1 /*
2 * Copyright © 2015 Red Hat
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 FROM,
20 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
21 * SOFTWARE.
22 */
23
24 #include "nir.h"
25 #include "nir_builder.h"
26
27 /* Lower gl_FragCoord (and ddy) to account for driver's requested coordinate-
28 * origin and pixel-center vs. shader. If transformation is required, a
29 * gl_FbWposYTransform uniform is inserted (with the specified state-slots)
30 * and additional instructions are inserted to transform gl_FragCoord (and
31 * ddy src arg).
32 */
33
34 typedef struct {
35 const nir_lower_wpos_ytransform_options *options;
36 nir_builder b;
37 nir_def *transform;
38 } lower_wpos_ytransform_state;
39
40 static nir_def *
get_transform(lower_wpos_ytransform_state * state)41 get_transform(lower_wpos_ytransform_state *state)
42 {
43 if (state->transform == NULL) {
44 /* NOTE: name must be prefixed w/ "gl_" to trigger slot based
45 * special handling in uniform setup:
46 */
47 nir_variable *var = nir_state_variable_create(state->b.shader,
48 glsl_vec4_type(),
49 "gl_FbWposYTransform",
50 state->options->state_tokens);
51
52 var->data.how_declared = nir_var_hidden;
53 state->b.cursor = nir_before_impl(nir_shader_get_entrypoint(state->b.shader));
54 state->transform = nir_load_var(&state->b, var);
55 }
56 return state->transform;
57 }
58
59 static bool
emit_wpos_adjustment(lower_wpos_ytransform_state * state,nir_intrinsic_instr * intr,bool invert,float adjX,float adjY[2])60 emit_wpos_adjustment(lower_wpos_ytransform_state *state,
61 nir_intrinsic_instr *intr, bool invert,
62 float adjX, float adjY[2])
63 {
64 nir_builder *b = &state->b;
65 unsigned c = 0;
66 if (nir_intrinsic_has_component(intr)) {
67 c = nir_intrinsic_component(intr);
68 /* this pass only alters the first two components */
69 if (c > 1)
70 return false;
71 }
72
73 if (c == 0 && intr->num_components == 1 && adjX == 0.0)
74 return false;
75
76 nir_def *wpostrans = get_transform(state);
77
78 b->cursor = nir_after_instr(&intr->instr);
79 nir_def *wpos[4] = { NULL };
80 for (unsigned i = 0; i < intr->num_components; i++)
81 wpos[i + c] = nir_channel(b, &intr->def, i);
82
83 /* First, apply the coordinate shift: */
84 if (wpos[0] && adjX)
85 wpos[0] = nir_fadd_imm(b, wpos[0], adjX);
86
87 if (wpos[1] && adjY[0] != adjY[1]) {
88 /* Adjust the y coordinate by adjY[1] or adjY[0] respectively
89 * depending on whether inversion is actually going to be applied
90 * or not, which is determined by testing against the inversion
91 * state variable used below, which will be either +1 or -1.
92 */
93 nir_def *cond = nir_flt_imm(b, nir_channel(b, wpostrans, invert ? 2 : 0), 0.0);
94 nir_def *adj_temp = nir_bcsel(b, cond,
95 nir_imm_float(b, adjY[0]),
96 nir_imm_float(b, adjY[1]));
97
98 wpos[1] = nir_fadd(b, wpos[1], adj_temp);
99 } else if (wpos[1] && adjY[0]) {
100 wpos[1] = nir_fadd_imm(b, wpos[1], adjY[0]);
101 }
102
103 if (wpos[1]) {
104 /* Now the conditional y flip: STATE_FB_WPOS_Y_TRANSFORM.xy/zw will be
105 * inversion/identity, or the other way around if we're drawing to an FBO.
106 */
107 unsigned base = invert ? 0 : 2;
108 /* wpos.y = wpos.y * trans.x/z + trans.y/w */
109 wpos[1] = nir_ffma(b, wpos[1], nir_channel(b, wpostrans, base),
110 nir_channel(b, wpostrans, base + 1));
111 }
112
113 nir_def *new_wpos = nir_vec(b, &wpos[c], intr->num_components);
114
115 nir_def_rewrite_uses_after(&intr->def, new_wpos,
116 new_wpos->parent_instr);
117
118 return true;
119 }
120
121 static bool
lower_fragcoord(lower_wpos_ytransform_state * state,nir_intrinsic_instr * intr)122 lower_fragcoord(lower_wpos_ytransform_state *state, nir_intrinsic_instr *intr)
123 {
124 const nir_lower_wpos_ytransform_options *options = state->options;
125 const struct shader_info *info = &state->b.shader->info;
126 float adjX = 0.0f;
127 float adjY[2] = { 0.0f, 0.0f };
128 bool invert = false;
129
130 /* Based on logic in emit_wpos():
131 *
132 * Query the pixel center conventions supported by the pipe driver and set
133 * adjX, adjY to help out if it cannot handle the requested one internally.
134 *
135 * The bias of the y-coordinate depends on whether y-inversion takes place
136 * (adjY[1]) or not (adjY[0]), which is in turn dependent on whether we are
137 * drawing to an FBO (causes additional inversion), and whether the pipe
138 * driver origin and the requested origin differ (the latter condition is
139 * stored in the 'invert' variable).
140 *
141 * For height = 100 (i = integer, h = half-integer, l = lower, u = upper):
142 *
143 * center shift only:
144 * i -> h: +0.5
145 * h -> i: -0.5
146 *
147 * inversion only:
148 * l,i -> u,i: ( 0.0 + 1.0) * -1 + 100 = 99
149 * l,h -> u,h: ( 0.5 + 0.0) * -1 + 100 = 99.5
150 * u,i -> l,i: (99.0 + 1.0) * -1 + 100 = 0
151 * u,h -> l,h: (99.5 + 0.0) * -1 + 100 = 0.5
152 *
153 * inversion and center shift:
154 * l,i -> u,h: ( 0.0 + 0.5) * -1 + 100 = 99.5
155 * l,h -> u,i: ( 0.5 + 0.5) * -1 + 100 = 99
156 * u,i -> l,h: (99.0 + 0.5) * -1 + 100 = 0.5
157 * u,h -> l,i: (99.5 + 0.5) * -1 + 100 = 0
158 */
159
160 if (info->fs.origin_upper_left) {
161 /* Fragment shader wants origin in upper-left */
162 if (options->fs_coord_origin_upper_left) {
163 /* the driver supports upper-left origin */
164 } else if (options->fs_coord_origin_lower_left) {
165 /* the driver supports lower-left origin, need to invert Y */
166 invert = true;
167 } else {
168 unreachable("invalid options");
169 }
170 } else {
171 /* Fragment shader wants origin in lower-left */
172 if (options->fs_coord_origin_lower_left) {
173 /* the driver supports lower-left origin */
174 } else if (options->fs_coord_origin_upper_left) {
175 /* the driver supports upper-left origin, need to invert Y */
176 invert = true;
177 } else {
178 unreachable("invalid options");
179 }
180 }
181
182 if (info->fs.pixel_center_integer) {
183 /* Fragment shader wants pixel center integer */
184 if (options->fs_coord_pixel_center_integer) {
185 /* the driver supports pixel center integer */
186 adjY[1] = 1.0f;
187 } else if (options->fs_coord_pixel_center_half_integer) {
188 /* the driver supports pixel center half integer, need to bias X,Y */
189 adjX = -0.5f;
190 adjY[0] = -0.5f;
191 adjY[1] = 0.5f;
192 } else {
193 unreachable("invalid options");
194 }
195 } else {
196 /* Fragment shader wants pixel center half integer */
197 if (options->fs_coord_pixel_center_half_integer) {
198 /* the driver supports pixel center half integer */
199 } else if (options->fs_coord_pixel_center_integer) {
200 /* the driver supports pixel center integer, need to bias X,Y */
201 adjX = adjY[0] = adjY[1] = 0.5f;
202 } else {
203 unreachable("invalid options");
204 }
205 }
206
207 return emit_wpos_adjustment(state, intr, invert, adjX, adjY);
208 }
209
210 /* turns 'ddy(p)' into 'ddy(fmul(p, transform.x))' */
211 static bool
lower_ddy(lower_wpos_ytransform_state * state,nir_intrinsic_instr * ddy)212 lower_ddy(lower_wpos_ytransform_state *state, nir_intrinsic_instr *ddy)
213 {
214 nir_builder *b = &state->b;
215 nir_def *wpostrans = get_transform(state);
216
217 b->cursor = nir_before_instr(&ddy->instr);
218
219 nir_def *p = ddy->src[0].ssa;
220 nir_def *trans = nir_f2fN(b, nir_channel(b, wpostrans, 0), p->bit_size);
221 nir_def *pt = nir_fmul(b, p, trans);
222
223 nir_src_rewrite(&ddy->src[0], pt);
224
225 return true;
226 }
227
228 /* Multiply interp_deref_at_offset's or load_barycentric_at_offset's offset
229 * by transform.x to flip it.
230 */
231 static bool
lower_interp_deref_or_load_baryc_at_offset(lower_wpos_ytransform_state * state,nir_intrinsic_instr * intr,unsigned offset_src)232 lower_interp_deref_or_load_baryc_at_offset(lower_wpos_ytransform_state *state,
233 nir_intrinsic_instr *intr,
234 unsigned offset_src)
235 {
236 nir_builder *b = &state->b;
237 nir_def *wpostrans = get_transform(state);
238
239 b->cursor = nir_before_instr(&intr->instr);
240
241 nir_def *offset = intr->src[offset_src].ssa;
242 nir_def *flip_y = nir_fmul(b, nir_channel(b, offset, 1),
243 nir_channel(b, wpostrans, 0));
244 offset = nir_vector_insert_imm(b, offset, flip_y, 1);
245 nir_src_rewrite(&intr->src[offset_src], offset);
246
247 return true;
248 }
249
250 static bool
lower_load_sample_pos(lower_wpos_ytransform_state * state,nir_intrinsic_instr * intr)251 lower_load_sample_pos(lower_wpos_ytransform_state *state,
252 nir_intrinsic_instr *intr)
253 {
254 nir_builder *b = &state->b;
255 nir_def *wpostrans = get_transform(state);
256 b->cursor = nir_after_instr(&intr->instr);
257
258 nir_def *pos = &intr->def;
259 nir_def *scale = nir_channel(b, wpostrans, 0);
260 nir_def *neg_scale = nir_channel(b, wpostrans, 2);
261 /* Either y or 1-y for scale equal to 1 or -1 respectively. */
262 nir_def *flipped_y = nir_ffma(b, nir_channel(b, pos, 1), scale,
263 nir_fmax(b, neg_scale, nir_imm_float(b, 0.0)));
264 nir_def *flipped_pos = nir_vector_insert_imm(b, pos, flipped_y, 1);
265
266 nir_def_rewrite_uses_after(&intr->def, flipped_pos,
267 flipped_pos->parent_instr);
268
269 return true;
270 }
271
272 static bool
lower_wpos_ytransform_instr(nir_builder * b,nir_intrinsic_instr * intr,void * data)273 lower_wpos_ytransform_instr(nir_builder *b, nir_intrinsic_instr *intr,
274 void *data)
275 {
276 lower_wpos_ytransform_state *state = data;
277 state->b = *b;
278
279 switch (intr->intrinsic) {
280 case nir_intrinsic_load_deref: {
281 nir_deref_instr *deref = nir_src_as_deref(intr->src[0]);
282 nir_variable *var = nir_deref_instr_get_variable(deref);
283 if ((var->data.mode == nir_var_shader_in &&
284 var->data.location == VARYING_SLOT_POS) ||
285 (var->data.mode == nir_var_system_value &&
286 var->data.location == SYSTEM_VALUE_FRAG_COORD)) {
287 /* gl_FragCoord should not have array/struct derefs: */
288 return lower_fragcoord(state, intr);
289 } else if (var->data.mode == nir_var_system_value &&
290 var->data.location == SYSTEM_VALUE_SAMPLE_POS) {
291 return lower_load_sample_pos(state, intr);
292 }
293 return false;
294 }
295 case nir_intrinsic_load_interpolated_input: {
296 nir_io_semantics sem = nir_intrinsic_io_semantics(intr);
297 if (sem.location == VARYING_SLOT_POS)
298 return lower_fragcoord(state, intr);
299 return false;
300 }
301 case nir_intrinsic_load_frag_coord:
302 return lower_fragcoord(state, intr);
303 case nir_intrinsic_load_sample_pos:
304 return lower_load_sample_pos(state, intr);
305 case nir_intrinsic_interp_deref_at_offset:
306 return lower_interp_deref_or_load_baryc_at_offset(state, intr, 1);
307 case nir_intrinsic_load_barycentric_at_offset:
308 return lower_interp_deref_or_load_baryc_at_offset(state, intr, 0);
309 case nir_intrinsic_ddy:
310 case nir_intrinsic_ddy_fine:
311 case nir_intrinsic_ddy_coarse:
312 return lower_ddy(state, intr);
313 default:
314 return false;
315 }
316 }
317
318 bool
nir_lower_wpos_ytransform(nir_shader * shader,const nir_lower_wpos_ytransform_options * options)319 nir_lower_wpos_ytransform(nir_shader *shader,
320 const nir_lower_wpos_ytransform_options *options)
321 {
322 lower_wpos_ytransform_state state = {
323 .options = options,
324 };
325
326 assert(shader->info.stage == MESA_SHADER_FRAGMENT);
327
328 return nir_shader_intrinsics_pass(shader,
329 lower_wpos_ytransform_instr,
330 nir_metadata_control_flow,
331 &state);
332 }
333