampler_index + sampler_offset.
2277     */
2278    unsigned sampler_index;
2279 } nir_tex_instr;
2280 
2281 /**
2282  * Returns true if the texture operation requires a sampler as a general rule
2283  *
2284  * Note that the specific hw/driver backend could require to a sampler
2285  * object/configuration packet in any case, for some other reason.
2286  *
2287  * @see nir_tex_instr::sampler_index.
2288  */
2289 bool nir_tex_instr_need_sampler(const nir_tex_instr *instr);
2290 
2291 /** Returns the number of components returned by this nir_tex_instr
2292  *
2293  * Useful for code building texture instructions when you don't want to think
2294  * about how many components a particular texture op returns.  This does not
2295  * include the sparse residency code.
2296  */
2297 unsigned
2298 nir_tex_instr_result_size(const nir_tex_instr *instr);
2299 
2300 /**
2301  * Returns the destination size of this nir_tex_instr including the sparse
2302  * residency code, if any.
2303  */
2304 static inline unsigned
nir_tex_instr_dest_size(const nir_tex_instr * instr)2305 nir_tex_instr_dest_size(const nir_tex_instr *instr)
2306 {
2307    /* One more component is needed for the residency code. */
2308    return nir_tex_instr_result_size(instr) + instr->is_sparse;
2309 }
2310 
2311 /**
2312  * Returns true if this texture operation queries something about the texture
2313  * rather than actually sampling it.
2314  */
2315 bool
2316 nir_tex_instr_is_query(const nir_tex_instr *instr);
2317 
2318 /** Returns true if this texture instruction does implicit derivatives
2319  *
2320  * This is important as there are extra control-flow rules around derivatives
2321  * and texture instructions which perform them implicitly.
2322  */
2323 bool
2324 nir_tex_instr_has_implicit_derivative(const nir_tex_instr *instr);
2325 
2326 /** Returns the ALU type of the given texture instruction source */
2327 nir_alu_type
2328 nir_tex_instr_src_type(const nir_tex_instr *instr, unsigned src);
2329 
2330 /**
2331  * Returns the number of components required by the given texture instruction
2332  * source
2333  */
2334 unsigned
2335 nir_tex_instr_src_size(const nir_tex_instr *instr, unsigned src);
2336 
2337 /**
2338  * Returns the index of the texture instruction source with the given
2339  * nir_tex_src_type or -1 if no such source exists.
2340  */
2341 static inline int
nir_tex_instr_src_index(const nir_tex_instr * instr,nir_tex_src_type type)2342 nir_tex_instr_src_index(const nir_tex_instr *instr, nir_tex_src_type type)
2343 {
2344    for (unsigned i = 0; i < instr->num_srcs; i++)
2345       if (instr->src[i].src_type == type)
2346          return (int) i;
2347 
2348    return -1;
2349 }
2350 
2351 /** Adds a source to a texture instruction */
2352 void nir_tex_instr_add_src(nir_tex_instr *tex,
2353                            nir_tex_src_type src_type,
2354                            nir_src src);
2355 
2356 /** Removes a source from a texture instruction */
2357 void nir_tex_instr_remove_src(nir_tex_instr *tex, unsigned src_idx);
2358 
2359 bool nir_tex_instr_has_explicit_tg4_offsets(nir_tex_instr *tex);
2360 
2361 typedef struct {
2362    nir_instr instr;
2363 
2364    nir_ssa_def def;
2365 
2366    nir_const_value value[];
2367 } nir_load_const_instr;
2368 
2369 typedef enum {
2370    /** Return from a function
2371     *
2372     * This instruction is a classic function return.  It jumps to
2373     * nir_function_impl::end_block.  No return value is provided in this
2374     * instruction.  Instead, the function is expected to write any return
2375     * data to a deref passed in from the caller.
2376     */
2377    nir_jump_return,
2378 
2379    /** Immediately exit the current shader
2380     *
2381     * This instruction is roughly the equivalent of C's "exit()" in that it
2382     * immediately terminates the current shader invocation.  From a CFG
2383     * perspective, it looks like a jump to nir_function_impl::end_block but
2384     * it actually jumps to the end block of the shader entrypoint.  A halt
2385     * instruction in the shader entrypoint itself is semantically identical
2386     * to a return.
2387     *
2388     * For shaders with built-in I/O, any outputs written prior to a halt
2389     * instruction remain written and any outputs not written prior to the
2390     * halt have undefined values.  It does NOT cause an implicit discard of
2391     * written results.  If one wants discard results in a fragment shader,
2392     * for instance, a discard or demote intrinsic is required.
2393     */
2394    nir_jump_halt,
2395 
2396    /** Break out of the inner-most loop
2397     *
2398     * This has the same semantics as C's "break" statement.
2399     */
2400    nir_jump_break,
2401 
2402    /** Jump back to the top of the inner-most loop
2403     *
2404     * This has the same semantics as C's "continue" statement assuming that a
2405     * NIR loop is implemented as "while (1) { body }".
2406     */
2407    nir_jump_continue,
2408 
2409    /** Jumps for unstructured CFG.
2410     *
2411     * As within an unstructured CFG we can't rely on block ordering we need to
2412     * place explicit jumps at the end of every block.
2413     */
2414    nir_jump_goto,
2415    nir_jump_goto_if,
2416 } nir_jump_type;
2417 
2418 typedef struct {
2419    nir_instr instr;
2420    nir_jump_type type;
2421    nir_src condition;
2422    struct nir_block *target;
2423    struct nir_block *else_target;
2424 } nir_jump_instr;
2425 
2426 /* creates a new SSA variable in an undefined state */
2427 
2428 typedef struct {
2429    nir_instr instr;
2430    nir_ssa_def def;
2431 } nir_ssa_undef_instr;
2432 
2433 typedef struct {
2434    struct exec_node node;
2435 
2436    /* The predecessor block corresponding to this source */
2437    struct nir_block *pred;
2438 
2439    nir_src src;
2440 } nir_phi_src;
2441 
2442 #define nir_foreach_phi_src(phi_src, phi) \
2443    foreach_list_typed(nir_phi_src, phi_src, node, &(phi)->srcs)
2444 #define nir_foreach_phi_src_safe(phi_src, phi) \
2445    foreach_list_typed_safe(nir_phi_src, phi_src, node, &(phi)->srcs)
2446 
2447 typedef struct {
2448    nir_instr instr;
2449 
2450    struct exec_list srcs; /** < list of nir_phi_src */
2451 
2452    nir_dest dest;
2453 } nir_phi_instr;
2454 
2455 static inline nir_phi_src *
nir_phi_get_src_from_block(nir_phi_instr * phi,struct nir_block * block)2456 nir_phi_get_src_from_block(nir_phi_instr *phi, struct nir_block *block)
2457 {
2458    nir_foreach_phi_src(src, phi) {
2459       if (src->pred == block)
2460          return src;
2461    }
2462 
2463    assert(!"Block is not a predecessor of phi.");
2464    return NULL;
2465 }
2466 
2467 typedef struct {
2468    struct exec_node node;
2469    nir_src src;
2470    nir_dest dest;
2471 } nir_parallel_copy_entry;
2472 
2473 #define nir_foreach_parallel_copy_entry(entry, pcopy) \
2474    foreach_list_typed(nir_parallel_copy_entry, entry, node, &(pcopy)->entries)
2475 
2476 typedef struct {
2477    nir_instr instr;
2478 
2479    /* A list of nir_parallel_copy_entrys.  The sources of all of the
2480     * entries are copied to the corresponding destinations "in parallel".
2481     * In other words, if we have two entries: a -> b and b -> a, the values
2482     * get swapped.
2483     */
2484    struct exec_list entries;
2485 } nir_parallel_copy_instr;
2486 
2487 NIR_DEFINE_CAST(nir_instr_as_alu, nir_instr, nir_alu_instr, instr,
2488                 type, nir_instr_type_alu)
2489 NIR_DEFINE_CAST(nir_instr_as_deref, nir_instr, nir_deref_instr, instr,
2490                 type, nir_instr_type_deref)
2491 NIR_DEFINE_CAST(nir_instr_as_call, nir_instr, nir_call_instr, instr,
2492                 type, nir_instr_type_call)
2493 NIR_DEFINE_CAST(nir_instr_as_jump, nir_instr, nir_jump_instr, instr,
2494                 type, nir_instr_type_jump)
2495 NIR_DEFINE_CAST(nir_instr_as_tex, nir_instr, nir_tex_instr, instr,
2496                 type, nir_instr_type_tex)
2497 NIR_DEFINE_CAST(nir_instr_as_intrinsic, nir_instr, nir_intrinsic_instr, instr,
2498                 type, nir_instr_type_intrinsic)
2499 NIR_DEFINE_CAST(nir_instr_as_load_const, nir_instr, nir_load_const_instr, instr,
2500                 type, nir_instr_type_load_const)
2501 NIR_DEFINE_CAST(nir_instr_as_ssa_undef, nir_instr, nir_ssa_undef_instr, instr,
2502                 type, nir_instr_type_ssa_undef)
2503 NIR_DEFINE_CAST(nir_instr_as_phi, nir_instr, nir_phi_instr, instr,
2504                 type, nir_instr_type_phi)
2505 NIR_DEFINE_CAST(nir_instr_as_parallel_copy, nir_instr,
2506                 nir_parallel_copy_instr, instr,
2507                 type, nir_instr_type_parallel_copy)
2508 
2509 
2510 #define NIR_DEFINE_SRC_AS_CONST(type, suffix)               \
2511 static inline type                                          \
2512 nir_src_comp_as_##suffix(nir_src src, unsigned comp)        \
2513 {                                                           \
2514    assert(nir_src_is_const(src));                           \
2515    nir_load_const_instr *load =                             \
2516       nir_instr_as_load_const(src.ssa->parent_instr);       \
2517    assert(comp < load->def.num_components);                 \
2518    return nir_const_value_as_##suffix(load->value[comp],    \
2519                                       load->def.bit_size);  \
2520 }                                                           \
2521                                                             \
2522 static inline type                                          \
2523 nir_src_as_##suffix(nir_src src)                            \
2524 {                                                           \
2525    assert(nir_src_num_components(src) == 1);                \
2526    return nir_src_comp_as_##suffix(src, 0);                 \
2527 }
2528 
2529 NIR_DEFINE_SRC_AS_CONST(int64_t,    int)
2530 NIR_DEFINE_SRC_AS_CONST(uint64_t,   uint)
2531 NIR_DEFINE_SRC_AS_CONST(bool,       bool)
2532 NIR_DEFINE_SRC_AS_CONST(double,     float)
2533 
2534 #undef NIR_DEFINE_SRC_AS_CONST
2535 
2536 
2537 typedef struct {
2538    nir_ssa_def *def;
2539    unsigned comp;
2540 } nir_ssa_scalar;
2541 
2542 static inline bool
nir_ssa_scalar_is_const(nir_ssa_scalar s)2543 nir_ssa_scalar_is_const(nir_ssa_scalar s)
2544 {
2545    return s.def->parent_instr->type == nir_instr_type_load_const;
2546 }
2547 
2548 static inline nir_const_value
nir_ssa_scalar_as_const_value(nir_ssa_scalar s)2549 nir_ssa_scalar_as_const_value(nir_ssa_scalar s)
2550 {
2551    assert(s.comp < s.def->num_components);
2552    nir_load_const_instr *load = nir_instr_as_load_const(s.def->parent_instr);
2553    return load->value[s.comp];
2554 }
2555 
2556 #define NIR_DEFINE_SCALAR_AS_CONST(type, suffix)                     \
2557 static inline type                                                   \
2558 nir_ssa_scalar_as_##suffix(nir_ssa_scalar s)                         \
2559 {                                                                    \
2560    return nir_const_value_as_##suffix(                               \
2561       nir_ssa_scalar_as_const_value(s), s.def->bit_size);            \
2562 }
2563 
NIR_DEFINE_SCALAR_AS_CONST(int64_t,int)2564 NIR_DEFINE_SCALAR_AS_CONST(int64_t,    int)
2565 NIR_DEFINE_SCALAR_AS_CONST(uint64_t,   uint)
2566 NIR_DEFINE_SCALAR_AS_CONST(bool,       bool)
2567 NIR_DEFINE_SCALAR_AS_CONST(double,     float)
2568 
2569 #undef NIR_DEFINE_SCALAR_AS_CONST
2570 
2571 static inline bool
2572 nir_ssa_scalar_is_alu(nir_ssa_scalar s)
2573 {
2574    return s.def->parent_instr->type == nir_instr_type_alu;
2575 }
2576 
2577 static inline nir_op
nir_ssa_scalar_alu_op(nir_ssa_scalar s)2578 nir_ssa_scalar_alu_op(nir_ssa_scalar s)
2579 {
2580    return nir_instr_as_alu(s.def->parent_instr)->op;
2581 }
2582 
2583 static inline nir_ssa_scalar
nir_ssa_scalar_chase_alu_src(nir_ssa_scalar s,unsigned alu_src_idx)2584 nir_ssa_scalar_chase_alu_src(nir_ssa_scalar s, unsigned alu_src_idx)
2585 {
2586    nir_ssa_scalar out = { NULL, 0 };
2587 
2588    nir_alu_instr *alu = nir_instr_as_alu(s.def->parent_instr);
2589    assert(alu_src_idx < nir_op_infos[alu->op].num_inputs);
2590 
2591    /* Our component must be written */
2592    assert(s.comp < s.def->num_components);
2593    assert(alu->dest.write_mask & (1u << s.comp));
2594 
2595    assert(alu->src[alu_src_idx].src.is_ssa);
2596    out.def = alu->src[alu_src_idx].src.ssa;
2597 
2598    if (nir_op_infos[alu->op].input_sizes[alu_src_idx] == 0) {
2599       /* The ALU src is unsized so the source component follows the
2600        * destination component.
2601        */
2602       out.comp = alu->src[alu_src_idx].swizzle[s.comp];
2603    } else {
2604       /* This is a sized source so all source components work together to
2605        * produce all the destination components.  Since we need to return a
2606        * scalar, this only works if the source is a scalar.
2607        */
2608       assert(nir_op_infos[alu->op].input_sizes[alu_src_idx] == 1);
2609       out.comp = alu->src[alu_src_idx].swizzle[0];
2610    }
2611    assert(out.comp < out.def->num_components);
2612 
2613    return out;
2614 }
2615 
2616 nir_ssa_scalar nir_ssa_scalar_chase_movs(nir_ssa_scalar s);
2617 
2618 static inline nir_ssa_scalar
nir_get_ssa_scalar(nir_ssa_def * def,unsigned channel)2619 nir_get_ssa_scalar(nir_ssa_def *def, unsigned channel)
2620 {
2621    nir_ssa_scalar s = { def, channel };
2622    return s;
2623 }
2624 
2625 /** Returns a nir_ssa_scalar where we've followed the bit-exact mov/vec use chain to the original definition */
2626 static inline nir_ssa_scalar
nir_ssa_scalar_resolved(nir_ssa_def * def,unsigned channel)2627 nir_ssa_scalar_resolved(nir_ssa_def *def, unsigned channel)
2628 {
2629    return nir_ssa_scalar_chase_movs(nir_get_ssa_scalar(def, channel));
2630 }
2631 
2632 
2633 typedef struct {
2634    bool success;
2635 
2636    nir_variable *var;
2637    unsigned desc_set;
2638    unsigned binding;
2639    unsigned num_indices;
2640    nir_src indices[4];
2641    bool read_first_invocation;
2642 } nir_binding;
2643 
2644 nir_binding nir_chase_binding(nir_src rsrc);
2645 nir_variable *nir_get_binding_variable(struct nir_shader *shader, nir_binding binding);
2646 
2647 
2648 /*
2649  * Control flow
2650  *
2651  * Control flow consists of a tree of control flow nodes, which include
2652  * if-statements and loops. The leaves of the tree are basic blocks, lists of
2653  * instructions that always run start-to-finish. Each basic block also keeps
2654  * track of its successors (blocks which may run immediately after the current
2655  * block) and predecessors (blocks which could have run immediately before the
2656  * current block). Each function also has a start block and an end block which
2657  * all return statements point to (which is always empty). Together, all the
2658  * blocks with their predecessors and successors make up the control flow
2659  * graph (CFG) of the function. There are helpers that modify the tree of
2660  * control flow nodes while modifying the CFG appropriately; these should be
2661  * used instead of modifying the tree directly.
2662  */
2663 
2664 typedef enum {
2665    nir_cf_node_block,
2666    nir_cf_node_if,
2667    nir_cf_node_loop,
2668    nir_cf_node_function
2669 } nir_cf_node_type;
2670 
2671 typedef struct nir_cf_node {
2672    struct exec_node node;
2673    nir_cf_node_type type;
2674    struct nir_cf_node *parent;
2675 } nir_cf_node;
2676 
2677 typedef struct nir_block {
2678    nir_cf_node cf_node;
2679 
2680    struct exec_list instr_list; /** < list of nir_instr */
2681 
2682    /** generic block index; generated by nir_index_blocks */
2683    unsigned index;
2684 
2685    /*
2686     * Each block can only have up to 2 successors, so we put them in a simple
2687     * array - no need for anything more complicated.
2688     */
2689    struct nir_block *successors[2];
2690 
2691    /* Set of nir_block predecessors in the CFG */
2692    struct set *predecessors;
2693 
2694    /*
2695     * this node's immediate dominator in the dominance tree - set to NULL for
2696     * the start block.
2697     */
2698    struct nir_block *imm_dom;
2699 
2700    /* This node's children in the dominance tree */
2701    unsigned num_dom_children;
2702    struct nir_block **dom_children;
2703 
2704    /* Set of nir_blocks on the dominance frontier of this block */
2705    struct set *dom_frontier;
2706 
2707    /*
2708     * These two indices have the property that dom_{pre,post}_index for each
2709     * child of this block in the dominance tree will always be between
2710     * dom_pre_index and dom_post_index for this block, which makes testing if
2711     * a given block is dominated by another block an O(1) operation.
2712     */
2713    uint32_t dom_pre_index, dom_post_index;
2714 
2715    /**
2716     * Value just before the first nir_instr->index in the block, but after
2717     * end_ip that of any predecessor block.
2718     */
2719    uint32_t start_ip;
2720    /**
2721     * Value just after the last nir_instr->index in the block, but before the
2722     * start_ip of any successor block.
2723     */
2724    uint32_t end_ip;
2725 
2726    /* SSA def live in and out for this block; used for liveness analysis.
2727     * Indexed by ssa_def->index
2728     */
2729    BITSET_WORD *live_in;
2730    BITSET_WORD *live_out;
2731 } nir_block;
2732 
2733 static inline bool
nir_block_is_reachable(nir_block * b)2734 nir_block_is_reachable(nir_block *b)
2735 {
2736    /* See also nir_block_dominates */
2737    return b->dom_post_index != 0;
2738 }
2739 
2740 static inline nir_instr *
nir_block_first_instr(nir_block * block)2741 nir_block_first_instr(nir_block *block)
2742 {
2743    struct exec_node *head = exec_list_get_head(&block->instr_list);
2744    return exec_node_data(nir_instr, head, node);
2745 }
2746 
2747 static inline nir_instr *
nir_block_last_instr(nir_block * block)2748 nir_block_last_instr(nir_block *block)
2749 {
2750    struct exec_node *tail = exec_list_get_tail(&block->instr_list);
2751    return exec_node_data(nir_instr, tail, node);
2752 }
2753 
2754 static inline bool
nir_block_ends_in_jump(nir_block * block)2755 nir_block_ends_in_jump(nir_block *block)
2756 {
2757    return !exec_list_is_empty(&block->instr_list) &&
2758           nir_block_last_instr(block)->type == nir_instr_type_jump;
2759 }
2760 
2761 static inline bool
nir_block_ends_in_return_or_halt(nir_block * block)2762 nir_block_ends_in_return_or_halt(nir_block *block)
2763 {
2764    if (exec_list_is_empty(&block->instr_list))
2765       return false;
2766 
2767    nir_instr *instr = nir_block_last_instr(block);
2768    if (instr->type != nir_instr_type_jump)
2769       return false;
2770 
2771    nir_jump_instr *jump_instr = nir_instr_as_jump(instr);
2772    return jump_instr->type == nir_jump_return ||
2773           jump_instr->type == nir_jump_halt;
2774 }
2775 
2776 static inline bool
nir_block_ends_in_break(nir_block * block)2777 nir_block_ends_in_break(nir_block *block)
2778 {
2779    if (exec_list_is_empty(&block->instr_list))
2780       return false;
2781 
2782    nir_instr *instr = nir_block_last_instr(block);
2783    return instr->type == nir_instr_type_jump &&
2784       nir_instr_as_jump(instr)->type == nir_jump_break;
2785 }
2786 
2787 #define nir_foreach_instr(instr, block) \
2788    foreach_list_typed(nir_instr, instr, node, &(block)->instr_list)
2789 #define nir_foreach_instr_reverse(instr, block) \
2790    foreach_list_typed_reverse(nir_instr, instr, node, &(block)->instr_list)
2791 #define nir_foreach_instr_safe(instr, block) \
2792    foreach_list_typed_safe(nir_instr, instr, node, &(block)->instr_list)
2793 #define nir_foreach_instr_reverse_safe(instr, block) \
2794    foreach_list_typed_reverse_safe(nir_instr, instr, node, &(block)->instr_list)
2795 
2796 static inline nir_phi_instr *
nir_block_last_phi_instr(nir_block * block)2797 nir_block_last_phi_instr(nir_block *block)
2798 {
2799    nir_phi_instr *last_phi = NULL;
2800    nir_foreach_instr(instr, block) {
2801       if (instr->type == nir_instr_type_phi)
2802          last_phi = nir_instr_as_phi(instr);
2803       else
2804          return last_phi;
2805    }
2806    return last_phi;
2807 }
2808 
2809 typedef enum {
2810    nir_selection_control_none = 0x0,
2811    nir_selection_control_flatten = 0x1,
2812    nir_selection_control_dont_flatten = 0x2,
2813 } nir_selection_control;
2814 
2815 typedef struct nir_if {
2816    nir_cf_node cf_node;
2817    nir_src condition;
2818    nir_selection_control control;
2819 
2820    struct exec_list then_list; /** < list of nir_cf_node */
2821    struct exec_list else_list; /** < list of nir_cf_node */
2822 } nir_if;
2823 
2824 typedef struct {
2825    nir_if *nif;
2826 
2827    /** Instruction that generates nif::condition. */
2828    nir_instr *conditional_instr;
2829 
2830    /** Block within ::nif that has the break instruction. */
2831    nir_block *break_block;
2832 
2833    /** Last block for the then- or else-path that does not contain the break. */
2834    nir_block *continue_from_block;
2835 
2836    /** True when ::break_block is in the else-path of ::nif. */
2837    bool continue_from_then;
2838    bool induction_rhs;
2839 
2840    /* This is true if the terminators exact trip count is unknown. For
2841     * example:
2842     *
2843     *    for (int i = 0; i < imin(x, 4); i++)
2844     *       ...
2845     *
2846     * Here loop analysis would have set a max_trip_count of 4 however we dont
2847     * know for sure that this is the exact trip count.
2848     */
2849    bool exact_trip_count_unknown;
2850 
2851    struct list_head loop_terminator_link;
2852 } nir_loop_terminator;
2853 
2854 typedef struct {
2855    /* Induction variable. */
2856    nir_ssa_def *def;
2857 
2858    /* Init statement with only uniform. */
2859    nir_src *init_src;
2860 
2861    /* Update statement with only uniform. */
2862    nir_alu_src *update_src;
2863 } nir_loop_induction_variable;
2864 
2865 typedef struct {
2866    /* Estimated cost (in number of instructions) of the loop */
2867    unsigned instr_cost;
2868 
2869    /* Guessed trip count based on array indexing */
2870    unsigned guessed_trip_count;
2871 
2872    /* Maximum number of times the loop is run (if known) */
2873    unsigned max_trip_count;
2874 
2875    /* Do we know the exact number of times the loop will be run */
2876    bool exact_trip_count_known;
2877 
2878    /* Unroll the loop regardless of its size */
2879    bool force_unroll;
2880 
2881    /* Does the loop contain complex loop terminators, continues or other
2882     * complex behaviours? If this is true we can't rely on
2883     * loop_terminator_list to be complete or accurate.
2884     */
2885    bool complex_loop;
2886 
2887    nir_loop_terminator *limiting_terminator;
2888 
2889    /* A list of loop_terminators terminating this loop. */
2890    struct list_head loop_terminator_list;
2891 
2892    /* array of induction variables for this loop */
2893    nir_loop_induction_variable *induction_vars;
2894    unsigned num_induction_vars;
2895 } nir_loop_info;
2896 
2897 typedef enum {
2898    nir_loop_control_none = 0x0,
2899    nir_loop_control_unroll = 0x1,
2900    nir_loop_control_dont_unroll = 0x2,
2901 } nir_loop_control;
2902 
2903 typedef struct {
2904    nir_cf_node cf_node;
2905 
2906    struct exec_list body; /** < list of nir_cf_node */
2907 
2908    nir_loop_info *info;
2909    nir_loop_control control;
2910    bool partially_unrolled;
2911    bool divergent;
2912 } nir_loop;
2913 
2914 /**
2915  * Various bits of metadata that can may be created or required by
2916  * optimization and analysis passes
2917  */
2918 typedef enum {
2919    nir_metadata_none = 0x0,
2920 
2921    /** Indicates that nir_block::index values are valid.
2922     *
2923     * The start block has index 0 and they increase through a natural walk of
2924     * the CFG.  nir_function_impl::num_blocks is the number of blocks and
2925     * every block index is in the range [0, nir_function_impl::num_blocks].
2926     *
2927     * A pass can preserve this metadata type if it doesn't touch the CFG.
2928     */
2929    nir_metadata_block_index = 0x1,
2930 
2931    /** Indicates that block dominance information is valid
2932     *
2933     * This includes:
2934     *
2935     *   - nir_block::num_dom_children
2936     *   - nir_block::dom_children
2937     *   - nir_block::dom_frontier
2938     *   - nir_block::dom_pre_index
2939     *   - nir_block::dom_post_index
2940     *
2941     * A pass can preserve this metadata type if it doesn't touch the CFG.
2942     */
2943    nir_metadata_dominance = 0x2,
2944 
2945    /** Indicates that SSA def data-flow liveness information is valid
2946     *
2947     * This includes:
2948     *
2949     *   - nir_block::live_in
2950     *   - nir_block::live_out
2951     *
2952     * A pass can preserve this metadata type if it never adds or removes any
2953     * SSA defs or uses of SSA defs (most passes shouldn't preserve this
2954     * metadata type).
2955     */
2956    nir_metadata_live_ssa_defs = 0x4,
2957 
2958    /** A dummy metadata value to track when a pass forgot to call
2959     * nir_metadata_preserve.
2960     *
2961     * A pass should always clear this value even if it doesn't make any
2962     * progress to indicate that it thought about preserving metadata.
2963     */
2964    nir_metadata_not_properly_reset = 0x8,
2965 
2966    /** Indicates that loop analysis information is valid.
2967     *
2968     * This includes everything pointed to by nir_loop::info.
2969     *
2970     * A pass can preserve this metadata type if it is guaranteed to not affect
2971     * any loop metadata.  However, since loop metadata includes things like
2972     * loop counts which depend on arithmetic in the loop, this is very hard to
2973     * determine.  Most passes shouldn't preserve this metadata type.
2974     */
2975    nir_metadata_loop_analysis = 0x10,
2976 
2977    /** Indicates that nir_instr::index values are valid.
2978     *
2979     * The start instruction has index 0 and they increase through a natural
2980     * walk of instructions in blocks in the CFG.  The indices my have holes
2981     * after passes such as DCE.
2982     *
2983     * A pass can preserve this metadata type if it never adds or moves any
2984     * instructions (most passes shouldn't preserve this metadata type), but
2985     * can preserve it if it only removes instructions.
2986     */
2987    nir_metadata_instr_index = 0x20,
2988 
2989    /** All metadata
2990     *
2991     * This includes all nir_metadata flags except not_properly_reset.  Passes
2992     * which do not change the shader in any way should call
2993     *
2994     *    nir_metadata_preserve(impl, nir_metadata_all);
2995     */
2996    nir_metadata_all = ~nir_metadata_not_properly_reset,
2997 } nir_metadata;
2998 MESA_DEFINE_CPP_ENUM_BITFIELD_OPERATORS(nir_metadata)
2999 
3000 typedef struct {
3001    nir_cf_node cf_node;
3002 
3003    /** pointer to the function of which this is an implementation */
3004    struct nir_function *function;
3005 
3006    /**
3007     * For entrypoints, a pointer to a nir_function_impl which runs before
3008     * it, once per draw or dispatch, communicating via store_preamble and
3009     * load_preamble intrinsics. If NULL then there is no preamble.
3010     */
3011    struct nir_function *preamble;
3012 
3013    struct exec_list body; /** < list of nir_cf_node */
3014 
3015    nir_block *end_block;
3016 
3017    /** list for all local variables in the function */
3018    struct exec_list locals;
3019 
3020    /** list of local registers in the function */
3021    struct exec_list registers;
3022 
3023    /** next available local register index */
3024    unsigned reg_alloc;
3025 
3026    /** next available SSA value index */
3027    unsigned ssa_alloc;
3028 
3029    /* total number of basic blocks, only valid when block_index_dirty = false */
3030    unsigned num_blocks;
3031 
3032    /** True if this nir_function_impl uses structured control-flow
3033     *
3034     * Structured nir_function_impls have different validation rules.
3035     */
3036    bool structured;
3037 
3038    nir_metadata valid_metadata;
3039 } nir_function_impl;
3040 
3041 #define nir_foreach_function_temp_variable(var, impl) \
3042    foreach_list_typed(nir_variable, var, node, &(impl)->locals)
3043 
3044 #define nir_foreach_function_temp_variable_safe(var, impl) \
3045    foreach_list_typed_safe(nir_variable, var, node, &(impl)->locals)
3046 
3047 ATTRIBUTE_RETURNS_NONNULL static inline nir_block *
nir_start_block(nir_function_impl * impl)3048 nir_start_block(nir_function_impl *impl)
3049 {
3050    return (nir_block *) impl->body.head_sentinel.next;
3051 }
3052 
3053 ATTRIBUTE_RETURNS_NONNULL static inline nir_block *
nir_impl_last_block(nir_function_impl * impl)3054 nir_impl_last_block(nir_function_impl *impl)
3055 {
3056    return (nir_block *) impl->body.tail_sentinel.prev;
3057 }
3058 
3059 static inline nir_cf_node *
nir_cf_node_next(nir_cf_node * node)3060 nir_cf_node_next(nir_cf_node *node)
3061 {
3062    struct exec_node *next = exec_node_get_next(&node->node);
3063    if (exec_node_is_tail_sentinel(next))
3064       return NULL;
3065    else
3066       return exec_node_data(nir_cf_node, next, node);
3067 }
3068 
3069 static inline nir_cf_node *
nir_cf_node_prev(nir_cf_node * node)3070 nir_cf_node_prev(nir_cf_node *node)
3071 {
3072    struct exec_node *prev = exec_node_get_prev(&node->node);
3073    if (exec_node_is_head_sentinel(prev))
3074       return NULL;
3075    else
3076       return exec_node_data(nir_cf_node, prev, node);
3077 }
3078 
3079 static inline bool
nir_cf_node_is_first(const nir_cf_node * node)3080 nir_cf_node_is_first(const nir_cf_node *node)
3081 {
3082    return exec_node_is_head_sentinel(node->node.prev);
3083 }
3084 
3085 static inline bool
nir_cf_node_is_last(const nir_cf_node * node)3086 nir_cf_node_is_last(const nir_cf_node *node)
3087 {
3088    return exec_node_is_tail_sentinel(node->node.next);
3089 }
3090 
NIR_DEFINE_CAST(nir_cf_node_as_block,nir_cf_node,nir_block,cf_node,type,nir_cf_node_block)3091 NIR_DEFINE_CAST(nir_cf_node_as_block, nir_cf_node, nir_block, cf_node,
3092                 type, nir_cf_node_block)
3093 NIR_DEFINE_CAST(nir_cf_node_as_if, nir_cf_node, nir_if, cf_node,
3094                 type, nir_cf_node_if)
3095 NIR_DEFINE_CAST(nir_cf_node_as_loop, nir_cf_node, nir_loop, cf_node,
3096                 type, nir_cf_node_loop)
3097 NIR_DEFINE_CAST(nir_cf_node_as_function, nir_cf_node,
3098                 nir_function_impl, cf_node, type, nir_cf_node_function)
3099 
3100 static inline nir_block *
3101 nir_if_first_then_block(nir_if *if_stmt)
3102 {
3103    struct exec_node *head = exec_list_get_head(&if_stmt->then_list);
3104    return nir_cf_node_as_block(exec_node_data(nir_cf_node, head, node));
3105 }
3106 
3107 static inline nir_block *
nir_if_last_then_block(nir_if * if_stmt)3108 nir_if_last_then_block(nir_if *if_stmt)
3109 {
3110    struct exec_node *tail = exec_list_get_tail(&if_stmt->then_list);
3111    return nir_cf_node_as_block(exec_node_data(nir_cf_node, tail, node));
3112 }
3113 
3114 static inline nir_block *
nir_if_first_else_block(nir_if * if_stmt)3115 nir_if_first_else_block(nir_if *if_stmt)
3116 {
3117    struct exec_node *head = exec_list_get_head(&if_stmt->else_list);
3118    return nir_cf_node_as_block(exec_node_data(nir_cf_node, head, node));
3119 }
3120 
3121 static inline nir_block *
nir_if_last_else_block(nir_if * if_stmt)3122 nir_if_last_else_block(nir_if *if_stmt)
3123 {
3124    struct exec_node *tail = exec_list_get_tail(&if_stmt->else_list);
3125    return nir_cf_node_as_block(exec_node_data(nir_cf_node, tail, node));
3126 }
3127 
3128 static inline nir_block *
nir_loop_first_block(nir_loop * loop)3129 nir_loop_first_block(nir_loop *loop)
3130 {
3131    struct exec_node *head = exec_list_get_head(&loop->body);
3132    return nir_cf_node_as_block(exec_node_data(nir_cf_node, head, node));
3133 }
3134 
3135 static inline nir_block *
nir_loop_last_block(nir_loop * loop)3136 nir_loop_last_block(nir_loop *loop)
3137 {
3138    struct exec_node *tail = exec_list_get_tail(&loop->body);
3139    return nir_cf_node_as_block(exec_node_data(nir_cf_node, tail, node));
3140 }
3141 
3142 /**
3143  * Return true if this list of cf_nodes contains a single empty block.
3144  */
3145 static inline bool
nir_cf_list_is_empty_block(struct exec_list * cf_list)3146 nir_cf_list_is_empty_block(struct exec_list *cf_list)
3147 {
3148    if (exec_list_is_singular(cf_list)) {
3149       struct exec_node *head = exec_list_get_head(cf_list);
3150       nir_block *block =
3151          nir_cf_node_as_block(exec_node_data(nir_cf_node, head, node));
3152       return exec_list_is_empty(&block->instr_list);
3153    }
3154    return false;
3155 }
3156 
3157 typedef struct {
3158    uint8_t num_components;
3159    uint8_t bit_size;
3160 } nir_parameter;
3161 
3162 typedef struct nir_printf_info {
3163    unsigned num_args;
3164    unsigned *arg_sizes;
3165    unsigned string_size;
3166    char *strings;
3167 } nir_printf_info;
3168 
3169 typedef struct nir_function {
3170    struct exec_node node;
3171 
3172    const char *name;
3173    struct nir_shader *shader;
3174 
3175    unsigned num_params;
3176    nir_parameter *params;
3177 
3178    /** The implementation of this function.
3179     *
3180     * If the function is only declared and not implemented, this is NULL.
3181     */
3182    nir_function_impl *impl;
3183 
3184    bool is_entrypoint;
3185    bool is_preamble;
3186 } nir_function;
3187 
3188 typedef enum {
3189    nir_lower_imul64 = (1 << 0),
3190    nir_lower_isign64 = (1 << 1),
3191    /** Lower all int64 modulus and division opcodes */
3192    nir_lower_divmod64 = (1 << 2),
3193    /** Lower all 64-bit umul_high and imul_high opcodes */
3194    nir_lower_imul_high64 = (1 << 3),
3195    nir_lower_mov64 = (1 << 4),
3196    nir_lower_icmp64 = (1 << 5),
3197    nir_lower_iadd64 = (1 << 6),
3198    nir_lower_iabs64 = (1 << 7),
3199    nir_lower_ineg64 = (1 << 8),
3200    nir_lower_logic64 = (1 << 9),
3201    nir_lower_minmax64 = (1 << 10),
3202    nir_lower_shift64 = (1 << 11),
3203    nir_lower_imul_2x32_64 = (1 << 12),
3204    nir_lower_extract64 = (1 << 13),
3205    nir_lower_ufind_msb64 = (1 << 14),
3206    nir_lower_bit_count64 = (1 << 15),
3207    nir_lower_subgroup_shuffle64 = (1 << 16),
3208    nir_lower_scan_reduce_bitwise64 = (1 << 17),
3209    nir_lower_scan_reduce_iadd64 = (1 << 18),
3210    nir_lower_vote_ieq64 = (1 << 19),
3211    nir_lower_usub_sat64 = (1 << 20),
3212    nir_lower_iadd_sat64 = (1 << 21),
3213 } nir_lower_int64_options;
3214 
3215 typedef enum {
3216    nir_lower_drcp = (1 << 0),
3217    nir_lower_dsqrt = (1 << 1),
3218    nir_lower_drsq = (1 << 2),
3219    nir_lower_dtrunc = (1 << 3),
3220    nir_lower_dfloor = (1 << 4),
3221    nir_lower_dceil = (1 << 5),
3222    nir_lower_dfract = (1 << 6),
3223    nir_lower_dround_even = (1 << 7),
3224    nir_lower_dmod = (1 << 8),
3225    nir_lower_dsub = (1 << 9),
3226    nir_lower_ddiv = (1 << 10),
3227    nir_lower_fp64_full_software = (1 << 11),
3228 } nir_lower_doubles_options;
3229 
3230 typedef enum {
3231    nir_divergence_single_prim_per_subgroup = (1 << 0),
3232    nir_divergence_single_patch_per_tcs_subgroup = (1 << 1),
3233    nir_divergence_single_patch_per_tes_subgroup = (1 << 2),
3234    nir_divergence_view_index_uniform = (1 << 3),
3235    nir_divergence_single_frag_shading_rate_per_subgroup = (1 << 4),
3236    nir_divergence_multiple_workgroup_per_compute_subgroup = (1 << 5),
3237 } nir_divergence_options;
3238 
3239 typedef enum {
3240    nir_pack_varying_interp_mode_none          = (1 << 0),
3241    nir_pack_varying_interp_mode_smooth        = (1 << 1),
3242    nir_pack_varying_interp_mode_flat          = (1 << 2),
3243    nir_pack_varying_interp_mode_noperspective = (1 << 3),
3244    nir_pack_varying_interp_loc_sample         = (1 << 16),
3245    nir_pack_varying_interp_loc_centroid       = (1 << 17),
3246    nir_pack_varying_interp_loc_center         = (1 << 18),
3247 } nir_pack_varying_options;
3248 
3249 /** An instruction filtering callback
3250  *
3251  * Returns true if the instruction should be processed and false otherwise.
3252  */
3253 typedef bool (*nir_instr_filter_cb)(const nir_instr *, const void *);
3254 
3255 /** A vectorization width callback
3256  *
3257  * Returns the maximum vectorization width per instruction.
3258  * 0, if the instruction must not be modified.
3259  *
3260  * The vectorization width must be a power of 2.
3261  */
3262 typedef uint8_t (*nir_vectorize_cb)(const nir_instr *, const void *);
3263 
3264 typedef struct nir_shader_compiler_options {
3265    bool lower_fdiv;
3266    bool lower_ffma16;
3267    bool lower_ffma32;
3268    bool lower_ffma64;
3269    bool fuse_ffma16;
3270    bool fuse_ffma32;
3271    bool fuse_ffma64;
3272    bool lower_flrp16;
3273    bool lower_flrp32;
3274    /** Lowers flrp when it does not support doubles */
3275    bool lower_flrp64;
3276    bool lower_fpow;
3277    bool lower_fsat;
3278    bool lower_fsqrt;
3279    bool lower_sincos;
3280    bool lower_fmod;
3281    /** Lowers ibitfield_extract/ubitfield_extract to ibfe/ubfe. */
3282    bool lower_bitfield_extract;
3283    /** Lowers ibitfield_extract/ubitfield_extract to compares, shifts. */
3284    bool lower_bitfield_extract_to_shifts;
3285    /** Lowers bitfield_insert to bfi/bfm */
3286    bool lower_bitfield_insert;
3287    /** Lowers bitfield_insert to compares, and shifts. */
3288    bool lower_bitfield_insert_to_shifts;
3289    /** Lowers bitfield_insert to bfm/bitfield_select. */
3290    bool lower_bitfield_insert_to_bitfield_select;
3291    /** Lowers bitfield_reverse to shifts. */
3292    bool lower_bitfield_reverse;
3293    /** Lowers bit_count to shifts. */
3294    bool lower_bit_count;
3295    /** Lowers ifind_msb to compare and ufind_msb */
3296    bool lower_ifind_msb;
3297    /** Lowers ifind_msb and ufind_msb to reverse variants */
3298    bool lower_find_msb_to_reverse;
3299    /** Lowers find_lsb to ufind_msb and logic ops */
3300    bool lower_find_lsb;
3301    bool lower_uadd_carry;
3302    bool lower_usub_borrow;
3303    /** Lowers imul_high/umul_high to 16-bit multiplies and carry operations. */
3304    bool lower_mul_high;
3305    /** lowers fneg to fmul(x, -1.0). Driver must call nir_opt_algebraic_late() */
3306    bool lower_fneg;
3307    /** lowers ineg to isub. Driver must call nir_opt_algebraic_late(). */
3308    bool lower_ineg;
3309    /** lowers fisnormal to alu ops. */
3310    bool lower_fisnormal;
3311 
3312    /* lower {slt,sge,seq,sne} to {flt,fge,feq,fneu} + b2f: */
3313    bool lower_scmp;
3314 
3315    /* lower b/fall_equalN/b/fany_nequalN (ex:fany_nequal4 to sne+fdot4+fsat) */
3316    bool lower_vector_cmp;
3317 
3318    /** enable rules to avoid bit ops */
3319    bool lower_bitops;
3320 
3321    /** enables rules to lower isign to imin+imax */
3322    bool lower_isign;
3323 
3324    /** enables rules to lower fsign to fsub and flt */
3325    bool lower_fsign;
3326 
3327    /** enables rules to lower iabs to ineg+imax */
3328    bool lower_iabs;
3329 
3330    /** enable rules that avoid generating umax from signed integer ops */
3331    bool lower_umax;
3332 
3333    /** enable rules that avoid generating umin from signed integer ops */
3334    bool lower_umin;
3335 
3336    /* lower fdph to fdot4 */
3337    bool lower_fdph;
3338 
3339    /** lower fdot to fmul and fsum/fadd. */
3340    bool lower_fdot;
3341 
3342    /* Does the native fdot instruction replicate its result for four
3343     * components?  If so, then opt_algebraic_late will turn all fdotN
3344     * instructions into fdotN_replicated instructions.
3345     */
3346    bool fdot_replicates;
3347 
3348    /** lowers ffloor to fsub+ffract: */
3349    bool lower_ffloor;
3350 
3351    /** lowers ffract to fsub+ffloor: */
3352    bool lower_ffract;
3353 
3354    /** lowers fceil to fneg+ffloor+fneg: */
3355    bool lower_fceil;
3356 
3357    bool lower_ftrunc;
3358 
3359    /** Lowers fround_even to ffract+feq+csel.
3360     *
3361     * Not correct in that it doesn't correctly handle the "_even" part of the
3362     * rounding, but good enough for DX9 array indexing handling on DX9-class
3363     * hardware.
3364     */
3365    bool lower_fround_even;
3366 
3367    bool lower_ldexp;
3368 
3369    bool lower_pack_half_2x16;
3370    bool lower_pack_unorm_2x16;
3371    bool lower_pack_snorm_2x16;
3372    bool lower_pack_unorm_4x8;
3373    bool lower_pack_snorm_4x8;
3374    bool lower_pack_64_2x32;
3375    bool lower_pack_64_4x16;
3376    bool lower_pack_32_2x16;
3377    bool lower_pack_64_2x32_split;
3378    bool lower_pack_32_2x16_split;
3379    bool lower_unpack_half_2x16;
3380    bool lower_unpack_unorm_2x16;
3381    bool lower_unpack_snorm_2x16;
3382    bool lower_unpack_unorm_4x8;
3383    bool lower_unpack_snorm_4x8;
3384    bool lower_unpack_64_2x32_split;
3385    bool lower_unpack_32_2x16_split;
3386 
3387    bool lower_pack_split;
3388 
3389    bool lower_extract_byte;
3390    bool lower_extract_word;
3391    bool lower_insert_byte;
3392    bool lower_insert_word;
3393 
3394    bool lower_all_io_to_temps;
3395    bool lower_all_io_to_elements;
3396 
3397    /* Indicates that the driver only has zero-based vertex id */
3398    bool vertex_id_zero_based;
3399 
3400    /**
3401     * If enabled, gl_BaseVertex will be lowered as:
3402     * is_indexed_draw (~0/0) & firstvertex
3403     */
3404    bool lower_base_vertex;
3405 
3406    /**
3407     * If enabled, gl_HelperInvocation will be lowered as:
3408     *
3409     *   !((1 << sample_id) & sample_mask_in))
3410     *
3411     * This depends on some possibly hw implementation details, which may
3412     * not be true for all hw.  In particular that the FS is only executed
3413     * for covered samples or for helper invocations.  So, do not blindly
3414     * enable this option.
3415     *
3416     * Note: See also issue #22 in ARB_shader_image_load_store
3417     */
3418    bool lower_helper_invocation;
3419 
3420    /**
3421     * Convert gl_SampleMaskIn to gl_HelperInvocation as follows:
3422     *
3423     *   gl_SampleMaskIn == 0 ---> gl_HelperInvocation
3424     *   gl_SampleMaskIn != 0 ---> !gl_HelperInvocation
3425     */
3426    bool optimize_sample_mask_in;
3427 
3428    bool lower_cs_local_index_to_id;
3429    bool lower_cs_local_id_to_index;
3430 
3431    /* Prevents lowering global_invocation_id to be in terms of workgroup_id */
3432    bool has_cs_global_id;
3433 
3434    bool lower_device_index_to_zero;
3435 
3436    /* Set if nir_lower_pntc_ytransform() should invert gl_PointCoord.
3437     * Either when frame buffer is flipped or GL_POINT_SPRITE_COORD_ORIGIN
3438     * is GL_LOWER_LEFT.
3439     */
3440    bool lower_wpos_pntc;
3441 
3442    /**
3443     * Set if nir_op_[iu]hadd and nir_op_[iu]rhadd instructions should be
3444     * lowered to simple arithmetic.
3445     *
3446     * If this flag is set, the lowering will be applied to all bit-sizes of
3447     * these instructions.
3448     *
3449     * \sa ::lower_hadd64
3450     */
3451    bool lower_hadd;
3452 
3453    /**
3454     * Set if only 64-bit nir_op_[iu]hadd and nir_op_[iu]rhadd instructions
3455     * should be lowered to simple arithmetic.
3456     *
3457     * If this flag is set, the lowering will be applied to only 64-bit
3458     * versions of these instructions.
3459     *
3460     * \sa ::lower_hadd
3461     */
3462    bool lower_hadd64;
3463 
3464    /**
3465     * Set if nir_op_uadd_sat should be lowered to simple arithmetic.
3466     *
3467     * If this flag is set, the lowering will be applied to all bit-sizes of
3468     * these instructions.
3469     */
3470    bool lower_uadd_sat;
3471 
3472    /**
3473     * Set if nir_op_usub_sat should be lowered to simple arithmetic.
3474     *
3475     * If this flag is set, the lowering will be applied to all bit-sizes of
3476     * these instructions.
3477     */
3478    bool lower_usub_sat;
3479 
3480    /**
3481     * Set if nir_op_iadd_sat and nir_op_isub_sat should be lowered to simple
3482     * arithmetic.
3483     *
3484     * If this flag is set, the lowering will be applied to all bit-sizes of
3485     * these instructions.
3486     */
3487    bool lower_iadd_sat;
3488 
3489    /**
3490     * Set if imul_32x16 and umul_32x16 should be lowered to simple
3491     * arithmetic.
3492     */
3493    bool lower_mul_32x16;
3494 
3495    /**
3496     * Should IO be re-vectorized?  Some scalar ISAs still operate on vec4's
3497     * for IO purposes and would prefer loads/stores be vectorized.
3498     */
3499    bool vectorize_io;
3500    bool lower_to_scalar;
3501    nir_instr_filter_cb lower_to_scalar_filter;
3502 
3503    /**
3504     * Disables potentially harmful algebraic transformations for architectures
3505     * with SIMD-within-a-register semantics.
3506     *
3507     * Note, to actually vectorize 16bit instructions, use nir_opt_vectorize()
3508     * with a suitable callback function.
3509     */
3510    bool vectorize_vec2_16bit;
3511 
3512    /**
3513     * Should the linker unify inputs_read/outputs_written between adjacent
3514     * shader stages which are linked into a single program?
3515     */
3516    bool unify_interfaces;
3517 
3518    /**
3519     * Should nir_lower_io() create load_interpolated_input intrinsics?
3520     *
3521     * If not, it generates regular load_input intrinsics and interpolation
3522     * information must be inferred from the list of input nir_variables.
3523     */
3524    bool use_interpolated_input_intrinsics;
3525 
3526 
3527    /**
3528     * Whether nir_lower_io() will lower interpolateAt functions to
3529     * load_interpolated_input intrinsics.
3530     *
3531     * Unlike use_interpolated_input_intrinsics this will only lower these
3532     * functions and leave input load intrinsics untouched.
3533     */
3534    bool lower_interpolate_at;
3535 
3536    /* Lowers when 32x32->64 bit multiplication is not supported */
3537    bool lower_mul_2x32_64;
3538 
3539    /* Lowers when rotate instruction is not supported */
3540    bool lower_rotate;
3541 
3542    /** Backend supports ternary addition */
3543    bool has_iadd3;
3544 
3545    /**
3546     * Backend supports imul24, and would like to use it (when possible)
3547     * for address/offset calculation.  If true, driver should call
3548     * nir_lower_amul().  (If not set, amul will automatically be lowered
3549     * to imul.)
3550     */
3551    bool has_imul24;
3552 
3553    /** Backend supports umul24, if not set  umul24 will automatically be lowered
3554     * to imul with masked inputs */
3555    bool has_umul24;
3556 
3557    /** Backend supports umad24, if not set  umad24 will automatically be lowered
3558     * to imul with masked inputs and iadd */
3559    bool has_umad24;
3560 
3561    /* Backend supports fused comapre against zero and csel */
3562    bool has_fused_comp_and_csel;
3563 
3564    /** Backend supports fsub, if not set fsub will automatically be lowered to
3565     * fadd(x, fneg(y)). If true, driver should call nir_opt_algebraic_late(). */
3566    bool has_fsub;
3567 
3568    /** Backend supports isub, if not set isub will automatically be lowered to
3569     * iadd(x, ineg(y)). If true, driver should call nir_opt_algebraic_late(). */
3570    bool has_isub;
3571 
3572    /** Backend supports pack_32_4x8 or pack_32_4x8_split. */
3573    bool has_pack_32_4x8;
3574 
3575    /** Backend supports txs, if not nir_lower_tex(..) uses txs-free variants
3576     * for rect texture lowering. */
3577    bool has_txs;
3578 
3579    /** Backend supports sdot_4x8 opcodes. */
3580    bool has_sdot_4x8;
3581 
3582    /** Backend supports udot_4x8 opcodes. */
3583    bool has_udot_4x8;
3584 
3585    /** Backend supports sudot_4x8 opcodes. */
3586    bool has_sudot_4x8;
3587 
3588    /** Backend supports sdot_2x16 and udot_2x16 opcodes. */
3589    bool has_dot_2x16;
3590 
3591    /* Whether to generate only scoped_barrier intrinsics instead of the set of
3592     * memory and control barrier intrinsics based on GLSL.
3593     */
3594    bool use_scoped_barrier;
3595 
3596    /** Backend supports fmulz (and ffmaz if lower_ffma32=false) */
3597    bool has_fmulz;
3598 
3599    /**
3600     * Is this the Intel vec4 backend?
3601     *
3602     * Used to inhibit algebraic optimizations that are known to be harmful on
3603     * the Intel vec4 backend.  This is generally applicable to any
3604     * optimization that might cause more immediate values to be used in
3605     * 3-source (e.g., ffma and flrp) instructions.
3606     */
3607    bool intel_vec4;
3608 
3609    /**
3610     * For most Intel GPUs, all ternary operations such as FMA and BFE cannot
3611     * have immediates, so two to three instructions may eventually be needed.
3612     */
3613    bool avoid_ternary_with_two_constants;
3614 
3615    /** Whether 8-bit ALU is supported. */
3616    bool support_8bit_alu;
3617 
3618    /** Whether 16-bit ALU is supported. */
3619    bool support_16bit_alu;
3620 
3621    unsigned max_unroll_iterations;
3622    unsigned max_unroll_iterations_aggressive;
3623 
3624    bool lower_uniforms_to_ubo;
3625 
3626    /* If the precision is ignored, backends that don't handle
3627     * different precisions when passing data between stages and use
3628     * vectorized IO can pack more varyings when linking. */
3629    bool linker_ignore_precision;
3630 
3631    /* Specifies if indirect sampler array access will trigger forced loop
3632     * unrolling.
3633     */
3634    bool force_indirect_unrolling_sampler;
3635 
3636    /* Some older drivers don't support GLSL versions with the concept of flat
3637     * varyings and also don't support integers. This setting helps us avoid
3638     * marking varyings as flat and potentially having them changed to ints via
3639     * varying packing.
3640     */
3641    bool no_integers;
3642 
3643    /**
3644     * Specifies which type of indirectly accessed variables should force
3645     * loop unrolling.
3646     */
3647    nir_variable_mode force_indirect_unrolling;
3648 
3649    nir_lower_int64_options lower_int64_options;
3650    nir_lower_doubles_options lower_doubles_options;
3651    nir_divergence_options divergence_analysis_options;
3652 
3653    /**
3654     * Support pack varyings with different interpolation location
3655     * (center, centroid, sample) and mode (flat, noperspective, smooth)
3656     * into same slot.
3657     */
3658    nir_pack_varying_options pack_varying_options;
3659 
3660    /**
3661     * Lower load_deref/store_deref of inputs and outputs into
3662     * load_input/store_input intrinsics. This is used by nir_lower_io_passes.
3663     */
3664    bool lower_io_variables;
3665 
3666    /**
3667     * Lower color inputs to load_colorN that are kind of like system values
3668     * if lower_io_variables is also set. shader_info will contain
3669     * the interpolation settings. This is used by nir_lower_io_passes.
3670     */
3671    bool lower_fs_color_inputs;
3672 
3673    /**
3674     * The masks of shader stages that support indirect indexing with
3675     * load_input and store_output intrinsics. It's used when
3676     * lower_io_variables is true. This is used by nir_lower_io_passes.
3677     */
3678    uint8_t support_indirect_inputs;
3679    uint8_t support_indirect_outputs;
3680 
3681    /**
3682     * Remove varying loaded from uniform, let fragment shader load the
3683     * uniform directly. GPU passing varying by memory can benifit from it
3684     * for sure; but GPU passing varying by on chip resource may not.
3685     * Because it saves on chip resource but may increase memory pressure when
3686     * fragment task is far more than vertex one, so better left it disabled.
3687     */
3688    bool lower_varying_from_uniform;
3689 } nir_shader_compiler_options;
3690 
3691 typedef struct nir_shader {
3692    /** list of uniforms (nir_variable) */
3693    struct exec_list variables;
3694 
3695    /** Set of driver-specific options for the shader.
3696     *
3697     * The memory for the options is expected to be kept in a single static
3698     * copy by the driver.
3699     */
3700    const struct nir_shader_compiler_options *options;
3701 
3702    /** Various bits of compile-time information about a given shader */
3703    struct shader_info info;
3704 
3705    struct exec_list functions; /** < list of nir_function */
3706 
3707    struct list_head gc_list; /** < list of all nir_instrs allocated on the shader but not yet freed. */
3708 
3709    /**
3710     * The size of the variable space for load_input_*, load_uniform_*, etc.
3711     * intrinsics.  This is in back-end specific units which is likely one of
3712     * bytes, dwords, or vec4s depending on context and back-end.
3713     */
3714    unsigned num_inputs, num_uniforms, num_outputs;
3715 
3716    /** Size in bytes of required implicitly bound global memory */
3717    unsigned global_mem_size;
3718 
3719    /** Size in bytes of required scratch space */
3720    unsigned scratch_size;
3721 
3722    /** Constant data associated with this shader.
3723     *
3724     * Constant data is loaded through load_constant intrinsics (as compared to
3725     * the NIR load_const instructions which have the constant value inlined
3726     * into them).  This is usually generated by nir_opt_large_constants (so
3727     * shaders don't have to load_const into a temporary array when they want
3728     * to indirect on a const array).
3729     */
3730    void *constant_data;
3731    /** Size of the constant data associated with the shader, in bytes */
3732    unsigned constant_data_size;
3733 
3734    struct nir_xfb_info *xfb_info;
3735 
3736    unsigned printf_info_count;
3737    nir_printf_info *printf_info;
3738 } nir_shader;
3739 
3740 #define nir_foreach_function(func, shader) \
3741    foreach_list_typed(nir_function, func, node, &(shader)->functions)
3742 
3743 static inline nir_function_impl *
nir_shader_get_entrypoint(const nir_shader * shader)3744 nir_shader_get_entrypoint(const nir_shader *shader)
3745 {
3746    nir_function *func = NULL;
3747 
3748    nir_foreach_function(function, shader) {
3749       assert(func == NULL);
3750       if (function->is_entrypoint) {
3751          func = function;
3752 #ifndef NDEBUG
3753          break;
3754 #endif
3755       }
3756    }
3757 
3758    if (!func)
3759       return NULL;
3760 
3761    assert(func->num_params == 0);
3762    assert(func->impl);
3763    return func->impl;
3764 }
3765 
3766 void nir_remove_non_entrypoints(nir_shader *shader);
3767 
3768 nir_shader *nir_shader_create(void *mem_ctx,
3769                               gl_shader_stage stage,
3770                               const nir_shader_compiler_options *options,
3771                               shader_info *si);
3772 
3773 nir_register *nir_local_reg_create(nir_function_impl *impl);
3774 
3775 void nir_reg_remove(nir_register *reg);
3776 
3777 /** Adds a variable to the appropriate list in nir_shader */
3778 void nir_shader_add_variable(nir_shader *shader, nir_variable *var);
3779 
3780 static inline void
nir_function_impl_add_variable(nir_function_impl * impl,nir_variable * var)3781 nir_function_impl_add_variable(nir_function_impl *impl, nir_variable *var)
3782 {
3783    assert(var->data.mode == nir_var_function_temp);
3784    exec_list_push_tail(&impl->locals, &var->node);
3785 }
3786 
3787 /** creates a variable, sets a few defaults, and adds it to the list */
3788 nir_variable *nir_variable_create(nir_shader *shader,
3789                                   nir_variable_mode mode,
3790                                   const struct glsl_type *type,
3791                                   const char *name);
3792 /** creates a local variable and adds it to the list */
3793 nir_variable *nir_local_variable_create(nir_function_impl *impl,
3794                                         const struct glsl_type *type,
3795                                         const char *name);
3796 
3797 nir_variable *nir_find_variable_with_location(nir_shader *shader,
3798                                               nir_variable_mode mode,
3799                                               unsigned location);
3800 
3801 nir_variable *nir_find_variable_with_driver_location(nir_shader *shader,
3802                                                      nir_variable_mode mode,
3803                                                      unsigned location);
3804 
3805 void nir_sort_variables_with_modes(nir_shader *shader,
3806                                    int (*compar)(const nir_variable *,
3807                                                  const nir_variable *),
3808                                    nir_variable_mode modes);
3809 
3810 /** creates a function and adds it to the shader's list of functions */
3811 nir_function *nir_function_create(nir_shader *shader, const char *name);
3812 
3813 nir_function_impl *nir_function_impl_create(nir_function *func);
3814 /** creates a function_impl that isn't tied to any particular function */
3815 nir_function_impl *nir_function_impl_create_bare(nir_shader *shader);
3816 
3817 nir_block *nir_block_create(nir_shader *shader);
3818 nir_if *nir_if_create(nir_shader *shader);
3819 nir_loop *nir_loop_create(nir_shader *shader);
3820 
3821 nir_function_impl *nir_cf_node_get_function(nir_cf_node *node);
3822 
3823 /** requests that the given pieces of metadata be generated */
3824 void nir_metadata_require(nir_function_impl *impl, nir_metadata required, ...);
3825 /** dirties all but the preserved metadata */
3826 void nir_metadata_preserve(nir_function_impl *impl, nir_metadata preserved);
3827 /** Preserves all metadata for the given shader */
3828 void nir_shader_preserve_all_metadata(nir_shader *shader);
3829 
3830 /** creates an instruction with default swizzle/writemask/etc. with NULL registers */
3831 nir_alu_instr *nir_alu_instr_create(nir_shader *shader, nir_op op);
3832 
3833 nir_deref_instr *nir_deref_instr_create(nir_shader *shader,
3834                                         nir_deref_type deref_type);
3835 
3836 nir_jump_instr *nir_jump_instr_create(nir_shader *shader, nir_jump_type type);
3837 
3838 nir_load_const_instr *nir_load_const_instr_create(nir_shader *shader,
3839                                                   unsigned num_components,
3840                                                   unsigned bit_size);
3841 
3842 nir_intrinsic_instr *nir_intrinsic_instr_create(nir_shader *shader,
3843                                                 nir_intrinsic_op op);
3844 
3845 nir_call_instr *nir_call_instr_create(nir_shader *shader,
3846                                       nir_function *callee);
3847 
3848 /** Creates a NIR texture instruction */
3849 nir_tex_instr *nir_tex_instr_create(nir_shader *shader, unsigned num_srcs);
3850 
3851 nir_phi_instr *nir_phi_instr_create(nir_shader *shader);
3852 nir_phi_src *nir_phi_instr_add_src(nir_phi_instr *instr, nir_block *pred, nir_src src);
3853 
3854 nir_parallel_copy_instr *nir_parallel_copy_instr_create(nir_shader *shader);
3855 
3856 nir_ssa_undef_instr *nir_ssa_undef_instr_create(nir_shader *shader,
3857                                                 unsigned num_components,
3858                                                 unsigned bit_size);
3859 
3860 nir_const_value nir_alu_binop_identity(nir_op binop, unsigned bit_size);
3861 
3862 /**
3863  * NIR Cursors and Instruction Insertion API
3864  * @{
3865  *
3866  * A tiny struct representing a point to insert/extract instructions or
3867  * control flow nodes.  Helps reduce the combinatorial explosion of possible
3868  * points to insert/extract.
3869  *
3870  * \sa nir_control_flow.h
3871  */
3872 typedef enum {
3873    nir_cursor_before_block,
3874    nir_cursor_after_block,
3875    nir_cursor_before_instr,
3876    nir_cursor_after_instr,
3877 } nir_cursor_option;
3878 
3879 typedef struct {
3880    nir_cursor_option option;
3881    union {
3882       nir_block *block;
3883       nir_instr *instr;
3884    };
3885 } nir_cursor;
3886 
3887 static inline nir_block *
nir_cursor_current_block(nir_cursor cursor)3888 nir_cursor_current_block(nir_cursor cursor)
3889 {
3890    if (cursor.option == nir_cursor_before_instr ||
3891        cursor.option == nir_cursor_after_instr) {
3892       return cursor.instr->block;
3893    } else {
3894       return cursor.block;
3895    }
3896 }
3897 
3898 bool nir_cursors_equal(nir_cursor a, nir_cursor b);
3899 
3900 static inline nir_cursor
nir_before_block(nir_block * block)3901 nir_before_block(nir_block *block)
3902 {
3903    nir_cursor cursor;
3904    cursor.option = nir_cursor_before_block;
3905    cursor.block = block;
3906    return cursor;
3907 }
3908 
3909 static inline nir_cursor
nir_after_block(nir_block * block)3910 nir_after_block(nir_block *block)
3911 {
3912    nir_cursor cursor;
3913    cursor.option = nir_cursor_after_block;
3914    cursor.block = block;
3915    return cursor;
3916 }
3917 
3918 static inline nir_cursor
nir_before_instr(nir_instr * instr)3919 nir_before_instr(nir_instr *instr)
3920 {
3921    nir_cursor cursor;
3922    cursor.option = nir_cursor_before_instr;
3923    cursor.instr = instr;
3924    return cursor;
3925 }
3926 
3927 static inline nir_cursor
nir_after_instr(nir_instr * instr)3928 nir_after_instr(nir_instr *instr)
3929 {
3930    nir_cursor cursor;
3931    cursor.option = nir_cursor_after_instr;
3932    cursor.instr = instr;
3933    return cursor;
3934 }
3935 
3936 static inline nir_cursor
nir_before_block_after_phis(nir_block * block)3937 nir_before_block_after_phis(nir_block *block)
3938 {
3939    nir_phi_instr *last_phi = nir_block_last_phi_instr(block);
3940    if (last_phi)
3941       return nir_after_instr(&last_phi->instr);
3942    else
3943       return nir_before_block(block);
3944 }
3945 
3946 static inline nir_cursor
nir_after_block_before_jump(nir_block * block)3947 nir_after_block_before_jump(nir_block *block)
3948 {
3949    nir_instr *last_instr = nir_block_last_instr(block);
3950    if (last_instr && last_instr->type == nir_instr_type_jump) {
3951       return nir_before_instr(last_instr);
3952    } else {
3953       return nir_after_block(block);
3954    }
3955 }
3956 
3957 static inline nir_cursor
nir_before_src(nir_src * src,bool is_if_condition)3958 nir_before_src(nir_src *src, bool is_if_condition)
3959 {
3960    if (is_if_condition) {
3961       nir_block *prev_block =
3962          nir_cf_node_as_block(nir_cf_node_prev(&src->parent_if->cf_node));
3963       assert(!nir_block_ends_in_jump(prev_block));
3964       return nir_after_block(prev_block);
3965    } else if (src->parent_instr->type == nir_instr_type_phi) {
3966 #ifndef NDEBUG
3967       nir_phi_instr *cond_phi = nir_instr_as_phi(src->parent_instr);
3968       bool found = false;
3969       nir_foreach_phi_src(phi_src, cond_phi) {
3970          if (phi_src->src.ssa == src->ssa) {
3971             found = true;
3972             break;
3973          }
3974       }
3975       assert(found);
3976 #endif
3977       /* The list_entry() macro is a generic container-of macro, it just happens
3978        * to have a more specific name.
3979        */
3980       nir_phi_src *phi_src = list_entry(src, nir_phi_src, src);
3981       return nir_after_block_before_jump(phi_src->pred);
3982    } else {
3983       return nir_before_instr(src->parent_instr);
3984    }
3985 }
3986 
3987 static inline nir_cursor
nir_before_cf_node(nir_cf_node * node)3988 nir_before_cf_node(nir_cf_node *node)
3989 {
3990    if (node->type == nir_cf_node_block)
3991       return nir_before_block(nir_cf_node_as_block(node));
3992 
3993    return nir_after_block(nir_cf_node_as_block(nir_cf_node_prev(node)));
3994 }
3995 
3996 static inline nir_cursor
nir_after_cf_node(nir_cf_node * node)3997 nir_after_cf_node(nir_cf_node *node)
3998 {
3999    if (node->type == nir_cf_node_block)
4000       return nir_after_block(nir_cf_node_as_block(node));
4001 
4002    return nir_before_block(nir_cf_node_as_block(nir_cf_node_next(node)));
4003 }
4004 
4005 static inline nir_cursor
nir_after_phis(nir_block * block)4006 nir_after_phis(nir_block *block)
4007 {
4008    nir_foreach_instr(instr, block) {
4009       if (instr->type != nir_instr_type_phi)
4010          return nir_before_instr(instr);
4011    }
4012    return nir_after_block(block);
4013 }
4014 
4015 static inline nir_cursor
nir_after_instr_and_phis(nir_instr * instr)4016 nir_after_instr_and_phis(nir_instr *instr)
4017 {
4018    if (instr->type == nir_instr_type_phi)
4019       return nir_after_phis(instr->block);
4020    else
4021       return nir_after_instr(instr);
4022 }
4023 
4024 static inline nir_cursor
nir_after_cf_node_and_phis(nir_cf_node * node)4025 nir_after_cf_node_and_phis(nir_cf_node *node)
4026 {
4027    if (node->type == nir_cf_node_block)
4028       return nir_after_block(nir_cf_node_as_block(node));
4029 
4030    nir_block *block = nir_cf_node_as_block(nir_cf_node_next(node));
4031 
4032    return nir_after_phis(block);
4033 }
4034 
4035 static inline nir_cursor
nir_before_cf_list(struct exec_list * cf_list)4036 nir_before_cf_list(struct exec_list *cf_list)
4037 {
4038    nir_cf_node *first_node = exec_node_data(nir_cf_node,
4039                                             exec_list_get_head(cf_list), node);
4040    return nir_before_cf_node(first_node);
4041 }
4042 
4043 static inline nir_cursor
nir_after_cf_list(struct exec_list * cf_list)4044 nir_after_cf_list(struct exec_list *cf_list)
4045 {
4046    nir_cf_node *last_node = exec_node_data(nir_cf_node,
4047                                            exec_list_get_tail(cf_list), node);
4048    return nir_after_cf_node(last_node);
4049 }
4050 
4051 /**
4052  * Insert a NIR instruction at the given cursor.
4053  *
4054  * Note: This does not update the cursor.
4055  */
4056 void nir_instr_insert(nir_cursor cursor, nir_instr *instr);
4057 
4058 bool nir_instr_move(nir_cursor cursor, nir_instr *instr);
4059 
4060 static inline void
nir_instr_insert_before(nir_instr * instr,nir_instr * before)4061 nir_instr_insert_before(nir_instr *instr, nir_instr *before)
4062 {
4063    nir_instr_insert(nir_before_instr(instr), before);
4064 }
4065 
4066 static inline void
nir_instr_insert_after(nir_instr * instr,nir_instr * after)4067 nir_instr_insert_after(nir_instr *instr, nir_instr *after)
4068 {
4069    nir_instr_insert(nir_after_instr(instr), after);
4070 }
4071 
4072 static inline void
nir_instr_insert_before_block(nir_block * block,nir_instr * before)4073 nir_instr_insert_before_block(nir_block *block, nir_instr *before)
4074 {
4075    nir_instr_insert(nir_before_block(block), before);
4076 }
4077 
4078 static inline void
nir_instr_insert_after_block(nir_block * block,nir_instr * after)4079 nir_instr_insert_after_block(nir_block *block, nir_instr *after)
4080 {
4081    nir_instr_insert(nir_after_block(block), after);
4082 }
4083 
4084 static inline void
nir_instr_insert_before_cf(nir_cf_node * node,nir_instr * before)4085 nir_instr_insert_before_cf(nir_cf_node *node, nir_instr *before)
4086 {
4087    nir_instr_insert(nir_before_cf_node(node), before);
4088 }
4089 
4090 static inline void
nir_instr_insert_after_cf(nir_cf_node * node,nir_instr * after)4091 nir_instr_insert_after_cf(nir_cf_node *node, nir_instr *after)
4092 {
4093    nir_instr_insert(nir_after_cf_node(node), after);
4094 }
4095 
4096 static inline void
nir_instr_insert_before_cf_list(struct exec_list * list,nir_instr * before)4097 nir_instr_insert_before_cf_list(struct exec_list *list, nir_instr *before)
4098 {
4099    nir_instr_insert(nir_before_cf_list(list), before);
4100 }
4101 
4102 static inline void
nir_instr_insert_after_cf_list(struct exec_list * list,nir_instr * after)4103 nir_instr_insert_after_cf_list(struct exec_list *list, nir_instr *after)
4104 {
4105    nir_instr_insert(nir_after_cf_list(list), after);
4106 }
4107 
4108 void nir_instr_remove_v(nir_instr *instr);
4109 void nir_instr_free(nir_instr *instr);
4110 void nir_instr_free_list(struct exec_list *list);
4111 
4112 static inline nir_cursor
nir_instr_remove(nir_instr * instr)4113 nir_instr_remove(nir_instr *instr)
4114 {
4115    nir_cursor cursor;
4116    nir_instr *prev = nir_instr_prev(instr);
4117    if (prev) {
4118       cursor = nir_after_instr(prev);
4119    } else {
4120       cursor = nir_before_block(instr->block);
4121    }
4122    nir_instr_remove_v(instr);
4123    return cursor;
4124 }
4125 
4126 nir_cursor nir_instr_free_and_dce(nir_instr *instr);
4127 
4128 /** @} */
4129 
4130 nir_ssa_def *nir_instr_ssa_def(nir_instr *instr);
4131 bool nir_instr_def_is_register(nir_instr *instr);
4132 
4133 typedef bool (*nir_foreach_ssa_def_cb)(nir_ssa_def *def, void *state);
4134 typedef bool (*nir_foreach_dest_cb)(nir_dest *dest, void *state);
4135 typedef bool (*nir_foreach_src_cb)(nir_src *src, void *state);
4136 bool nir_foreach_ssa_def(nir_instr *instr, nir_foreach_ssa_def_cb cb,
4137                          void *state);
4138 static inline bool nir_foreach_dest(nir_instr *instr, nir_foreach_dest_cb cb, void *state);
4139 static inline bool nir_foreach_src(nir_instr *instr, nir_foreach_src_cb cb, void *state);
4140 bool nir_foreach_phi_src_leaving_block(nir_block *instr,
4141                                        nir_foreach_src_cb cb,
4142                                        void *state);
4143 
4144 nir_const_value *nir_src_as_const_value(nir_src src);
4145 
4146 #define NIR_SRC_AS_(name, c_type, type_enum, cast_macro)                \
4147 static inline c_type *                                                  \
4148 nir_src_as_ ## name (nir_src src)                                       \
4149 {                                                                       \
4150     return src.is_ssa && src.ssa->parent_instr->type == type_enum       \
4151            ? cast_macro(src.ssa->parent_instr) : NULL;                  \
4152 }
4153 
4154 NIR_SRC_AS_(alu_instr, nir_alu_instr, nir_instr_type_alu, nir_instr_as_alu)
4155 NIR_SRC_AS_(intrinsic, nir_intrinsic_instr,
4156             nir_instr_type_intrinsic, nir_instr_as_intrinsic)
4157 NIR_SRC_AS_(deref, nir_deref_instr, nir_instr_type_deref, nir_instr_as_deref)
4158 
4159 bool nir_src_is_always_uniform(nir_src src);
4160 bool nir_srcs_equal(nir_src src1, nir_src src2);
4161 bool nir_instrs_equal(const nir_instr *instr1, const nir_instr *instr2);
4162 
4163 static inline void
nir_instr_rewrite_src_ssa(ASSERTED nir_instr * instr,nir_src * src,nir_ssa_def * new_ssa)4164 nir_instr_rewrite_src_ssa(ASSERTED nir_instr *instr,
4165                           nir_src *src, nir_ssa_def *new_ssa)
4166 {
4167    assert(src->parent_instr == instr);
4168    assert(src->is_ssa && src->ssa);
4169    list_del(&src->use_link);
4170    src->ssa = new_ssa;
4171    list_addtail(&src->use_link, &new_ssa->uses);
4172 }
4173 
4174 void nir_instr_rewrite_src(nir_instr *instr, nir_src *src, nir_src new_src);
4175 void nir_instr_move_src(nir_instr *dest_instr, nir_src *dest, nir_src *src);
4176 
4177 static inline void
nir_if_rewrite_condition_ssa(ASSERTED nir_if * if_stmt,nir_src * src,nir_ssa_def * new_ssa)4178 nir_if_rewrite_condition_ssa(ASSERTED nir_if *if_stmt,
4179                              nir_src *src, nir_ssa_def *new_ssa)
4180 {
4181    assert(src->parent_if == if_stmt);
4182    assert(src->is_ssa && src->ssa);
4183    list_del(&src->use_link);
4184    src->ssa = new_ssa;
4185    list_addtail(&src->use_link, &new_ssa->if_uses);
4186 }
4187 
4188 void nir_if_rewrite_condition(nir_if *if_stmt, nir_src new_src);
4189 void nir_instr_rewrite_dest(nir_instr *instr, nir_dest *dest,
4190                             nir_dest new_dest);
4191 
4192 void nir_ssa_dest_init(nir_instr *instr, nir_dest *dest,
4193                        unsigned num_components, unsigned bit_size,
4194                        const char *name);
4195 void nir_ssa_def_init(nir_instr *instr, nir_ssa_def *def,
4196                       unsigned num_components, unsigned bit_size);
4197 static inline void
nir_ssa_dest_init_for_type(nir_instr * instr,nir_dest * dest,const struct glsl_type * type,const char * name)4198 nir_ssa_dest_init_for_type(nir_instr *instr, nir_dest *dest,
4199                            const struct glsl_type *type,
4200                            const char *name)
4201 {
4202    assert(glsl_type_is_vector_or_scalar(type));
4203    nir_ssa_dest_init(instr, dest, glsl_get_components(type),
4204                      glsl_get_bit_size(type), name);
4205 }
4206 void nir_ssa_def_rewrite_uses(nir_ssa_def *def, nir_ssa_def *new_ssa);
4207 void nir_ssa_def_rewrite_uses_src(nir_ssa_def *def, nir_src new_src);
4208 void nir_ssa_def_rewrite_uses_after(nir_ssa_def *def, nir_ssa_def *new_ssa,
4209                                     nir_instr *after_me);
4210 
4211 nir_component_mask_t nir_src_components_read(const nir_src *src);
4212 nir_component_mask_t nir_ssa_def_components_read(const nir_ssa_def *def);
4213 
4214 static inline bool
nir_ssa_def_is_unused(nir_ssa_def * ssa)4215 nir_ssa_def_is_unused(nir_ssa_def *ssa)
4216 {
4217    return list_is_empty(&ssa->uses) && list_is_empty(&ssa->if_uses);
4218 }
4219 
4220 
4221 /** Returns the next block, disregarding structure
4222  *
4223  * The ordering is deterministic but has no guarantees beyond that.  In
4224  * particular, it is not guaranteed to be dominance-preserving.
4225  */
4226 nir_block *nir_block_unstructured_next(nir_block *block);
4227 nir_block *nir_unstructured_start_block(nir_function_impl *impl);
4228 
4229 #define nir_foreach_block_unstructured(block, impl) \
4230    for (nir_block *block = nir_unstructured_start_block(impl); block != NULL; \
4231         block = nir_block_unstructured_next(block))
4232 
4233 #define nir_foreach_block_unstructured_safe(block, impl) \
4234    for (nir_block *block = nir_unstructured_start_block(impl), \
4235         *next = nir_block_unstructured_next(block); \
4236         block != NULL; \
4237         block = next, next = nir_block_unstructured_next(block))
4238 
4239 /*
4240  * finds the next basic block in source-code order, returns NULL if there is
4241  * none
4242  */
4243 
4244 nir_block *nir_block_cf_tree_next(nir_block *block);
4245 
4246 /* Performs the opposite of nir_block_cf_tree_next() */
4247 
4248 nir_block *nir_block_cf_tree_prev(nir_block *block);
4249 
4250 /* Gets the first block in a CF node in source-code order */
4251 
4252 nir_block *nir_cf_node_cf_tree_first(nir_cf_node *node);
4253 
4254 /* Gets the last block in a CF node in source-code order */
4255 
4256 nir_block *nir_cf_node_cf_tree_last(nir_cf_node *node);
4257 
4258 /* Gets the next block after a CF node in source-code order */
4259 
4260 nir_block *nir_cf_node_cf_tree_next(nir_cf_node *node);
4261 
4262 /* Macros for loops that visit blocks in source-code order */
4263 
4264 #define nir_foreach_block(block, impl) \
4265    for (nir_block *block = nir_start_block(impl); block != NULL; \
4266         block = nir_block_cf_tree_next(block))
4267 
4268 #define nir_foreach_block_safe(block, impl) \
4269    for (nir_block *block = nir_start_block(impl), \
4270         *next = nir_block_cf_tree_next(block); \
4271         block != NULL; \
4272         block = next, next = nir_block_cf_tree_next(block))
4273 
4274 #define nir_foreach_block_reverse(block, impl) \
4275    for (nir_block *block = nir_impl_last_block(impl); block != NULL; \
4276         block = nir_block_cf_tree_prev(block))
4277 
4278 #define nir_foreach_block_reverse_safe(block, impl) \
4279    for (nir_block *block = nir_impl_last_block(impl), \
4280         *prev = nir_block_cf_tree_prev(block); \
4281         block != NULL; \
4282         block = prev, prev = nir_block_cf_tree_prev(block))
4283 
4284 #define nir_foreach_block_in_cf_node(block, node) \
4285    for (nir_block *block = nir_cf_node_cf_tree_first(node); \
4286         block != nir_cf_node_cf_tree_next(node); \
4287         block = nir_block_cf_tree_next(block))
4288 
4289 /* If the following CF node is an if, this function returns that if.
4290  * Otherwise, it returns NULL.
4291  */
4292 nir_if *nir_block_get_following_if(nir_block *block);
4293 
4294 nir_loop *nir_block_get_following_loop(nir_block *block);
4295 
4296 nir_block **nir_block_get_predecessors_sorted(const nir_block *block, void *mem_ctx);
4297 
4298 void nir_index_local_regs(nir_function_impl *impl);
4299 void nir_index_ssa_defs(nir_function_impl *impl);
4300 unsigned nir_index_instrs(nir_function_impl *impl);
4301 
4302 void nir_index_blocks(nir_function_impl *impl);
4303 
4304 unsigned nir_shader_index_vars(nir_shader *shader, nir_variable_mode modes);
4305 unsigned nir_function_impl_index_vars(nir_function_impl *impl);
4306 
4307 void nir_print_shader(nir_shader *shader, FILE *fp);
4308 void nir_print_shader_annotated(nir_shader *shader, FILE *fp, struct hash_table *errors);
4309 void nir_print_instr(const nir_instr *instr, FILE *fp);
4310 void nir_print_deref(const nir_deref_instr *deref, FILE *fp);
4311 void nir_log_shader_annotated_tagged(enum mesa_log_level level, const char *tag, nir_shader *shader, struct hash_table *annotations);
4312 #define nir_log_shadere(s) nir_log_shader_annotated_tagged(MESA_LOG_ERROR, (MESA_LOG_TAG), (s), NULL)
4313 #define nir_log_shaderw(s) nir_log_shader_annotated_tagged(MESA_LOG_WARN, (MESA_LOG_TAG), (s), NULL)
4314 #define nir_log_shaderi(s) nir_log_shader_annotated_tagged(MESA_LOG_INFO, (MESA_LOG_TAG), (s), NULL)
4315 #define nir_log_shader_annotated(s, annotations) nir_log_shader_annotated_tagged(MESA_LOG_ERROR, (MESA_LOG_TAG), (s), annotations)
4316 
4317 char *nir_shader_as_str(nir_shader *nir, void *mem_ctx);
4318 char *nir_shader_as_str_annotated(nir_shader *nir, struct hash_table *annotations, void *mem_ctx);
4319 
4320 /** Shallow clone of a single instruction. */
4321 nir_instr *nir_instr_clone(nir_shader *s, const nir_instr *orig);
4322 
4323 /** Clone a single instruction, including a remap table to rewrite sources. */
4324 nir_instr *nir_instr_clone_deep(nir_shader *s, const nir_instr *orig,
4325                                 struct hash_table *remap_table);
4326 
4327 /** Shallow clone of a single ALU instruction. */
4328 nir_alu_instr *nir_alu_instr_clone(nir_shader *s, const nir_alu_instr *orig);
4329 
4330 nir_shader *nir_shader_clone(void *mem_ctx, const nir_shader *s);
4331 nir_function_impl *nir_function_impl_clone(nir_shader *shader,
4332                                            const nir_function_impl *fi);
4333 nir_constant *nir_constant_clone(const nir_constant *c, nir_variable *var);
4334 nir_variable *nir_variable_clone(const nir_variable *c, nir_shader *shader);
4335 
4336 void nir_shader_replace(nir_shader *dest, nir_shader *src);
4337 
4338 void nir_shader_serialize_deserialize(nir_shader *s);
4339 
4340 #ifndef NDEBUG
4341 void nir_validate_shader(nir_shader *shader, const char *when);
4342 void nir_validate_ssa_dominance(nir_shader *shader, const char *when);
4343 void nir_metadata_set_validation_flag(nir_shader *shader);
4344 void nir_metadata_check_validation_flag(nir_shader *shader);
4345 
4346 static inline bool
should_skip_nir(const char * name)4347 should_skip_nir(const char *name)
4348 {
4349    static const char *list = NULL;
4350    if (!list) {
4351       /* Comma separated list of names to skip. */
4352       list = getenv("NIR_SKIP");
4353       if (!list)
4354          list = "";
4355    }
4356 
4357    if (!list[0])
4358       return false;
4359 
4360    return comma_separated_list_contains(list, name);
4361 }
4362 
4363 static inline bool
should_print_nir(nir_shader * shader)4364 should_print_nir(nir_shader *shader)
4365 {
4366    if (shader->info.internal ||
4367        shader->info.stage < 0 ||
4368        shader->info.stage > MESA_SHADER_KERNEL)
4369       return false;
4370 
4371    return unlikely(nir_debug_print_shader[shader->info.stage]);
4372 }
4373 #else
nir_validate_shader(nir_shader * shader,const char * when)4374 static inline void nir_validate_shader(nir_shader *shader, const char *when) { (void) shader; (void)when; }
nir_validate_ssa_dominance(nir_shader * shader,const char * when)4375 static inline void nir_validate_ssa_dominance(nir_shader *shader, const char *when) { (void) shader; (void)when; }
nir_metadata_set_validation_flag(nir_shader * shader)4376 static inline void nir_metadata_set_validation_flag(nir_shader *shader) { (void) shader; }
nir_metadata_check_validation_flag(nir_shader * shader)4377 static inline void nir_metadata_check_validation_flag(nir_shader *shader) { (void) shader; }
should_skip_nir(UNUSED const char * pass_name)4378 static inline bool should_skip_nir(UNUSED const char *pass_name) { return false; }
should_print_nir(UNUSED nir_shader * shader)4379 static inline bool should_print_nir(UNUSED nir_shader *shader) { return false; }
4380 #endif /* NDEBUG */
4381 
4382 #define _PASS(pass, nir, do_pass) do {                               \
4383    if (should_skip_nir(#pass)) {                                     \
4384       printf("skipping %s\n", #pass);                                \
4385       break;                                                         \
4386    }                                                                 \
4387    do_pass                                                           \
4388    if (NIR_DEBUG(CLONE)) {                                           \
4389       nir_shader *clone = nir_shader_clone(ralloc_parent(nir), nir); \
4390       nir_shader_replace(nir, clone);                                \
4391    }                                                                 \
4392    if (NIR_DEBUG(SERIALIZE)) {                                       \
4393       nir_shader_serialize_deserialize(nir);                         \
4394    }                                                                 \
4395 } while (0)
4396 
4397 #define NIR_PASS(progress, nir, pass, ...) _PASS(pass, nir,          \
4398    nir_metadata_set_validation_flag(nir);                            \
4399    if (should_print_nir(nir))                                        \
4400       printf("%s\n", #pass);                                         \
4401    if (pass(nir, ##__VA_ARGS__)) {                                   \
4402       nir_validate_shader(nir, "after " #pass " in " __FILE__);      \
4403       UNUSED bool _;                                                 \
4404       progress = true;                                               \
4405       if (should_print_nir(nir))                                     \
4406          nir_print_shader(nir, stdout);                              \
4407       nir_metadata_check_validation_flag(nir);                       \
4408    }                                                                 \
4409 )
4410 
4411 #define NIR_PASS_V(nir, pass, ...) _PASS(pass, nir,                  \
4412    if (should_print_nir(nir))                                        \
4413       printf("%s\n", #pass);                                         \
4414    pass(nir, ##__VA_ARGS__);                                         \
4415    nir_validate_shader(nir, "after " #pass " in " __FILE__);         \
4416    if (should_print_nir(nir))                                        \
4417       nir_print_shader(nir, stdout);                                 \
4418 )
4419 
4420 #define NIR_SKIP(name) should_skip_nir(#name)
4421 
4422 /** An instruction filtering callback with writemask
4423  *
4424  * Returns true if the instruction should be processed with the associated
4425  * writemask and false otherwise.
4426  */
4427 typedef bool (*nir_instr_writemask_filter_cb)(const nir_instr *,
4428                                               unsigned writemask, const void *);
4429 
4430 /** A simple instruction lowering callback
4431  *
4432  * Many instruction lowering passes can be written as a simple function which
4433  * takes an instruction as its input and returns a sequence of instructions
4434  * that implement the consumed instruction.  This function type represents
4435  * such a lowering function.  When called, a function with this prototype
4436  * should either return NULL indicating that no lowering needs to be done or
4437  * emit a sequence of instructions using the provided builder (whose cursor
4438  * will already be placed after the instruction to be lowered) and return the
4439  * resulting nir_ssa_def.
4440  */
4441 typedef nir_ssa_def *(*nir_lower_instr_cb)(struct nir_builder *,
4442                                            nir_instr *, void *);
4443 
4444 /**
4445  * Special return value for nir_lower_instr_cb when some progress occurred
4446  * (like changing an input to the instr) that didn't result in a replacement
4447  * SSA def being generated.
4448  */
4449 #define NIR_LOWER_INSTR_PROGRESS ((nir_ssa_def *)(uintptr_t)1)
4450 
4451 /**
4452  * Special return value for nir_lower_instr_cb when some progress occurred
4453  * that should remove the current instruction that doesn't create an output
4454  * (like a store)
4455  */
4456 
4457 #define NIR_LOWER_INSTR_PROGRESS_REPLACE ((nir_ssa_def *)(uintptr_t)2)
4458 
4459 /** Iterate over all the instructions in a nir_function_impl and lower them
4460  *  using the provided callbacks
4461  *
4462  * This function implements the guts of a standard lowering pass for you.  It
4463  * iterates over all of the instructions in a nir_function_impl and calls the
4464  * filter callback on each one.  If the filter callback returns true, it then
4465  * calls the lowering call back on the instruction.  (Splitting it this way
4466  * allows us to avoid some save/restore work for instructions we know won't be
4467  * lowered.)  If the instruction is dead after the lowering is complete, it
4468  * will be removed.  If new instructions are added, the lowering callback will
4469  * also be called on them in case multiple lowerings are required.
4470  *
4471  * If the callback indicates that the original instruction is replaced (either
4472  * through a new SSA def or NIR_LOWER_INSTR_PROGRESS_REPLACE), then the
4473  * instruction is removed along with any now-dead SSA defs it used.
4474  *
4475  * The metadata for the nir_function_impl will also be updated.  If any blocks
4476  * are added (they cannot be removed), dominance and block indices will be
4477  * invalidated.
4478  */
4479 bool nir_function_impl_lower_instructions(nir_function_impl *impl,
4480                                           nir_instr_filter_cb filter,
4481                                           nir_lower_instr_cb lower,
4482                                           void *cb_data);
4483 bool nir_shader_lower_instructions(nir_shader *shader,
4484                                    nir_instr_filter_cb filter,
4485                                    nir_lower_instr_cb lower,
4486                                    void *cb_data);
4487 
4488 void nir_calc_dominance_impl(nir_function_impl *impl);
4489 void nir_calc_dominance(nir_shader *shader);
4490 
4491 nir_block *nir_dominance_lca(nir_block *b1, nir_block *b2);
4492 bool nir_block_dominates(nir_block *parent, nir_block *child);
4493 bool nir_block_is_unreachable(nir_block *block);
4494 
4495 void nir_dump_dom_tree_impl(nir_function_impl *impl, FILE *fp);
4496 void nir_dump_dom_tree(nir_shader *shader, FILE *fp);
4497 
4498 void nir_dump_dom_frontier_impl(nir_function_impl *impl, FILE *fp);
4499 void nir_dump_dom_frontier(nir_shader *shader, FILE *fp);
4500 
4501 void nir_dump_cfg_impl(nir_function_impl *impl, FILE *fp);
4502 void nir_dump_cfg(nir_shader *shader, FILE *fp);
4503 
4504 void nir_gs_count_vertices_and_primitives(const nir_shader *shader,
4505                                           int *out_vtxcnt,
4506                                           int *out_prmcnt,
4507                                           unsigned num_streams);
4508 
4509 typedef enum {
4510    nir_group_all,
4511    nir_group_same_resource_only,
4512 } nir_load_grouping;
4513 
4514 void nir_group_loads(nir_shader *shader, nir_load_grouping grouping,
4515                      unsigned max_distance);
4516 
4517 bool nir_shrink_vec_array_vars(nir_shader *shader, nir_variable_mode modes);
4518 bool nir_split_array_vars(nir_shader *shader, nir_variable_mode modes);
4519 bool nir_split_var_copies(nir_shader *shader);
4520 bool nir_split_per_member_structs(nir_shader *shader);
4521 bool nir_split_struct_vars(nir_shader *shader, nir_variable_mode modes);
4522 
4523 bool nir_lower_returns_impl(nir_function_impl *impl);
4524 bool nir_lower_returns(nir_shader *shader);
4525 
4526 void nir_inline_function_impl(struct nir_builder *b,
4527                               const nir_function_impl *impl,
4528                               nir_ssa_def **params,
4529                               struct hash_table *shader_var_remap);
4530 bool nir_inline_functions(nir_shader *shader);
4531 
4532 void nir_find_inlinable_uniforms(nir_shader *shader);
4533 void nir_inline_uniforms(nir_shader *shader, unsigned num_uniforms,
4534                          const uint32_t *uniform_values,
4535                          const uint16_t *uniform_dw_offsets);
4536 
4537 bool nir_propagate_invariant(nir_shader *shader, bool invariant_prim);
4538 
4539 void nir_lower_var_copy_instr(nir_intrinsic_instr *copy, nir_shader *shader);
4540 void nir_lower_deref_copy_instr(struct nir_builder *b,
4541                                 nir_intrinsic_instr *copy);
4542 bool nir_lower_var_copies(nir_shader *shader);
4543 
4544 bool nir_opt_memcpy(nir_shader *shader);
4545 bool nir_lower_memcpy(nir_shader *shader);
4546 
4547 void nir_fixup_deref_modes(nir_shader *shader);
4548 
4549 bool nir_lower_global_vars_to_local(nir_shader *shader);
4550 
4551 typedef enum {
4552    nir_lower_direct_array_deref_of_vec_load     = (1 << 0),
4553    nir_lower_indirect_array_deref_of_vec_load   = (1 << 1),
4554    nir_lower_direct_array_deref_of_vec_store    = (1 << 2),
4555    nir_lower_indirect_array_deref_of_vec_store  = (1 << 3),
4556 } nir_lower_array_deref_of_vec_options;
4557 
4558 bool nir_lower_array_deref_of_vec(nir_shader *shader, nir_variable_mode modes,
4559                                   nir_lower_array_deref_of_vec_options options);
4560 
4561 bool nir_lower_indirect_derefs(nir_shader *shader, nir_variable_mode modes,
4562                                uint32_t max_lower_array_len);
4563 
4564 bool nir_lower_indirect_var_derefs(nir_shader *shader,
4565                                    const struct set *vars);
4566 
4567 bool nir_lower_locals_to_regs(nir_shader *shader);
4568 
4569 void nir_lower_io_to_temporaries(nir_shader *shader,
4570                                  nir_function_impl *entrypoint,
4571                                  bool outputs, bool inputs);
4572 
4573 bool nir_lower_vars_to_scratch(nir_shader *shader,
4574                                nir_variable_mode modes,
4575                                int size_threshold,
4576                                glsl_type_size_align_func size_align);
4577 
4578 void nir_lower_clip_halfz(nir_shader *shader);
4579 
4580 void nir_shader_gather_info(nir_shader *shader, nir_function_impl *entrypoint);
4581 
4582 void nir_gather_ssa_types(nir_function_impl *impl,
4583                           BITSET_WORD *float_types,
4584                           BITSET_WORD *int_types);
4585 
4586 void nir_assign_var_locations(nir_shader *shader, nir_variable_mode mode,
4587                               unsigned *size,
4588                               int (*type_size)(const struct glsl_type *, bool));
4589 
4590 /* Some helpers to do very simple linking */
4591 bool nir_remove_unused_varyings(nir_shader *producer, nir_shader *consumer);
4592 bool nir_remove_unused_io_vars(nir_shader *shader, nir_variable_mode mode,
4593                                uint64_t *used_by_other_stage,
4594                                uint64_t *used_by_other_stage_patches);
4595 void nir_compact_varyings(nir_shader *producer, nir_shader *consumer,
4596                           bool default_to_smooth_interp);
4597 void nir_link_xfb_varyings(nir_shader *producer, nir_shader *consumer);
4598 bool nir_link_opt_varyings(nir_shader *producer, nir_shader *consumer);
4599 void nir_link_varying_precision(nir_shader *producer, nir_shader *consumer);
4600 
4601 bool nir_slot_is_sysval_output(gl_varying_slot slot);
4602 bool nir_slot_is_varying(gl_varying_slot slot);
4603 bool nir_slot_is_sysval_output_and_varying(gl_varying_slot slot);
4604 void nir_remove_varying(nir_intrinsic_instr *intr);
4605 void nir_remove_sysval_output(nir_intrinsic_instr *intr);
4606 
4607 bool nir_lower_amul(nir_shader *shader,
4608                     int (*type_size)(const struct glsl_type *, bool));
4609 
4610 bool nir_lower_ubo_vec4(nir_shader *shader);
4611 
4612 void nir_assign_io_var_locations(nir_shader *shader,
4613                                  nir_variable_mode mode,
4614                                  unsigned *size,
4615                                  gl_shader_stage stage);
4616 
4617 typedef struct {
4618    uint8_t num_linked_io_vars;
4619    uint8_t num_linked_patch_io_vars;
4620 } nir_linked_io_var_info;
4621 
4622 nir_linked_io_var_info
4623 nir_assign_linked_io_var_locations(nir_shader *producer,
4624                                    nir_shader *consumer);
4625 
4626 typedef enum {
4627    /* If set, this causes all 64-bit IO operations to be lowered on-the-fly
4628     * to 32-bit operations.  This is only valid for nir_var_shader_in/out
4629     * modes.
4630     */
4631    nir_lower_io_lower_64bit_to_32 = (1 << 0),
4632 
4633    /* If set, this forces all non-flat fragment shader inputs to be
4634     * interpolated as if with the "sample" qualifier.  This requires
4635     * nir_shader_compiler_options::use_interpolated_input_intrinsics.
4636     */
4637    nir_lower_io_force_sample_interpolation = (1 << 1),
4638 } nir_lower_io_options;
4639 bool nir_lower_io(nir_shader *shader,
4640                   nir_variable_mode modes,
4641                   int (*type_size)(const struct glsl_type *, bool),
4642                   nir_lower_io_options);
4643 
4644 bool nir_io_add_const_offset_to_base(nir_shader *nir, nir_variable_mode modes);
4645 
4646 void
4647 nir_lower_io_passes(nir_shader *nir);
4648 
4649 bool nir_io_add_intrinsic_xfb_info(nir_shader *nir);
4650 
4651 bool
4652 nir_lower_vars_to_explicit_types(nir_shader *shader,
4653                                  nir_variable_mode modes,
4654                                  glsl_type_size_align_func type_info);
4655 void
4656 nir_gather_explicit_io_initializers(nir_shader *shader,
4657                                     void *dst, size_t dst_size,
4658                                     nir_variable_mode mode);
4659 
4660 bool nir_lower_vec3_to_vec4(nir_shader *shader, nir_variable_mode modes);
4661 
4662 typedef enum {
4663    /**
4664     * An address format which is a simple 32-bit global GPU address.
4665     */
4666    nir_address_format_32bit_global,
4667 
4668    /**
4669     * An address format which is a simple 64-bit global GPU address.
4670     */
4671    nir_address_format_64bit_global,
4672 
4673    /**
4674     * An address format which is a 64-bit global GPU address encoded as a
4675     * 2x32-bit vector.
4676     */
4677    nir_address_format_2x32bit_global,
4678 
4679    /**
4680     * An address format which is a 64-bit global base address and a 32-bit
4681     * offset.
4682     *
4683     * The address is comprised as a 32-bit vec4 where .xy are a uint64_t base
4684     * address stored with the low bits in .x and high bits in .y, .z is
4685     * undefined, and .w is an offset.  This is intended to match
4686     * 64bit_bounded_global but without the bounds checking.
4687     */
4688    nir_address_format_64bit_global_32bit_offset,
4689 
4690    /**
4691     * An address format which is a bounds-checked 64-bit global GPU address.
4692     *
4693     * The address is comprised as a 32-bit vec4 where .xy are a uint64_t base
4694     * address stored with the low bits in .x and high bits in .y, .z is a
4695     * size, and .w is an offset.  When the final I/O operation is lowered, .w
4696     * is checked against .z and the operation is predicated on the result.
4697     */
4698    nir_address_format_64bit_bounded_global,
4699 
4700    /**
4701     * An address format which is comprised of a vec2 where the first
4702     * component is a buffer index and the second is an offset.
4703     */
4704    nir_address_format_32bit_index_offset,
4705 
4706    /**
4707     * An address format which is a 64-bit value, where the high 32 bits
4708     * are a buffer index, and the low 32 bits are an offset.
4709     */
4710     nir_address_format_32bit_index_offset_pack64,
4711 
4712    /**
4713     * An address format which is comprised of a vec3 where the first two
4714     * components specify the buffer and the third is an offset.
4715     */
4716    nir_address_format_vec2_index_32bit_offset,
4717 
4718    /**
4719     * An address format which represents generic pointers with a 62-bit
4720     * pointer and a 2-bit enum in the top two bits.  The top two bits have
4721     * the following meanings:
4722     *
4723     *  - 0x0: Global memory
4724     *  - 0x1: Shared memory
4725     *  - 0x2: Scratch memory
4726     *  - 0x3: Global memory
4727     *
4728     * The redundancy between 0x0 and 0x3 is because of Intel sign-extension of
4729     * addresses.  Valid global memory addresses may naturally have either 0 or
4730     * ~0 as their high bits.
4731     *
4732     * Shared and scratch pointers are represented as 32-bit offsets with the
4733     * top 32 bits only being used for the enum.  This allows us to avoid
4734     * 64-bit address calculations in a bunch of cases.
4735     */
4736    nir_address_format_62bit_generic,
4737 
4738    /**
4739     * An address format which is a simple 32-bit offset.
4740     */
4741    nir_address_format_32bit_offset,
4742 
4743    /**
4744     * An address format which is a simple 32-bit offset cast to 64-bit.
4745     */
4746     nir_address_format_32bit_offset_as_64bit,
4747 
4748    /**
4749     * An address format representing a purely logical addressing model.  In
4750     * this model, all deref chains must be complete from the dereference
4751     * operation to the variable.  Cast derefs are not allowed.  These
4752     * addresses will be 32-bit scalars but the format is immaterial because
4753     * you can always chase the chain.
4754     */
4755    nir_address_format_logical,
4756 } nir_address_format;
4757 
4758 unsigned
4759 nir_address_format_bit_size(nir_address_format addr_format);
4760 
4761 unsigned
4762 nir_address_format_num_components(nir_address_format addr_format);
4763 
4764 static inline const struct glsl_type *
nir_address_format_to_glsl_type(nir_address_format addr_format)4765 nir_address_format_to_glsl_type(nir_address_format addr_format)
4766 {
4767    unsigned bit_size = nir_address_format_bit_size(addr_format);
4768    assert(bit_size == 32 || bit_size == 64);
4769    return glsl_vector_type(bit_size == 32 ? GLSL_TYPE_UINT : GLSL_TYPE_UINT64,
4770                            nir_address_format_num_components(addr_format));
4771 }
4772 
4773 const nir_const_value *nir_address_format_null_value(nir_address_format addr_format);
4774 
4775 nir_ssa_def *nir_build_addr_ieq(struct nir_builder *b, nir_ssa_def *addr0, nir_ssa_def *addr1,
4776                                 nir_address_format addr_format);
4777 
4778 nir_ssa_def *nir_build_addr_isub(struct nir_builder *b, nir_ssa_def *addr0, nir_ssa_def *addr1,
4779                                  nir_address_format addr_format);
4780 
4781 nir_ssa_def * nir_explicit_io_address_from_deref(struct nir_builder *b,
4782                                                  nir_deref_instr *deref,
4783                                                  nir_ssa_def *base_addr,
4784                                                  nir_address_format addr_format);
4785 
4786 bool nir_get_explicit_deref_align(nir_deref_instr *deref,
4787                                   bool default_to_type_align,
4788                                   uint32_t *align_mul,
4789                                   uint32_t *align_offset);
4790 
4791 void nir_lower_explicit_io_instr(struct nir_builder *b,
4792                                  nir_intrinsic_instr *io_instr,
4793                                  nir_ssa_def *addr,
4794                                  nir_address_format addr_format);
4795 
4796 bool nir_lower_explicit_io(nir_shader *shader,
4797                            nir_variable_mode modes,
4798                            nir_address_format);
4799 
4800 bool
4801 nir_lower_shader_calls(nir_shader *shader,
4802                        nir_address_format address_format,
4803                        unsigned stack_alignment,
4804                        nir_shader ***resume_shaders_out,
4805                        uint32_t *num_resume_shaders_out,
4806                        void *mem_ctx);
4807 
4808 nir_src *nir_get_io_offset_src(nir_intrinsic_instr *instr);
4809 nir_src *nir_get_io_arrayed_index_src(nir_intrinsic_instr *instr);
4810 nir_src *nir_get_shader_call_payload_src(nir_intrinsic_instr *call);
4811 
4812 bool nir_is_arrayed_io(const nir_variable *var, gl_shader_stage stage);
4813 
4814 bool nir_lower_regs_to_ssa_impl(nir_function_impl *impl);
4815 bool nir_lower_regs_to_ssa(nir_shader *shader);
4816 bool nir_lower_vars_to_ssa(nir_shader *shader);
4817 
4818 bool nir_remove_dead_derefs(nir_shader *shader);
4819 bool nir_remove_dead_derefs_impl(nir_function_impl *impl);
4820 
4821 typedef struct nir_remove_dead_variables_options {
4822    bool (*can_remove_var)(nir_variable *var, void *data);
4823    void *can_remove_var_data;
4824 } nir_remove_dead_variables_options;
4825 
4826 bool nir_remove_dead_variables(nir_shader *shader, nir_variable_mode modes,
4827                                const nir_remove_dead_variables_options *options);
4828 
4829 bool nir_lower_variable_initializers(nir_shader *shader,
4830                                      nir_variable_mode modes);
4831 bool nir_zero_initialize_shared_memory(nir_shader *shader,
4832                                        const unsigned shared_size,
4833                                        const unsigned chunk_size);
4834 
4835 bool nir_move_vec_src_uses_to_dest(nir_shader *shader);
4836 bool nir_lower_vec_to_movs(nir_shader *shader, nir_instr_writemask_filter_cb cb,
4837                            const void *_data);
4838 void nir_lower_alpha_test(nir_shader *shader, enum compare_func func,
4839                           bool alpha_to_one,
4840                           const gl_state_index16 *alpha_ref_state_tokens);
4841 bool nir_lower_alu(nir_shader *shader);
4842 
4843 bool nir_lower_flrp(nir_shader *shader, unsigned lowering_mask,
4844                     bool always_precise);
4845 
4846 bool nir_scale_fdiv(nir_shader *shader);
4847 
4848 bool nir_lower_alu_to_scalar(nir_shader *shader, nir_instr_filter_cb cb, const void *data);
4849 bool nir_lower_alu_width(nir_shader *shader, nir_vectorize_cb cb, const void *data);
4850 bool nir_lower_bool_to_bitsize(nir_shader *shader);
4851 bool nir_lower_bool_to_float(nir_shader *shader);
4852 bool nir_lower_bool_to_int32(nir_shader *shader);
4853 bool nir_opt_simplify_convert_alu_types(nir_shader *shader);
4854 bool nir_lower_const_arrays_to_uniforms(nir_shader *shader,
4855                                         unsigned max_uniform_components);
4856 bool nir_lower_convert_alu_types(nir_shader *shader,
4857                                  bool (*should_lower)(nir_intrinsic_instr *));
4858 bool nir_lower_constant_convert_alu_types(nir_shader *shader);
4859 bool nir_lower_alu_conversion_to_intrinsic(nir_shader *shader);
4860 bool nir_lower_int_to_float(nir_shader *shader);
4861 bool nir_lower_load_const_to_scalar(nir_shader *shader);
4862 bool nir_lower_read_invocation_to_scalar(nir_shader *shader);
4863 bool nir_lower_phis_to_scalar(nir_shader *shader, bool lower_all);
4864 void nir_lower_io_arrays_to_elements(nir_shader *producer, nir_shader *consumer);
4865 void nir_lower_io_arrays_to_elements_no_indirects(nir_shader *shader,
4866                                                   bool outputs_only);
4867 void nir_lower_io_to_scalar(nir_shader *shader, nir_variable_mode mask);
4868 bool nir_lower_io_to_scalar_early(nir_shader *shader, nir_variable_mode mask);
4869 bool nir_lower_io_to_vector(nir_shader *shader, nir_variable_mode mask);
4870 bool nir_vectorize_tess_levels(nir_shader *shader);
4871 
4872 bool nir_lower_fragcolor(nir_shader *shader, unsigned max_cbufs);
4873 bool nir_lower_fragcoord_wtrans(nir_shader *shader);
4874 void nir_lower_viewport_transform(nir_shader *shader);
4875 bool nir_lower_uniforms_to_ubo(nir_shader *shader, bool dword_packed, bool load_vec4);
4876 
4877 bool nir_lower_is_helper_invocation(nir_shader *shader);
4878 
4879 bool nir_lower_single_sampled(nir_shader *shader);
4880 
4881 typedef struct nir_lower_subgroups_options {
4882    uint8_t subgroup_size;
4883    uint8_t ballot_bit_size;
4884    uint8_t ballot_components;
4885    bool lower_to_scalar:1;
4886    bool lower_vote_trivial:1;
4887    bool lower_vote_eq:1;
4888    bool lower_subgroup_masks:1;
4889    bool lower_relative_shuffle:1;
4890    bool lower_shuffle_to_32bit:1;
4891    bool lower_shuffle_to_swizzle_amd:1;
4892    bool lower_shuffle:1;
4893    bool lower_quad:1;
4894    bool lower_quad_broadcast_dynamic:1;
4895    bool lower_quad_broadcast_dynamic_to_const:1;
4896    bool lower_elect:1;
4897    bool lower_read_invocation_to_cond:1;
4898 } nir_lower_subgroups_options;
4899 
4900 bool nir_lower_subgroups(nir_shader *shader,
4901                          const nir_lower_subgroups_options *options);
4902 
4903 bool nir_lower_system_values(nir_shader *shader);
4904 
4905 typedef struct nir_lower_compute_system_values_options {
4906    bool has_base_global_invocation_id:1;
4907    bool has_base_workgroup_id:1;
4908    bool shuffle_local_ids_for_quad_derivatives:1;
4909    bool lower_local_invocation_index:1;
4910    bool lower_cs_local_id_to_index:1;
4911    bool lower_workgroup_id_to_index:1;
4912 } nir_lower_compute_system_values_options;
4913 
4914 bool nir_lower_compute_system_values(nir_shader *shader,
4915                                      const nir_lower_compute_system_values_options *options);
4916 
4917 struct nir_lower_sysvals_to_varyings_options {
4918    bool frag_coord:1;
4919    bool front_face:1;
4920    bool point_coord:1;
4921 };
4922 
4923 bool
4924 nir_lower_sysvals_to_varyings(nir_shader *shader,
4925                               const struct nir_lower_sysvals_to_varyings_options *options);
4926 
4927 enum PACKED nir_lower_tex_packing {
4928    /** No packing */
4929    nir_lower_tex_packing_none = 0,
4930    /**
4931     * The sampler returns up to 2 32-bit words of half floats or 16-bit signed
4932     * or unsigned ints based on the sampler type
4933     */
4934    nir_lower_tex_packing_16,
4935    /** The sampler returns 1 32-bit word of 4x8 unorm */
4936    nir_lower_tex_packing_8,
4937 };
4938 
4939 typedef struct nir_lower_tex_options {
4940    /**
4941     * bitmask of (1 << GLSL_SAMPLER_DIM_x) to control for which
4942     * sampler types a texture projector is lowered.
4943     */
4944    unsigned lower_txp;
4945 
4946    /**
4947     * If true, lower texture projector for any array sampler dims
4948     */
4949    bool lower_txp_array;
4950 
4951    /**
4952     * If true, lower away nir_tex_src_offset for all texelfetch instructions.
4953     */
4954    bool lower_txf_offset;
4955 
4956    /**
4957     * If true, lower away nir_tex_src_offset for all rect textures.
4958     */
4959    bool lower_rect_offset;
4960 
4961    /**
4962     * If not NULL, this filter will return true for tex instructions that
4963     * should lower away nir_tex_src_offset.
4964     */
4965    nir_instr_filter_cb lower_offset_filter;
4966 
4967    /**
4968     * If true, lower rect textures to 2D, using txs to fetch the
4969     * texture dimensions and dividing the texture coords by the
4970     * texture dims to normalize.
4971     */
4972    bool lower_rect;
4973 
4974    /**
4975     * If true, convert yuv to rgb.
4976     */
4977    unsigned lower_y_uv_external;
4978    unsigned lower_y_u_v_external;
4979    unsigned lower_yx_xuxv_external;
4980    unsigned lower_xy_uxvx_external;
4981    unsigned lower_ayuv_external;
4982    unsigned lower_xyuv_external;
4983    unsigned lower_yuv_external;
4984    unsigned lower_yu_yv_external;
4985    unsigned lower_y41x_external;
4986    unsigned bt709_external;
4987    unsigned bt2020_external;
4988    unsigned yuv_full_range_external;
4989 
4990    /**
4991     * To emulate certain texture wrap modes, this can be used
4992     * to saturate the specified tex coord to [0.0, 1.0].  The
4993     * bits are according to sampler #, ie. if, for example:
4994     *
4995     *   (conf->saturate_s & (1 << n))
4996     *
4997     * is true, then the s coord for sampler n is saturated.
4998     *
4999     * Note that clamping must happen *after* projector lowering
5000     * so any projected texture sample instruction with a clamped
5001     * coordinate gets automatically lowered, regardless of the
5002     * 'lower_txp' setting.
5003     */
5004    unsigned saturate_s;
5005    unsigned saturate_t;
5006    unsigned saturate_r;
5007 
5008    /* Bitmask of textures that need swizzling.
5009     *
5010     * If (swizzle_result & (1 << texture_index)), then the swizzle in
5011     * swizzles[texture_index] is applied to the result of the texturing
5012     * operation.
5013     */
5014    unsigned swizzle_result;
5015 
5016    /* A swizzle for each texture.  Values 0-3 represent x, y, z, or w swizzles
5017     * while 4 and 5 represent 0 and 1 respectively.
5018     *
5019     * Indexed by texture-id.
5020     */
5021    uint8_t swizzles[32][4];
5022 
5023    /* Can be used to scale sampled values in range required by the
5024     * format.
5025     *
5026     * Indexed by texture-id.
5027     */
5028    float scale_factors[32];
5029 
5030    /**
5031     * Bitmap of textures that need srgb to linear conversion.  If
5032     * (lower_srgb & (1 << texture_index)) then the rgb (xyz) components
5033     * of the texture are lowered to linear.
5034     */
5035    unsigned lower_srgb;
5036 
5037    /**
5038     * If true, lower nir_texop_txd on cube maps with nir_texop_txl.
5039     */
5040    bool lower_txd_cube_map;
5041 
5042    /**
5043     * If true, lower nir_texop_txd on 3D surfaces with nir_texop_txl.
5044     */
5045    bool lower_txd_3d;
5046 
5047    /**
5048     * If true, lower nir_texop_txd any array surfaces with nir_texop_txl.
5049     */
5050    bool lower_txd_array;
5051 
5052    /**
5053     * If true, lower nir_texop_txd on shadow samplers (except cube maps)
5054     * with nir_texop_txl. Notice that cube map shadow samplers are lowered
5055     * with lower_txd_cube_map.
5056     */
5057    bool lower_txd_shadow;
5058 
5059    /**
5060     * If true, lower nir_texop_txd on all samplers to a nir_texop_txl.
5061     * Implies lower_txd_cube_map and lower_txd_shadow.
5062     */
5063    bool lower_txd;
5064 
5065    /**
5066     * If true, lower nir_texop_txb that try to use shadow compare and min_lod
5067     * at the same time to a nir_texop_lod, some math, and nir_texop_tex.
5068     */
5069    bool lower_txb_shadow_clamp;
5070 
5071    /**
5072     * If true, lower nir_texop_txd on shadow samplers when it uses min_lod
5073     * with nir_texop_txl.  This includes cube maps.
5074     */
5075    bool lower_txd_shadow_clamp;
5076 
5077    /**
5078     * If true, lower nir_texop_txd on when it uses both offset and min_lod
5079     * with nir_texop_txl.  This includes cube maps.
5080     */
5081    bool lower_txd_offset_clamp;
5082 
5083    /**
5084     * If true, lower nir_texop_txd with min_lod to a nir_texop_txl if the
5085     * sampler is bindless.
5086     */
5087    bool lower_txd_clamp_bindless_sampler;
5088 
5089    /**
5090     * If true, lower nir_texop_txd with min_lod to a nir_texop_txl if the
5091     * sampler index is not statically determinable to be less than 16.
5092     */
5093    bool lower_txd_clamp_if_sampler_index_not_lt_16;
5094 
5095    /**
5096     * If true, lower nir_texop_txs with a non-0-lod into nir_texop_txs with
5097     * 0-lod followed by a nir_ishr.
5098     */
5099    bool lower_txs_lod;
5100 
5101    /**
5102     * If true, lower nir_texop_txs for cube arrays to a nir_texop_txs with a
5103     * 2D array type followed by a nir_idiv by 6.
5104     */
5105    bool lower_txs_cube_array;
5106 
5107    /**
5108     * If true, apply a .bagr swizzle on tg4 results to handle Broadcom's
5109     * mixed-up tg4 locations.
5110     */
5111    bool lower_tg4_broadcom_swizzle;
5112 
5113    /**
5114     * If true, lowers tg4 with 4 constant offsets to 4 tg4 calls
5115     */
5116    bool lower_tg4_offsets;
5117 
5118    /**
5119     * Lower txf_ms to fragment_mask_fetch and fragment_fetch and samples_identical to
5120     * fragment_mask_fetch.
5121     */
5122    bool lower_to_fragment_fetch_amd;
5123 
5124    /**
5125     * To lower packed sampler return formats.
5126     *
5127     * Indexed by sampler-id.
5128     */
5129    enum nir_lower_tex_packing lower_tex_packing[32];
5130 
5131    /**
5132     * If true, lower nir_texop_lod to return -FLT_MAX if the sum of the
5133     * absolute values of derivatives is 0 for all coordinates.
5134     */
5135    bool lower_lod_zero_width;
5136 
5137    /* Turns nir_op_tex and other ops with an implicit derivative, in stages
5138     * without implicit derivatives (like the vertex shader) to have an explicit
5139     * LOD with a value of 0.
5140     */
5141    bool lower_invalid_implicit_lod;
5142 
5143    /* If true, round the layer component of the coordinates source to the nearest
5144     * integer for all array ops.
5145     */
5146    bool lower_array_layer_round_even;
5147 
5148    /**
5149     * Payload data to be sent to callback / filter functions.
5150     */
5151    void *callback_data;
5152 } nir_lower_tex_options;
5153 
5154 /** Lowers complex texture instructions to simpler ones */
5155 bool nir_lower_tex(nir_shader *shader,
5156                    const nir_lower_tex_options *options);
5157 
5158 
5159 typedef struct nir_lower_tex_shadow_swizzle {
5160    unsigned swizzle_r:3;
5161    unsigned swizzle_g:3;
5162    unsigned swizzle_b:3;
5163    unsigned swizzle_a:3;
5164 } nir_lower_tex_shadow_swizzle;
5165 
5166 bool
5167 nir_lower_tex_shadow(nir_shader *s,
5168                      unsigned n_states,
5169                      enum compare_func *compare_func,
5170                      nir_lower_tex_shadow_swizzle *tex_swizzles);
5171 
5172 typedef struct nir_lower_image_options {
5173    /**
5174     * If true, lower cube size operations.
5175     */
5176    bool lower_cube_size;
5177 } nir_lower_image_options;
5178 
5179 bool nir_lower_image(nir_shader *nir,
5180                      const nir_lower_image_options *options);
5181 
5182 bool nir_lower_readonly_images_to_tex(nir_shader *shader, bool per_variable);
5183 
5184 enum nir_lower_non_uniform_access_type {
5185    nir_lower_non_uniform_ubo_access     = (1 << 0),
5186    nir_lower_non_uniform_ssbo_access    = (1 << 1),
5187    nir_lower_non_uniform_texture_access = (1 << 2),
5188    nir_lower_non_uniform_image_access   = (1 << 3),
5189 };
5190 
5191 /* Given the nir_src used for the resource, return the channels which might be non-uniform. */
5192 typedef nir_component_mask_t (*nir_lower_non_uniform_access_callback)(const nir_src *, void *);
5193 
5194 typedef struct nir_lower_non_uniform_access_options {
5195    enum nir_lower_non_uniform_access_type types;
5196    nir_lower_non_uniform_access_callback callback;
5197    void *callback_data;
5198 } nir_lower_non_uniform_access_options;
5199 
5200 bool nir_lower_non_uniform_access(nir_shader *shader,
5201                                   const nir_lower_non_uniform_access_options *options);
5202 
5203 typedef struct {
5204    /* If true, a 32-bit division lowering based on NV50LegalizeSSA::handleDIV()
5205     * is used. It is the faster of the two but it is not exact in some cases
5206     * (for example, 1091317713u / 1034u gives 5209173 instead of 1055432).
5207     *
5208     * If false, a lowering based on AMDGPUTargetLowering::LowerUDIVREM() and
5209     * AMDGPUTargetLowering::LowerSDIVREM() is used. It requires more
5210     * instructions than the nv50 path and many of them are integer
5211     * multiplications, so it is probably slower. It should always return the
5212     * correct result, though.
5213     */
5214    bool imprecise_32bit_lowering;
5215 
5216    /* Whether 16-bit floating point arithmetic should be allowed in 8-bit
5217     * division lowering
5218     */
5219    bool allow_fp16;
5220 } nir_lower_idiv_options;
5221 
5222 bool nir_lower_idiv(nir_shader *shader, const nir_lower_idiv_options *options);
5223 
5224 typedef struct nir_input_attachment_options {
5225    bool use_fragcoord_sysval;
5226    bool use_layer_id_sysval;
5227    bool use_view_id_for_layer;
5228 } nir_input_attachment_options;
5229 
5230 bool nir_lower_input_attachments(nir_shader *shader,
5231                                  const nir_input_attachment_options *options);
5232 
5233 bool nir_lower_clip_vs(nir_shader *shader, unsigned ucp_enables,
5234                        bool use_vars,
5235                        bool use_clipdist_array,
5236                        const gl_state_index16 clipplane_state_tokens[][STATE_LENGTH]);
5237 bool nir_lower_clip_gs(nir_shader *shader, unsigned ucp_enables,
5238                        bool use_clipdist_array,
5239                        const gl_state_index16 clipplane_state_tokens[][STATE_LENGTH]);
5240 bool nir_lower_clip_fs(nir_shader *shader, unsigned ucp_enables,
5241                        bool use_clipdist_array);
5242 bool nir_lower_clip_cull_distance_arrays(nir_shader *nir);
5243 bool nir_lower_clip_disable(nir_shader *shader, unsigned clip_plane_enable);
5244 
5245 void nir_lower_point_size_mov(nir_shader *shader,
5246                               const gl_state_index16 *pointsize_state_tokens);
5247 
5248 bool nir_lower_frexp(nir_shader *nir);
5249 
5250 void nir_lower_two_sided_color(nir_shader *shader, bool face_sysval);
5251 
5252 bool nir_lower_clamp_color_outputs(nir_shader *shader);
5253 
5254 bool nir_lower_flatshade(nir_shader *shader);
5255 
5256 void nir_lower_passthrough_edgeflags(nir_shader *shader);
5257 bool nir_lower_patch_vertices(nir_shader *nir, unsigned static_count,
5258                               const gl_state_index16 *uniform_state_tokens);
5259 
5260 typedef struct nir_lower_wpos_ytransform_options {
5261    gl_state_index16 state_tokens[STATE_LENGTH];
5262    bool fs_coord_origin_upper_left :1;
5263    bool fs_coord_origin_lower_left :1;
5264    bool fs_coord_pixel_center_integer :1;
5265    bool fs_coord_pixel_center_half_integer :1;
5266 } nir_lower_wpos_ytransform_options;
5267 
5268 bool nir_lower_wpos_ytransform(nir_shader *shader,
5269                                const nir_lower_wpos_ytransform_options *options);
5270 bool nir_lower_wpos_center(nir_shader *shader);
5271 
5272 bool nir_lower_pntc_ytransform(nir_shader *shader,
5273                                const gl_state_index16 clipplane_state_tokens[][STATE_LENGTH]);
5274 
5275 bool nir_lower_wrmasks(nir_shader *shader, nir_instr_filter_cb cb, const void *data);
5276 
5277 bool nir_lower_fb_read(nir_shader *shader);
5278 
5279 typedef struct nir_lower_drawpixels_options {
5280    gl_state_index16 texcoord_state_tokens[STATE_LENGTH];
5281    gl_state_index16 scale_state_tokens[STATE_LENGTH];
5282    gl_state_index16 bias_state_tokens[STATE_LENGTH];
5283    unsigned drawpix_sampler;
5284    unsigned pixelmap_sampler;
5285    bool pixel_maps :1;
5286    bool scale_and_bias :1;
5287 } nir_lower_drawpixels_options;
5288 
5289 void nir_lower_drawpixels(nir_shader *shader,
5290                           const nir_lower_drawpixels_options *options);
5291 
5292 typedef struct nir_lower_bitmap_options {
5293    unsigned sampler;
5294    bool swizzle_xxxx;
5295 } nir_lower_bitmap_options;
5296 
5297 void nir_lower_bitmap(nir_shader *shader, const nir_lower_bitmap_options *options);
5298 
5299 bool nir_lower_atomics_to_ssbo(nir_shader *shader, unsigned offset_align_state);
5300 
5301 typedef enum  {
5302    nir_lower_int_source_mods = 1 << 0,
5303    nir_lower_fabs_source_mods = 1 << 1,
5304    nir_lower_fneg_source_mods = 1 << 2,
5305    nir_lower_64bit_source_mods = 1 << 3,
5306    nir_lower_triop_abs = 1 << 4,
5307    nir_lower_all_source_mods = (1 << 5) - 1
5308 } nir_lower_to_source_mods_flags;
5309 
5310 #define nir_lower_float_source_mods (nir_lower_fabs_source_mods | nir_lower_fneg_source_mods)
5311 
5312 bool nir_lower_to_source_mods(nir_shader *shader, nir_lower_to_source_mods_flags options);
5313 
5314 typedef enum {
5315    nir_lower_gs_intrinsics_per_stream = 1 << 0,
5316    nir_lower_gs_intrinsics_count_primitives = 1 << 1,
5317    nir_lower_gs_intrinsics_count_vertices_per_primitive = 1 << 2,
5318    nir_lower_gs_intrinsics_overwrite_incomplete = 1 << 3,
5319 } nir_lower_gs_intrinsics_flags;
5320 
5321 bool nir_lower_gs_intrinsics(nir_shader *shader, nir_lower_gs_intrinsics_flags options);
5322 
5323 typedef struct {
5324    bool payload_to_shared_for_atomics : 1;
5325 } nir_lower_task_shader_options;
5326 
5327 bool nir_lower_task_shader(nir_shader *shader, nir_lower_task_shader_options options);
5328 
5329 typedef unsigned (*nir_lower_bit_size_callback)(const nir_instr *, void *);
5330 
5331 bool nir_lower_bit_size(nir_shader *shader,
5332                         nir_lower_bit_size_callback callback,
5333                         void *callback_data);
5334 bool nir_lower_64bit_phis(nir_shader *shader);
5335 
5336 bool nir_split_64bit_vec3_and_vec4(nir_shader *shader);
5337 
5338 nir_lower_int64_options nir_lower_int64_op_to_options_mask(nir_op opcode);
5339 bool nir_lower_int64(nir_shader *shader);
5340 
5341 nir_lower_doubles_options nir_lower_doubles_op_to_options_mask(nir_op opcode);
5342 bool nir_lower_doubles(nir_shader *shader, const nir_shader *softfp64,
5343                        nir_lower_doubles_options options);
5344 bool nir_lower_pack(nir_shader *shader);
5345 
5346 bool nir_recompute_io_bases(nir_shader *nir, nir_variable_mode modes);
5347 bool nir_lower_mediump_io(nir_shader *nir, nir_variable_mode modes,
5348                           uint64_t varying_mask, bool use_16bit_slots);
5349 bool nir_force_mediump_io(nir_shader *nir, nir_variable_mode modes,
5350                           nir_alu_type types);
5351 bool nir_unpack_16bit_varying_slots(nir_shader *nir, nir_variable_mode modes);
5352 
5353 struct nir_fold_tex_srcs_options {
5354    unsigned sampler_dims;
5355    unsigned src_types;
5356 };
5357 
5358 struct nir_fold_16bit_tex_image_options {
5359    nir_rounding_mode rounding_mode;
5360    bool fold_tex_dest;
5361    bool fold_image_load_store_data;
5362    unsigned fold_srcs_options_count;
5363    struct nir_fold_tex_srcs_options *fold_srcs_options;
5364 };
5365 
5366 bool nir_fold_16bit_tex_image(nir_shader *nir,
5367                               struct nir_fold_16bit_tex_image_options *options);
5368 
5369 typedef struct {
5370    bool legalize_type;         /* whether this src should be legalized */
5371    uint8_t bit_size;           /* bit_size to enforce */
5372    nir_tex_src_type match_src; /* if bit_size is 0, match bit size of this */
5373 } nir_tex_src_type_constraint, nir_tex_src_type_constraints[nir_num_tex_src_types];
5374 
5375 bool nir_legalize_16bit_sampler_srcs(nir_shader *nir,
5376                                      nir_tex_src_type_constraints constraints);
5377 
5378 bool nir_lower_point_size(nir_shader *shader, float min, float max);
5379 
5380 void nir_lower_texcoord_replace(nir_shader *s, unsigned coord_replace,
5381                                 bool point_coord_is_sysval, bool yinvert);
5382 
5383 typedef enum {
5384    nir_lower_interpolation_at_sample = (1 << 1),
5385    nir_lower_interpolation_at_offset = (1 << 2),
5386    nir_lower_interpolation_centroid  = (1 << 3),
5387    nir_lower_interpolation_pixel     = (1 << 4),
5388    nir_lower_interpolation_sample    = (1 << 5),
5389 } nir_lower_interpolation_options;
5390 
5391 bool nir_lower_interpolation(nir_shader *shader,
5392                              nir_lower_interpolation_options options);
5393 
5394 bool nir_lower_discard_if(nir_shader *shader);
5395 
5396 bool nir_lower_discard_or_demote(nir_shader *shader,
5397                                  bool force_correct_quad_ops_after_discard);
5398 
5399 bool nir_lower_memory_model(nir_shader *shader);
5400 
5401 bool nir_lower_goto_ifs(nir_shader *shader);
5402 
5403 bool nir_shader_uses_view_index(nir_shader *shader);
5404 bool nir_can_lower_multiview(nir_shader *shader);
5405 bool nir_lower_multiview(nir_shader *shader, uint32_t view_mask);
5406 
5407 
5408 bool nir_lower_fp16_casts(nir_shader *shader);
5409 bool nir_normalize_cubemap_coords(nir_shader *shader);
5410 
5411 bool nir_shader_supports_implicit_lod(nir_shader *shader);
5412 
5413 void nir_live_ssa_defs_impl(nir_function_impl *impl);
5414 
5415 const BITSET_WORD *nir_get_live_ssa_defs(nir_cursor cursor, void *mem_ctx);
5416 
5417 void nir_loop_analyze_impl(nir_function_impl *impl,
5418                            nir_variable_mode indirect_mask,
5419                            bool force_unroll_sampler_indirect);
5420 
5421 bool nir_ssa_defs_interfere(nir_ssa_def *a, nir_ssa_def *b);
5422 
5423 bool nir_repair_ssa_impl(nir_function_impl *impl);
5424 bool nir_repair_ssa(nir_shader *shader);
5425 
5426 void nir_convert_loop_to_lcssa(nir_loop *loop);
5427 bool nir_convert_to_lcssa(nir_shader *shader, bool skip_invariants, bool skip_bool_invariants);
5428 void nir_divergence_analysis(nir_shader *shader);
5429 bool nir_update_instr_divergence(nir_shader *shader, nir_instr *instr);
5430 bool nir_has_divergent_loop(nir_shader *shader);
5431 
5432 /* If phi_webs_only is true, only convert SSA values involved in phi nodes to
5433  * registers.  If false, convert all values (even those not involved in a phi
5434  * node) to registers.
5435  */
5436 bool nir_convert_from_ssa(nir_shader *shader, bool phi_webs_only);
5437 
5438 bool nir_lower_phis_to_regs_block(nir_block *block);
5439 bool nir_lower_ssa_defs_to_regs_block(nir_block *block);
5440 bool nir_rematerialize_derefs_in_use_blocks_impl(nir_function_impl *impl);
5441 
5442 bool nir_lower_samplers(nir_shader *shader);
5443 bool nir_lower_ssbo(nir_shader *shader);
5444 
5445 typedef struct nir_lower_printf_options {
5446    bool treat_doubles_as_floats : 1;
5447    unsigned max_buffer_size;
5448 } nir_lower_printf_options;
5449 
5450 bool nir_lower_printf(nir_shader *nir, const nir_lower_printf_options *options);
5451 
5452 /* This is here for unit tests. */
5453 bool nir_opt_comparison_pre_impl(nir_function_impl *impl);
5454 
5455 bool nir_opt_comparison_pre(nir_shader *shader);
5456 
5457 typedef struct nir_opt_access_options {
5458    bool is_vulkan;
5459    bool infer_non_readable;
5460 } nir_opt_access_options;
5461 
5462 bool nir_opt_access(nir_shader *shader, const nir_opt_access_options *options);
5463 bool nir_opt_algebraic(nir_shader *shader);
5464 bool nir_opt_algebraic_before_ffma(nir_shader *shader);
5465 bool nir_opt_algebraic_late(nir_shader *shader);
5466 bool nir_opt_algebraic_distribute_src_mods(nir_shader *shader);
5467 bool nir_opt_constant_folding(nir_shader *shader);
5468 
5469 /* Try to combine a and b into a.  Return true if combination was possible,
5470  * which will result in b being removed by the pass.  Return false if
5471  * combination wasn't possible.
5472  */
5473 typedef bool (*nir_combine_memory_barrier_cb)(
5474    nir_intrinsic_instr *a, nir_intrinsic_instr *b, void *data);
5475 
5476 bool nir_opt_combine_memory_barriers(nir_shader *shader,
5477                                      nir_combine_memory_barrier_cb combine_cb,
5478                                      void *data);
5479 
5480 bool nir_opt_combine_stores(nir_shader *shader, nir_variable_mode modes);
5481 
5482 bool nir_copy_prop_impl(nir_function_impl *impl);
5483 bool nir_copy_prop(nir_shader *shader);
5484 
5485 bool nir_opt_copy_prop_vars(nir_shader *shader);
5486 
5487 bool nir_opt_cse(nir_shader *shader);
5488 
5489 bool nir_opt_dce(nir_shader *shader);
5490 
5491 bool nir_opt_dead_cf(nir_shader *shader);
5492 
5493 bool nir_opt_dead_write_vars(nir_shader *shader);
5494 
5495 bool nir_opt_deref_impl(nir_function_impl *impl);
5496 bool nir_opt_deref(nir_shader *shader);
5497 
5498 bool nir_opt_find_array_copies(nir_shader *shader);
5499 
5500 bool nir_opt_fragdepth(nir_shader *shader);
5501 
5502 bool nir_opt_gcm(nir_shader *shader, bool value_number);
5503 
5504 bool nir_opt_idiv_const(nir_shader *shader, unsigned min_bit_size);
5505 
5506 typedef enum {
5507    nir_opt_if_aggressive_last_continue = (1 << 0),
5508    nir_opt_if_optimize_phi_true_false = (1 << 1),
5509 } nir_opt_if_options;
5510 
5511 bool nir_opt_if(nir_shader *shader, nir_opt_if_options options);
5512 
5513 bool nir_opt_intrinsics(nir_shader *shader);
5514 
5515 bool nir_opt_large_constants(nir_shader *shader,
5516                              glsl_type_size_align_func size_align,
5517                              unsigned threshold);
5518 
5519 bool nir_opt_loop_unroll(nir_shader *shader);
5520 
5521 typedef enum {
5522     nir_move_const_undef  = (1 << 0),
5523     nir_move_load_ubo     = (1 << 1),
5524     nir_move_load_input   = (1 << 2),
5525     nir_move_comparisons  = (1 << 3),
5526     nir_move_copies       = (1 << 4),
5527     nir_move_load_ssbo    = (1 << 5),
5528     nir_move_load_uniform = (1 << 6),
5529 } nir_move_options;
5530 
5531 bool nir_can_move_instr(nir_instr *instr, nir_move_options options);
5532 
5533 bool nir_opt_sink(nir_shader *shader, nir_move_options options);
5534 
5535 bool nir_opt_move(nir_shader *shader, nir_move_options options);
5536 
5537 typedef struct {
5538    /** nir_load_uniform max base offset */
5539    uint32_t uniform_max;
5540 
5541    /** nir_load_ubo_vec4 max base offset */
5542    uint32_t ubo_vec4_max;
5543 
5544    /** nir_var_mem_shared max base offset */
5545    uint32_t shared_max;
5546 
5547    /** nir_load/store_buffer_amd max base offset */
5548    uint32_t buffer_max;
5549 } nir_opt_offsets_options;
5550 
5551 bool nir_opt_offsets(nir_shader *shader, const nir_opt_offsets_options *options);
5552 
5553 bool nir_opt_peephole_select(nir_shader *shader, unsigned limit,
5554                              bool indirect_load_ok, bool expensive_alu_ok);
5555 
5556 bool nir_opt_rematerialize_compares(nir_shader *shader);
5557 
5558 bool nir_opt_remove_phis(nir_shader *shader);
5559 bool nir_opt_remove_phis_block(nir_block *block);
5560 
5561 bool nir_opt_phi_precision(nir_shader *shader);
5562 
5563 bool nir_opt_shrink_stores(nir_shader *shader, bool shrink_image_store);
5564 
5565 bool nir_opt_shrink_vectors(nir_shader *shader);
5566 
5567 bool nir_opt_trivial_continues(nir_shader *shader);
5568 
5569 bool nir_opt_undef(nir_shader *shader);
5570 
5571 bool nir_lower_undef_to_zero(nir_shader *shader);
5572 
5573 bool nir_opt_uniform_atomics(nir_shader *shader);
5574 
5575 bool nir_opt_vectorize(nir_shader *shader, nir_vectorize_cb filter,
5576                        void *data);
5577 
5578 bool nir_opt_conditional_discard(nir_shader *shader);
5579 bool nir_opt_move_discards_to_top(nir_shader *shader);
5580 
5581 bool nir_opt_ray_queries(nir_shader *shader);
5582 
5583 typedef bool (*nir_should_vectorize_mem_func)(unsigned align_mul,
5584                                               unsigned align_offset,
5585                                               unsigned bit_size,
5586                                               unsigned num_components,
5587                                               nir_intrinsic_instr *low, nir_intrinsic_instr *high,
5588                                               void *data);
5589 
5590 typedef struct {
5591    nir_should_vectorize_mem_func callback;
5592    nir_variable_mode modes;
5593    nir_variable_mode robust_modes;
5594    void *cb_data;
5595    bool has_shared2_amd;
5596 } nir_load_store_vectorize_options;
5597 
5598 bool nir_opt_load_store_vectorize(nir_shader *shader, const nir_load_store_vectorize_options *options);
5599 
5600 void nir_sweep(nir_shader *shader);
5601 
5602 void nir_remap_dual_slot_attributes(nir_shader *shader,
5603                                     uint64_t *dual_slot_inputs);
5604 uint64_t nir_get_single_slot_attribs_mask(uint64_t attribs, uint64_t dual_slot);
5605 
5606 nir_intrinsic_op nir_intrinsic_from_system_value(gl_system_value val);
5607 gl_system_value nir_system_value_from_intrinsic(nir_intrinsic_op intrin);
5608 
5609 static inline bool
nir_variable_is_in_ubo(const nir_variable * var)5610 nir_variable_is_in_ubo(const nir_variable *var)
5611 {
5612    return (var->data.mode == nir_var_mem_ubo &&
5613            var->interface_type != NULL);
5614 }
5615 
5616 static inline bool
nir_variable_is_in_ssbo(const nir_variable * var)5617 nir_variable_is_in_ssbo(const nir_variable *var)
5618 {
5619    return (var->data.mode == nir_var_mem_ssbo &&
5620            var->interface_type != NULL);
5621 }
5622 
5623 static inline bool
nir_variable_is_in_block(const nir_variable * var)5624 nir_variable_is_in_block(const nir_variable *var)
5625 {
5626    return nir_variable_is_in_ubo(var) || nir_variable_is_in_ssbo(var);
5627 }
5628 
5629 typedef struct nir_unsigned_upper_bound_config {
5630    unsigned min_subgroup_size;
5631    unsigned max_subgroup_size;
5632    unsigned max_workgroup_invocations;
5633    unsigned max_workgroup_count[3];
5634    unsigned max_workgroup_size[3];
5635 
5636    uint32_t vertex_attrib_max[32];
5637 } nir_unsigned_upper_bound_config;
5638 
5639 uint32_t
5640 nir_unsigned_upper_bound(nir_shader *shader, struct hash_table *range_ht,
5641                          nir_ssa_scalar scalar,
5642                          const nir_unsigned_upper_bound_config *config);
5643 
5644 bool
5645 nir_addition_might_overflow(nir_shader *shader, struct hash_table *range_ht,
5646                             nir_ssa_scalar ssa, unsigned const_val,
5647                             const nir_unsigned_upper_bound_config *config);
5648 
5649 typedef enum {
5650    nir_ray_query_value_intersection_type,
5651    nir_ray_query_value_intersection_t,
5652    nir_ray_query_value_intersection_instance_custom_index,
5653    nir_ray_query_value_intersection_instance_id,
5654    nir_ray_query_value_intersection_instance_sbt_index,
5655    nir_ray_query_value_intersection_geometry_index,
5656    nir_ray_query_value_intersection_primitive_index,
5657    nir_ray_query_value_intersection_barycentrics,
5658    nir_ray_query_value_intersection_front_face,
5659    nir_ray_query_value_intersection_object_ray_direction,
5660    nir_ray_query_value_intersection_object_ray_origin,
5661    nir_ray_query_value_intersection_object_to_world,
5662    nir_ray_query_value_intersection_world_to_object,
5663    nir_ray_query_value_intersection_candidate_aabb_opaque,
5664    nir_ray_query_value_tmin,
5665    nir_ray_query_value_flags,
5666    nir_ray_query_value_world_ray_direction,
5667    nir_ray_query_value_world_ray_origin,
5668 } nir_ray_query_value;
5669 
5670 typedef struct {
5671    /* True if gl_DrawID is considered uniform, i.e. if the preamble is run
5672     * at least once per "internal" draw rather than per user-visible draw.
5673     */
5674    bool drawid_uniform;
5675 
5676    /* True if the subgroup size is uniform. */
5677    bool subgroup_size_uniform;
5678 
5679    /* size/align for load/store_preamble. */
5680    void (*def_size)(nir_ssa_def *def, unsigned *size, unsigned *align);
5681 
5682    /* Total available size for load/store_preamble storage, in units
5683     * determined by def_size.
5684     */
5685    unsigned preamble_storage_size;
5686 
5687    /* Give the cost for an instruction. nir_opt_preamble will prioritize
5688     * instructions with higher costs. Instructions with cost 0 may still be
5689     * lifted, but only when required to lift other instructions with non-0
5690     * cost (e.g. a load_const source of an expression).
5691     */
5692    float (*instr_cost_cb)(nir_instr *instr, const void *data);
5693 
5694    /* Give the cost of rewriting the instruction to use load_preamble. This
5695     * may happen from inserting move instructions, etc. If the benefit doesn't
5696     * exceed the cost here then we won't rewrite it.
5697     */
5698    float (*rewrite_cost_cb)(nir_ssa_def *def, const void *data);
5699 
5700    /* Instructions whose definitions should not be rewritten. These could
5701     * still be moved to the preamble, but they shouldn't be the root of a
5702     * replacement expression. Instructions with cost 0 and derefs are
5703     * automatically included by the pass.
5704     */
5705    nir_instr_filter_cb avoid_instr_cb;
5706 
5707    const void *cb_data;
5708 } nir_opt_preamble_options;
5709 
5710 bool
5711 nir_opt_preamble(nir_shader *shader,
5712                  const nir_opt_preamble_options *options,
5713                  unsigned *size);
5714 
5715 nir_function_impl *nir_shader_get_preamble(nir_shader *shader);
5716 
5717 bool nir_lower_point_smooth(nir_shader *shader);
5718 bool nir_lower_poly_line_smooth(nir_shader *shader, unsigned num_smooth_aa_sample);
5719 
5720 #include "nir_inline_helpers.h"
5721 
5722 #ifdef __cplusplus
5723 } /* extern "C" */
5724 #endif
5725 
5726 #endif /* NIR_H */
5727