nir/scheduler: Move nir_scheduler to its own header
[mesa.git] / src / compiler / nir / nir_lower_wpos_ytransform.c
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 #include "program/prog_instruction.h"
27
28 /* Lower gl_FragCoord (and fddy) to account for driver's requested coordinate-
29 * origin and pixel-center vs. shader. If transformation is required, a
30 * gl_FbWposYTransform uniform is inserted (with the specified state-slots)
31 * and additional instructions are inserted to transform gl_FragCoord (and
32 * fddy src arg).
33 *
34 * This is based on the logic in emit_wpos()/emit_wpos_adjustment() in TGSI
35 * compiler.
36 *
37 * Run before nir_lower_io.
38 */
39
40 typedef struct {
41 const nir_lower_wpos_ytransform_options *options;
42 nir_shader *shader;
43 nir_builder b;
44 nir_variable *transform;
45 } lower_wpos_ytransform_state;
46
47 static nir_ssa_def *
48 get_transform(lower_wpos_ytransform_state *state)
49 {
50 if (state->transform == NULL) {
51 /* NOTE: name must be prefixed w/ "gl_" to trigger slot based
52 * special handling in uniform setup:
53 */
54 nir_variable *var = nir_variable_create(state->shader,
55 nir_var_uniform,
56 glsl_vec4_type(),
57 "gl_FbWposYTransform");
58
59 var->num_state_slots = 1;
60 var->state_slots = ralloc_array(var, nir_state_slot, 1);
61 var->state_slots[0].swizzle = SWIZZLE_XYZW;
62 memcpy(var->state_slots[0].tokens, state->options->state_tokens,
63 sizeof(var->state_slots[0].tokens));
64 var->data.how_declared = nir_var_hidden;
65 state->transform = var;
66 }
67 return nir_load_var(&state->b, state->transform);
68 }
69
70 /* NIR equiv of TGSI CMP instruction: */
71 static nir_ssa_def *
72 nir_cmp(nir_builder *b, nir_ssa_def *src0, nir_ssa_def *src1, nir_ssa_def *src2)
73 {
74 return nir_bcsel(b, nir_flt(b, src0, nir_imm_float(b, 0.0)), src1, src2);
75 }
76
77 /* see emit_wpos_adjustment() in st_mesa_to_tgsi.c */
78 static void
79 emit_wpos_adjustment(lower_wpos_ytransform_state *state,
80 nir_intrinsic_instr *intr, bool invert,
81 float adjX, float adjY[2])
82 {
83 nir_builder *b = &state->b;
84 nir_ssa_def *wpostrans, *wpos_temp, *wpos_temp_y, *wpos_input;
85
86 assert(intr->dest.is_ssa);
87 wpos_input = &intr->dest.ssa;
88
89 b->cursor = nir_after_instr(&intr->instr);
90
91 wpostrans = get_transform(state);
92
93 /* First, apply the coordinate shift: */
94 if (adjX || adjY[0] || adjY[1]) {
95 if (adjY[0] != adjY[1]) {
96 /* Adjust the y coordinate by adjY[1] or adjY[0] respectively
97 * depending on whether inversion is actually going to be applied
98 * or not, which is determined by testing against the inversion
99 * state variable used below, which will be either +1 or -1.
100 */
101 nir_ssa_def *adj_temp;
102
103 adj_temp = nir_cmp(b,
104 nir_channel(b, wpostrans, invert ? 2 : 0),
105 nir_imm_vec4(b, adjX, adjY[0], 0.0f, 0.0f),
106 nir_imm_vec4(b, adjX, adjY[1], 0.0f, 0.0f));
107
108 wpos_temp = nir_fadd(b, wpos_input, adj_temp);
109 } else {
110 wpos_temp = nir_fadd(b,
111 wpos_input,
112 nir_imm_vec4(b, adjX, adjY[0], 0.0f, 0.0f));
113 }
114 wpos_input = wpos_temp;
115 } else {
116 /* MOV wpos_temp, input[wpos]
117 */
118 wpos_temp = wpos_input;
119 }
120
121 /* Now the conditional y flip: STATE_FB_WPOS_Y_TRANSFORM.xy/zw will be
122 * inversion/identity, or the other way around if we're drawing to an FBO.
123 */
124 if (invert) {
125 /* wpos_temp.y = wpos_input * wpostrans.xxxx + wpostrans.yyyy */
126 wpos_temp_y = nir_fadd(b, nir_fmul(b, nir_channel(b, wpos_temp, 1),
127 nir_channel(b, wpostrans, 0)),
128 nir_channel(b, wpostrans, 1));
129 } else {
130 /* wpos_temp.y = wpos_input * wpostrans.zzzz + wpostrans.wwww */
131 wpos_temp_y = nir_fadd(b, nir_fmul(b, nir_channel(b, wpos_temp, 1),
132 nir_channel(b, wpostrans, 2)),
133 nir_channel(b, wpostrans, 3));
134 }
135
136 wpos_temp = nir_vec4(b,
137 nir_channel(b, wpos_temp, 0),
138 wpos_temp_y,
139 nir_channel(b, wpos_temp, 2),
140 nir_channel(b, wpos_temp, 3));
141
142 nir_ssa_def_rewrite_uses_after(&intr->dest.ssa,
143 nir_src_for_ssa(wpos_temp),
144 wpos_temp->parent_instr);
145 }
146
147 static void
148 lower_fragcoord(lower_wpos_ytransform_state *state, nir_intrinsic_instr *intr)
149 {
150 const nir_lower_wpos_ytransform_options *options = state->options;
151 float adjX = 0.0f;
152 float adjY[2] = { 0.0f, 0.0f };
153 bool invert = false;
154
155 /* Based on logic in emit_wpos():
156 *
157 * Query the pixel center conventions supported by the pipe driver and set
158 * adjX, adjY to help out if it cannot handle the requested one internally.
159 *
160 * The bias of the y-coordinate depends on whether y-inversion takes place
161 * (adjY[1]) or not (adjY[0]), which is in turn dependent on whether we are
162 * drawing to an FBO (causes additional inversion), and whether the pipe
163 * driver origin and the requested origin differ (the latter condition is
164 * stored in the 'invert' variable).
165 *
166 * For height = 100 (i = integer, h = half-integer, l = lower, u = upper):
167 *
168 * center shift only:
169 * i -> h: +0.5
170 * h -> i: -0.5
171 *
172 * inversion only:
173 * l,i -> u,i: ( 0.0 + 1.0) * -1 + 100 = 99
174 * l,h -> u,h: ( 0.5 + 0.0) * -1 + 100 = 99.5
175 * u,i -> l,i: (99.0 + 1.0) * -1 + 100 = 0
176 * u,h -> l,h: (99.5 + 0.0) * -1 + 100 = 0.5
177 *
178 * inversion and center shift:
179 * l,i -> u,h: ( 0.0 + 0.5) * -1 + 100 = 99.5
180 * l,h -> u,i: ( 0.5 + 0.5) * -1 + 100 = 99
181 * u,i -> l,h: (99.0 + 0.5) * -1 + 100 = 0.5
182 * u,h -> l,i: (99.5 + 0.5) * -1 + 100 = 0
183 */
184
185 if (state->shader->info.fs.origin_upper_left) {
186 /* Fragment shader wants origin in upper-left */
187 if (options->fs_coord_origin_upper_left) {
188 /* the driver supports upper-left origin */
189 } else if (options->fs_coord_origin_lower_left) {
190 /* the driver supports lower-left origin, need to invert Y */
191 invert = true;
192 } else {
193 unreachable("invalid options");
194 }
195 } else {
196 /* Fragment shader wants origin in lower-left */
197 if (options->fs_coord_origin_lower_left) {
198 /* the driver supports lower-left origin */
199 } else if (options->fs_coord_origin_upper_left) {
200 /* the driver supports upper-left origin, need to invert Y */
201 invert = true;
202 } else {
203 unreachable("invalid options");
204 }
205 }
206
207 if (state->shader->info.fs.pixel_center_integer) {
208 /* Fragment shader wants pixel center integer */
209 if (options->fs_coord_pixel_center_integer) {
210 /* the driver supports pixel center integer */
211 adjY[1] = 1.0f;
212 } else if (options->fs_coord_pixel_center_half_integer) {
213 /* the driver supports pixel center half integer, need to bias X,Y */
214 adjX = -0.5f;
215 adjY[0] = -0.5f;
216 adjY[1] = 0.5f;
217 } else {
218 unreachable("invalid options");
219 }
220 } else {
221 /* Fragment shader wants pixel center half integer */
222 if (options->fs_coord_pixel_center_half_integer) {
223 /* the driver supports pixel center half integer */
224 } else if (options->fs_coord_pixel_center_integer) {
225 /* the driver supports pixel center integer, need to bias X,Y */
226 adjX = adjY[0] = adjY[1] = 0.5f;
227 } else {
228 unreachable("invalid options");
229 }
230 }
231
232 emit_wpos_adjustment(state, intr, invert, adjX, adjY);
233 }
234
235 static void
236 lower_load_pointcoord(lower_wpos_ytransform_state *state,
237 nir_intrinsic_instr *intr)
238 {
239 nir_builder *b = &state->b;
240 b->cursor = nir_after_instr(&intr->instr);
241
242 nir_ssa_def *pntc = &intr->dest.ssa;
243 nir_ssa_def *transform = get_transform(state);
244 nir_ssa_def *y = nir_channel(b, pntc, 1);
245 /* The offset is 1 if we're flipping, 0 otherwise. */
246 nir_ssa_def *offset = nir_fmax(b, nir_channel(b, transform, 2),
247 nir_imm_float(b, 0.0));
248 /* Flip the sign of y if we're flipping. */
249 nir_ssa_def *scaled = nir_fmul(b, y, nir_channel(b, transform, 0));
250
251 /* Reassemble the vector. */
252 nir_ssa_def *flipped_pntc = nir_vec2(b,
253 nir_channel(b, pntc, 0),
254 nir_fadd(b, offset, scaled));
255
256 nir_ssa_def_rewrite_uses_after(&intr->dest.ssa, nir_src_for_ssa(flipped_pntc),
257 flipped_pntc->parent_instr);
258 }
259
260 /* turns 'fddy(p)' into 'fddy(fmul(p, transform.x))' */
261 static void
262 lower_fddy(lower_wpos_ytransform_state *state, nir_alu_instr *fddy)
263 {
264 nir_builder *b = &state->b;
265 nir_ssa_def *p, *pt, *trans;
266
267 b->cursor = nir_before_instr(&fddy->instr);
268
269 p = nir_ssa_for_alu_src(b, fddy, 0);
270 trans = nir_channel(b, get_transform(state), 0);
271 if (p->bit_size == 16)
272 trans = nir_f2f16(b, trans);
273
274 pt = nir_fmul(b, p, trans);
275
276 nir_instr_rewrite_src(&fddy->instr,
277 &fddy->src[0].src,
278 nir_src_for_ssa(pt));
279
280 for (unsigned i = 0; i < 4; i++)
281 fddy->src[0].swizzle[i] = MIN2(i, pt->num_components - 1);
282 }
283
284 /* Multiply interp_deref_at_offset's offset by transform.x to flip it. */
285 static void
286 lower_interp_deref_at_offset(lower_wpos_ytransform_state *state,
287 nir_intrinsic_instr *interp)
288 {
289 nir_builder *b = &state->b;
290 nir_ssa_def *offset;
291 nir_ssa_def *flip_y;
292
293 b->cursor = nir_before_instr(&interp->instr);
294
295 offset = nir_ssa_for_src(b, interp->src[1], 2);
296 flip_y = nir_fmul(b, nir_channel(b, offset, 1),
297 nir_channel(b, get_transform(state), 0));
298 nir_instr_rewrite_src(&interp->instr, &interp->src[1],
299 nir_src_for_ssa(nir_vec2(b, nir_channel(b, offset, 0),
300 flip_y)));
301 }
302
303 static void
304 lower_load_sample_pos(lower_wpos_ytransform_state *state,
305 nir_intrinsic_instr *intr)
306 {
307 nir_builder *b = &state->b;
308 b->cursor = nir_after_instr(&intr->instr);
309
310 nir_ssa_def *pos = &intr->dest.ssa;
311 nir_ssa_def *scale = nir_channel(b, get_transform(state), 0);
312 nir_ssa_def *neg_scale = nir_channel(b, get_transform(state), 2);
313 /* Either y or 1-y for scale equal to 1 or -1 respectively. */
314 nir_ssa_def *flipped_y =
315 nir_fadd(b, nir_fmax(b, neg_scale, nir_imm_float(b, 0.0)),
316 nir_fmul(b, nir_channel(b, pos, 1), scale));
317 nir_ssa_def *flipped_pos = nir_vec2(b, nir_channel(b, pos, 0), flipped_y);
318
319 nir_ssa_def_rewrite_uses_after(&intr->dest.ssa, nir_src_for_ssa(flipped_pos),
320 flipped_pos->parent_instr);
321 }
322
323 static void
324 lower_wpos_ytransform_block(lower_wpos_ytransform_state *state, nir_block *block)
325 {
326 nir_foreach_instr_safe(instr, block) {
327 if (instr->type == nir_instr_type_intrinsic) {
328 nir_intrinsic_instr *intr = nir_instr_as_intrinsic(instr);
329 if (intr->intrinsic == nir_intrinsic_load_deref) {
330 nir_deref_instr *deref = nir_src_as_deref(intr->src[0]);
331 nir_variable *var = nir_deref_instr_get_variable(deref);
332
333 if ((var->data.mode == nir_var_shader_in &&
334 var->data.location == VARYING_SLOT_POS) ||
335 (var->data.mode == nir_var_system_value &&
336 var->data.location == SYSTEM_VALUE_FRAG_COORD)) {
337 /* gl_FragCoord should not have array/struct derefs: */
338 lower_fragcoord(state, intr);
339 } else if (var->data.mode == nir_var_system_value &&
340 var->data.location == SYSTEM_VALUE_SAMPLE_POS) {
341 lower_load_sample_pos(state, intr);
342 } else if (var->data.mode == nir_var_shader_in &&
343 var->data.location == VARYING_SLOT_PNTC &&
344 state->shader->options->lower_wpos_pntc) {
345 lower_load_pointcoord(state, intr);
346 }
347 } else if (intr->intrinsic == nir_intrinsic_load_frag_coord) {
348 lower_fragcoord(state, intr);
349 } else if (intr->intrinsic == nir_intrinsic_load_sample_pos) {
350 lower_load_sample_pos(state, intr);
351 } else if (intr->intrinsic == nir_intrinsic_interp_deref_at_offset) {
352 lower_interp_deref_at_offset(state, intr);
353 }
354 } else if (instr->type == nir_instr_type_alu) {
355 nir_alu_instr *alu = nir_instr_as_alu(instr);
356 if (alu->op == nir_op_fddy ||
357 alu->op == nir_op_fddy_fine ||
358 alu->op == nir_op_fddy_coarse)
359 lower_fddy(state, alu);
360 }
361 }
362 }
363
364 static void
365 lower_wpos_ytransform_impl(lower_wpos_ytransform_state *state, nir_function_impl *impl)
366 {
367 nir_builder_init(&state->b, impl);
368
369 nir_foreach_block(block, impl) {
370 lower_wpos_ytransform_block(state, block);
371 }
372 nir_metadata_preserve(impl, nir_metadata_block_index |
373 nir_metadata_dominance);
374 }
375
376 bool
377 nir_lower_wpos_ytransform(nir_shader *shader,
378 const nir_lower_wpos_ytransform_options *options)
379 {
380 lower_wpos_ytransform_state state = {
381 .options = options,
382 .shader = shader,
383 };
384
385 assert(shader->info.stage == MESA_SHADER_FRAGMENT);
386
387 nir_foreach_function(function, shader) {
388 if (function->impl)
389 lower_wpos_ytransform_impl(&state, function->impl);
390 }
391
392 return state.transform != NULL;
393 }