• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * Copyright © 2015 Rob Clark <robclark@freedesktop.org>
3  * SPDX-License-Identifier: MIT
4  *
5  * Authors:
6  *    Rob Clark <robclark@freedesktop.org>
7  */
8 
9 #include "util/ralloc.h"
10 
11 #include "freedreno_dev_info.h"
12 
13 #include "ir3_compiler.h"
14 
15 static const struct debug_named_value shader_debug_options[] = {
16    /* clang-format off */
17    {"vs",         IR3_DBG_SHADER_VS,  "Print shader disasm for vertex shaders"},
18    {"tcs",        IR3_DBG_SHADER_TCS, "Print shader disasm for tess ctrl shaders"},
19    {"tes",        IR3_DBG_SHADER_TES, "Print shader disasm for tess eval shaders"},
20    {"gs",         IR3_DBG_SHADER_GS,  "Print shader disasm for geometry shaders"},
21    {"fs",         IR3_DBG_SHADER_FS,  "Print shader disasm for fragment shaders"},
22    {"cs",         IR3_DBG_SHADER_CS,  "Print shader disasm for compute shaders"},
23    {"internal",   IR3_DBG_SHADER_INTERNAL, "Print shader disasm for internal shaders (normally not included in vs/fs/cs/etc)"},
24    {"disasm",     IR3_DBG_DISASM,     "Dump NIR and adreno shader disassembly"},
25    {"optmsgs",    IR3_DBG_OPTMSGS,    "Enable optimizer debug messages"},
26    {"forces2en",  IR3_DBG_FORCES2EN,  "Force s2en mode for tex sampler instructions"},
27    {"nouboopt",   IR3_DBG_NOUBOOPT,   "Disable lowering UBO to uniform"},
28    {"nofp16",     IR3_DBG_NOFP16,     "Don't lower mediump to fp16"},
29    {"nocache",    IR3_DBG_NOCACHE,    "Disable shader cache"},
30    {"spillall",   IR3_DBG_SPILLALL,   "Spill as much as possible to test the spiller"},
31    {"nopreamble", IR3_DBG_NOPREAMBLE, "Disable the preamble pass"},
32    {"fullsync",   IR3_DBG_FULLSYNC,   "Add (sy) + (ss) after each cat5/cat6"},
33    {"fullnop",    IR3_DBG_FULLNOP,    "Add nops before each instruction"},
34    {"noearlypreamble", IR3_DBG_NOEARLYPREAMBLE, "Disable early preambles"},
35    {"nodescprefetch", IR3_DBG_NODESCPREFETCH, "Disable descriptor prefetch optimization"},
36    {"expandrpt",  IR3_DBG_EXPANDRPT,  "Expand rptN instructions"},
37 #if MESA_DEBUG
38    /* MESA_DEBUG-only options: */
39    {"schedmsgs",  IR3_DBG_SCHEDMSGS,  "Enable scheduler debug messages"},
40    {"ramsgs",     IR3_DBG_RAMSGS,     "Enable register-allocation debug messages"},
41 #endif
42    DEBUG_NAMED_VALUE_END
43    /* clang-format on */
44 };
45 
46 DEBUG_GET_ONCE_FLAGS_OPTION(ir3_shader_debug, "IR3_SHADER_DEBUG",
47                             shader_debug_options, 0)
48 DEBUG_GET_ONCE_OPTION(ir3_shader_override_path, "IR3_SHADER_OVERRIDE_PATH",
49                       NULL)
50 
51 enum ir3_shader_debug ir3_shader_debug = 0;
52 const char *ir3_shader_override_path = NULL;
53 
54 void
ir3_compiler_destroy(struct ir3_compiler * compiler)55 ir3_compiler_destroy(struct ir3_compiler *compiler)
56 {
57    disk_cache_destroy(compiler->disk_cache);
58    ralloc_free(compiler);
59 }
60 
61 static const nir_shader_compiler_options ir3_base_options = {
62    .compact_arrays = true,
63    .lower_fpow = true,
64    .lower_scmp = true,
65    .lower_flrp16 = true,
66    .lower_flrp32 = true,
67    .lower_flrp64 = true,
68    .lower_ffract = true,
69    .lower_fmod = true,
70    .lower_fdiv = true,
71    .lower_isign = true,
72    .lower_ldexp = true,
73    .lower_uadd_carry = true,
74    .lower_usub_borrow = true,
75    .lower_mul_high = true,
76    .lower_mul_2x32_64 = true,
77    .fuse_ffma16 = true,
78    .fuse_ffma32 = true,
79    .fuse_ffma64 = true,
80    .vertex_id_zero_based = false,
81    .lower_extract_byte = true,
82    .lower_extract_word = true,
83    .lower_insert_byte = true,
84    .lower_insert_word = true,
85    .lower_helper_invocation = true,
86    .lower_bitfield_insert = true,
87    .lower_bitfield_extract = true,
88    .lower_pack_half_2x16 = true,
89    .lower_pack_snorm_4x8 = true,
90    .lower_pack_snorm_2x16 = true,
91    .lower_pack_unorm_4x8 = true,
92    .lower_pack_unorm_2x16 = true,
93    .lower_unpack_half_2x16 = true,
94    .lower_unpack_snorm_4x8 = true,
95    .lower_unpack_snorm_2x16 = true,
96    .lower_unpack_unorm_4x8 = true,
97    .lower_unpack_unorm_2x16 = true,
98    .lower_pack_split = true,
99    .lower_to_scalar = true,
100    .has_imul24 = true,
101    .has_icsel_eqz32 = true,
102    .has_icsel_eqz16 = true,
103    .has_fsub = true,
104    .has_isub = true,
105    .force_indirect_unrolling_sampler = true,
106    .lower_uniforms_to_ubo = true,
107    .max_unroll_iterations = 32,
108 
109    .lower_cs_local_index_to_id = true,
110    .lower_wpos_pntc = true,
111 
112    .lower_hadd = true,
113    .lower_hadd64 = true,
114    .lower_fisnormal = true,
115 
116    .lower_int64_options = (nir_lower_int64_options)~0,
117    .lower_doubles_options = (nir_lower_doubles_options)~0,
118 
119    .divergence_analysis_options = nir_divergence_uniform_load_tears,
120    .scalarize_ddx = true,
121 
122    .per_view_unique_driver_locations = true,
123    .compact_view_index = true,
124 };
125 
126 struct ir3_compiler *
ir3_compiler_create(struct fd_device * dev,const struct fd_dev_id * dev_id,const struct fd_dev_info * dev_info,const struct ir3_compiler_options * options)127 ir3_compiler_create(struct fd_device *dev, const struct fd_dev_id *dev_id,
128                     const struct fd_dev_info *dev_info,
129                     const struct ir3_compiler_options *options)
130 {
131    struct ir3_compiler *compiler = rzalloc(NULL, struct ir3_compiler);
132 
133    ir3_shader_debug = debug_get_option_ir3_shader_debug();
134    ir3_shader_override_path =
135       __normal_user() ? debug_get_option_ir3_shader_override_path() : NULL;
136 
137    if (ir3_shader_override_path) {
138       ir3_shader_debug |= IR3_DBG_NOCACHE;
139    }
140 
141    compiler->dev = dev;
142    compiler->dev_id = dev_id;
143    compiler->gen = fd_dev_gen(dev_id);
144    compiler->is_64bit = fd_dev_64b(dev_id);
145    compiler->options = *options;
146 
147    /* TODO see if older GPU's were different here */
148    compiler->branchstack_size = 64;
149    compiler->wave_granularity = dev_info->wave_granularity;
150    compiler->max_waves = dev_info->max_waves;
151 
152    compiler->max_variable_workgroup_size = 1024;
153 
154    compiler->local_mem_size = dev_info->cs_shared_mem_size;
155 
156    compiler->num_predicates = 1;
157    compiler->bitops_can_write_predicates = false;
158    compiler->has_branch_and_or = false;
159    compiler->has_rpt_bary_f = false;
160 
161    if (compiler->gen >= 6) {
162       compiler->samgq_workaround = true;
163       /* a6xx split the pipeline state into geometry and fragment state, in
164        * order to let the VS run ahead of the FS. As a result there are now
165        * separate const files for the the fragment shader and everything
166        * else, and separate limits. There seems to be a shared limit, but
167        * it's higher than the vert or frag limits.
168        *
169        * Also, according to the observation on a630/a650/a660, max_const_pipeline
170        * has to be 512 when all geometry stages are present. Otherwise a gpu hang
171        * happens. Accordingly maximum safe size for each stage should be under
172        * (max_const_pipeline / 5 (stages)) with 4 vec4's alignment considered for
173        * const files.
174        *
175        * Only when VS and FS stages are present, the limit is 640.
176        *
177        * TODO: The shared limit seems to be different on different models.
178        */
179       compiler->max_const_pipeline = 512;
180       compiler->max_const_frag = 512;
181       compiler->max_const_geom = 512;
182       compiler->max_const_safe = 100;
183 
184       /* Compute shaders don't share a const file with the FS. Instead they
185        * have their own file, which is smaller than the FS one. On a7xx the size
186        * was doubled, although this doesn't work on X1-85.
187        *
188        * TODO: is this true on earlier gen's?
189        */
190       compiler->max_const_compute =
191          (compiler->gen >= 7 && !dev_info->a7xx.compute_constlen_quirk) ? 512 : 256;
192 
193       /* TODO: implement clip+cull distances on earlier gen's */
194       compiler->has_clip_cull = true;
195 
196       compiler->has_preamble = true;
197 
198       compiler->tess_use_shared = dev_info->a6xx.tess_use_shared;
199 
200       compiler->has_getfiberid = dev_info->a6xx.has_getfiberid;
201 
202       compiler->has_dp2acc = dev_info->a6xx.has_dp2acc;
203       compiler->has_dp4acc = dev_info->a6xx.has_dp4acc;
204       compiler->has_compliant_dp4acc = dev_info->a7xx.has_compliant_dp4acc;
205 
206       if (compiler->gen == 6 && options->shared_push_consts) {
207          compiler->shared_consts_base_offset = 504;
208          compiler->shared_consts_size = 8;
209          compiler->geom_shared_consts_size_quirk = 16;
210       } else {
211          compiler->shared_consts_base_offset = -1;
212          compiler->shared_consts_size = 0;
213          compiler->geom_shared_consts_size_quirk = 0;
214       }
215 
216       compiler->has_fs_tex_prefetch = dev_info->a6xx.has_fs_tex_prefetch;
217       compiler->stsc_duplication_quirk = dev_info->a7xx.stsc_duplication_quirk;
218       compiler->load_shader_consts_via_preamble = dev_info->a7xx.load_shader_consts_via_preamble;
219       compiler->load_inline_uniforms_via_preamble_ldgk = dev_info->a7xx.load_inline_uniforms_via_preamble_ldgk;
220       compiler->num_predicates = 4;
221       compiler->bitops_can_write_predicates = true;
222       compiler->has_branch_and_or = true;
223       compiler->has_predication = true;
224       compiler->predtf_nop_quirk = dev_info->a6xx.predtf_nop_quirk;
225       compiler->prede_nop_quirk = dev_info->a6xx.prede_nop_quirk;
226       compiler->has_scalar_alu = dev_info->a6xx.has_scalar_alu;
227       compiler->has_isam_v = dev_info->a6xx.has_isam_v;
228       compiler->has_ssbo_imm_offsets = dev_info->a6xx.has_ssbo_imm_offsets;
229       compiler->fs_must_have_non_zero_constlen_quirk = dev_info->a7xx.fs_must_have_non_zero_constlen_quirk;
230       compiler->has_early_preamble = dev_info->a6xx.has_early_preamble;
231       compiler->has_rpt_bary_f = true;
232       compiler->has_shfl = true;
233       compiler->reading_shading_rate_requires_smask_quirk =
234          dev_info->a7xx.reading_shading_rate_requires_smask_quirk;
235    } else {
236       compiler->max_const_pipeline = 512;
237       compiler->max_const_geom = 512;
238       compiler->max_const_frag = 512;
239       compiler->max_const_compute = 512;
240 
241       /* Note: this will have to change if/when we support tess+GS on
242        * earlier gen's.
243        */
244       compiler->max_const_safe = 256;
245 
246       compiler->has_scalar_alu = false;
247       compiler->has_isam_v = false;
248       compiler->has_ssbo_imm_offsets = false;
249       compiler->has_early_preamble = false;
250    }
251 
252    /* This is just a guess for a4xx. */
253    compiler->pvtmem_per_fiber_align = compiler->gen >= 4 ? 512 : 128;
254    /* TODO: implement private memory on earlier gen's */
255    compiler->has_pvtmem = compiler->gen >= 5;
256 
257    compiler->has_isam_ssbo = compiler->gen >= 6;
258 
259    if (compiler->gen >= 6) {
260       compiler->reg_size_vec4 = dev_info->a6xx.reg_size_vec4;
261    } else if (compiler->gen >= 4) {
262       /* On a4xx-a5xx, using r24.x and above requires using the smallest
263        * threadsize.
264        */
265       compiler->reg_size_vec4 = 48;
266    } else {
267       /* TODO: confirm this */
268       compiler->reg_size_vec4 = 96;
269    }
270 
271    compiler->threadsize_base = dev_info->threadsize_base;
272 
273    if (compiler->gen >= 4) {
274       /* need special handling for "flat" */
275       compiler->flat_bypass = true;
276       compiler->levels_add_one = false;
277       compiler->unminify_coords = false;
278       compiler->txf_ms_with_isaml = false;
279       compiler->array_index_add_half = true;
280       compiler->instr_align = 16;
281       compiler->const_upload_unit = 4;
282    } else {
283       /* no special handling for "flat" */
284       compiler->flat_bypass = false;
285       compiler->levels_add_one = true;
286       compiler->unminify_coords = true;
287       compiler->txf_ms_with_isaml = true;
288       compiler->array_index_add_half = false;
289       compiler->instr_align = 4;
290       compiler->const_upload_unit = 8;
291    }
292 
293    compiler->bool_type = (compiler->gen >= 5) ? TYPE_U16 : TYPE_U32;
294    compiler->has_shared_regfile = compiler->gen >= 5;
295    compiler->has_bitwise_triops = compiler->gen >= 5;
296 
297    /* The driver can't request this unless preambles are supported. */
298    if (options->push_ubo_with_preamble)
299       assert(compiler->has_preamble);
300 
301    /* Set up nir shader compiler options, using device-specific overrides of our base settings. */
302    compiler->nir_options = ir3_base_options;
303    compiler->nir_options.has_iadd3 = dev_info->a6xx.has_sad;
304 
305    if (compiler->gen >= 6) {
306       compiler->nir_options.vectorize_io = true,
307       compiler->nir_options.force_indirect_unrolling = nir_var_all,
308       compiler->nir_options.lower_device_index_to_zero = true;
309 
310       if (dev_info->a6xx.has_dp2acc || dev_info->a6xx.has_dp4acc) {
311          compiler->nir_options.has_udot_4x8 =
312             compiler->nir_options.has_udot_4x8_sat = true;
313          compiler->nir_options.has_sudot_4x8 =
314             compiler->nir_options.has_sudot_4x8_sat = true;
315       }
316 
317       if (dev_info->a6xx.has_dp4acc && dev_info->a7xx.has_compliant_dp4acc) {
318          compiler->nir_options.has_sdot_4x8 =
319             compiler->nir_options.has_sdot_4x8_sat = true;
320       }
321    } else if (compiler->gen >= 3 && compiler->gen <= 5) {
322       compiler->nir_options.vertex_id_zero_based = true;
323    } else if (compiler->gen <= 2) {
324       /* a2xx compiler doesn't handle indirect: */
325       compiler->nir_options.force_indirect_unrolling = nir_var_all;
326    }
327 
328    if (options->lower_base_vertex) {
329       compiler->nir_options.lower_base_vertex = true;
330    }
331 
332    /* 16-bit ALU op generation is mostly controlled by frontend compiler options, but
333     * this core NIR option enables some optimizations of 16-bit operations.
334     */
335    if (compiler->gen >= 5 && !(ir3_shader_debug & IR3_DBG_NOFP16))
336       compiler->nir_options.support_16bit_alu = true;
337 
338    compiler->nir_options.support_indirect_inputs = (uint8_t)BITFIELD_MASK(PIPE_SHADER_TYPES);
339    compiler->nir_options.support_indirect_outputs = (uint8_t)BITFIELD_MASK(PIPE_SHADER_TYPES);
340 
341    if (!options->disable_cache)
342       ir3_disk_cache_init(compiler);
343 
344    return compiler;
345 }
346 
347 const nir_shader_compiler_options *
ir3_get_compiler_options(struct ir3_compiler * compiler)348 ir3_get_compiler_options(struct ir3_compiler *compiler)
349 {
350    return &compiler->nir_options;
351 }
352