2 * Copyright 2017 Advanced Micro Devices, Inc.
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 * on the rights to use, copy, modify, merge, publish, distribute, sub
8 * license, and/or sell copies of the Software, and to permit persons to whom
9 * the Software is furnished to do so, subject to the following conditions:
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
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 NON-INFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
19 * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
20 * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
21 * USE OR OTHER DEALINGS IN THE SOFTWARE.
24 #include "si_shader.h"
25 #include "si_shader_internal.h"
27 #include "ac_nir_to_llvm.h"
29 #include "tgsi/tgsi_from_mesa.h"
31 #include "compiler/nir/nir.h"
32 #include "compiler/nir_types.h"
36 type_size(const struct glsl_type
*type
)
38 return glsl_count_attribute_slots(type
, false);
41 static void scan_instruction(struct tgsi_shader_info
*info
,
44 if (instr
->type
== nir_instr_type_alu
) {
45 nir_alu_instr
*alu
= nir_instr_as_alu(instr
);
50 case nir_op_fddx_fine
:
51 case nir_op_fddy_fine
:
52 case nir_op_fddx_coarse
:
53 case nir_op_fddy_coarse
:
54 info
->uses_derivatives
= true;
59 } else if (instr
->type
== nir_instr_type_tex
) {
60 nir_tex_instr
*tex
= nir_instr_as_tex(instr
);
63 info
->samplers_declared
|=
64 u_bit_consecutive(tex
->sampler_index
, 1);
71 info
->uses_derivatives
= true;
76 } else if (instr
->type
== nir_instr_type_intrinsic
) {
77 nir_intrinsic_instr
*intr
= nir_instr_as_intrinsic(instr
);
79 switch (intr
->intrinsic
) {
80 case nir_intrinsic_load_front_face
:
81 info
->uses_frontface
= 1;
83 case nir_intrinsic_load_instance_id
:
84 info
->uses_instanceid
= 1;
86 case nir_intrinsic_load_invocation_id
:
87 info
->uses_invocationid
= true;
89 case nir_intrinsic_load_vertex_id
:
90 info
->uses_vertexid
= 1;
92 case nir_intrinsic_load_vertex_id_zero_base
:
93 info
->uses_vertexid_nobase
= 1;
95 case nir_intrinsic_load_base_vertex
:
96 info
->uses_basevertex
= 1;
98 case nir_intrinsic_load_primitive_id
:
99 info
->uses_primid
= 1;
101 case nir_intrinsic_load_sample_mask_in
:
102 info
->reads_samplemask
= true;
104 case nir_intrinsic_load_tess_level_inner
:
105 case nir_intrinsic_load_tess_level_outer
:
106 info
->reads_tess_factors
= true;
108 case nir_intrinsic_image_store
:
109 case nir_intrinsic_image_atomic_add
:
110 case nir_intrinsic_image_atomic_min
:
111 case nir_intrinsic_image_atomic_max
:
112 case nir_intrinsic_image_atomic_and
:
113 case nir_intrinsic_image_atomic_or
:
114 case nir_intrinsic_image_atomic_xor
:
115 case nir_intrinsic_image_atomic_exchange
:
116 case nir_intrinsic_image_atomic_comp_swap
:
117 case nir_intrinsic_store_ssbo
:
118 case nir_intrinsic_ssbo_atomic_add
:
119 case nir_intrinsic_ssbo_atomic_imin
:
120 case nir_intrinsic_ssbo_atomic_umin
:
121 case nir_intrinsic_ssbo_atomic_imax
:
122 case nir_intrinsic_ssbo_atomic_umax
:
123 case nir_intrinsic_ssbo_atomic_and
:
124 case nir_intrinsic_ssbo_atomic_or
:
125 case nir_intrinsic_ssbo_atomic_xor
:
126 case nir_intrinsic_ssbo_atomic_exchange
:
127 case nir_intrinsic_ssbo_atomic_comp_swap
:
128 info
->writes_memory
= true;
136 void si_nir_scan_tess_ctrl(const struct nir_shader
*nir
,
137 const struct tgsi_shader_info
*info
,
138 struct tgsi_tessctrl_info
*out
)
140 memset(out
, 0, sizeof(*out
));
142 if (nir
->info
.stage
!= MESA_SHADER_TESS_CTRL
)
145 /* Initial value = true. Here the pass will accumulate results from
146 * multiple segments surrounded by barriers. If tess factors aren't
147 * written at all, it's a shader bug and we don't care if this will be
150 out
->tessfactors_are_def_in_all_invocs
= true;
152 /* TODO: Implement scanning of tess factors, see tgsi backend. */
155 void si_nir_scan_shader(const struct nir_shader
*nir
,
156 struct tgsi_shader_info
*info
)
161 assert(nir
->info
.stage
== MESA_SHADER_VERTEX
||
162 nir
->info
.stage
== MESA_SHADER_GEOMETRY
||
163 nir
->info
.stage
== MESA_SHADER_TESS_CTRL
||
164 nir
->info
.stage
== MESA_SHADER_TESS_EVAL
||
165 nir
->info
.stage
== MESA_SHADER_FRAGMENT
);
167 info
->processor
= pipe_shader_type_from_mesa(nir
->info
.stage
);
168 info
->num_tokens
= 2; /* indicate that the shader is non-empty */
169 info
->num_instructions
= 2;
171 if (nir
->info
.stage
== MESA_SHADER_TESS_CTRL
) {
172 info
->properties
[TGSI_PROPERTY_TCS_VERTICES_OUT
] =
173 nir
->info
.tess
.tcs_vertices_out
;
176 if (nir
->info
.stage
== MESA_SHADER_TESS_EVAL
) {
177 if (nir
->info
.tess
.primitive_mode
== GL_ISOLINES
)
178 info
->properties
[TGSI_PROPERTY_TES_PRIM_MODE
] = PIPE_PRIM_LINES
;
180 info
->properties
[TGSI_PROPERTY_TES_PRIM_MODE
] = nir
->info
.tess
.primitive_mode
;
182 STATIC_ASSERT((TESS_SPACING_EQUAL
+ 1) % 3 == PIPE_TESS_SPACING_EQUAL
);
183 STATIC_ASSERT((TESS_SPACING_FRACTIONAL_ODD
+ 1) % 3 ==
184 PIPE_TESS_SPACING_FRACTIONAL_ODD
);
185 STATIC_ASSERT((TESS_SPACING_FRACTIONAL_EVEN
+ 1) % 3 ==
186 PIPE_TESS_SPACING_FRACTIONAL_EVEN
);
188 info
->properties
[TGSI_PROPERTY_TES_SPACING
] = (nir
->info
.tess
.spacing
+ 1) % 3;
189 info
->properties
[TGSI_PROPERTY_TES_VERTEX_ORDER_CW
] = !nir
->info
.tess
.ccw
;
190 info
->properties
[TGSI_PROPERTY_TES_POINT_MODE
] = nir
->info
.tess
.point_mode
;
193 if (nir
->info
.stage
== MESA_SHADER_GEOMETRY
) {
194 info
->properties
[TGSI_PROPERTY_GS_INPUT_PRIM
] = nir
->info
.gs
.input_primitive
;
195 info
->properties
[TGSI_PROPERTY_GS_OUTPUT_PRIM
] = nir
->info
.gs
.output_primitive
;
196 info
->properties
[TGSI_PROPERTY_GS_MAX_OUTPUT_VERTICES
] = nir
->info
.gs
.vertices_out
;
197 info
->properties
[TGSI_PROPERTY_GS_INVOCATIONS
] = nir
->info
.gs
.invocations
;
201 uint64_t processed_inputs
= 0;
202 unsigned num_inputs
= 0;
203 nir_foreach_variable(variable
, &nir
->inputs
) {
204 unsigned semantic_name
, semantic_index
;
205 unsigned attrib_count
= glsl_count_attribute_slots(variable
->type
,
206 nir
->info
.stage
== MESA_SHADER_VERTEX
);
208 /* Vertex shader inputs don't have semantics. The state
209 * tracker has already mapped them to attributes via
210 * variable->data.driver_location.
212 if (nir
->info
.stage
== MESA_SHADER_VERTEX
)
215 assert(nir
->info
.stage
!= MESA_SHADER_FRAGMENT
||
216 (attrib_count
== 1 && "not implemented"));
218 /* Fragment shader position is a system value. */
219 if (nir
->info
.stage
== MESA_SHADER_FRAGMENT
&&
220 variable
->data
.location
== VARYING_SLOT_POS
) {
221 if (variable
->data
.pixel_center_integer
)
222 info
->properties
[TGSI_PROPERTY_FS_COORD_PIXEL_CENTER
] =
223 TGSI_FS_COORD_PIXEL_CENTER_INTEGER
;
229 i
= variable
->data
.driver_location
;
230 if (processed_inputs
& ((uint64_t)1 << i
))
233 processed_inputs
|= ((uint64_t)1 << i
);
236 tgsi_get_gl_varying_semantic(variable
->data
.location
, true,
237 &semantic_name
, &semantic_index
);
239 info
->input_semantic_name
[i
] = semantic_name
;
240 info
->input_semantic_index
[i
] = semantic_index
;
242 if (variable
->data
.sample
)
243 info
->input_interpolate_loc
[i
] = TGSI_INTERPOLATE_LOC_SAMPLE
;
244 else if (variable
->data
.centroid
)
245 info
->input_interpolate_loc
[i
] = TGSI_INTERPOLATE_LOC_CENTROID
;
247 info
->input_interpolate_loc
[i
] = TGSI_INTERPOLATE_LOC_CENTER
;
249 enum glsl_base_type base_type
=
250 glsl_get_base_type(glsl_without_array(variable
->type
));
252 switch (variable
->data
.interpolation
) {
253 case INTERP_MODE_NONE
:
254 if (glsl_base_type_is_integer(base_type
)) {
255 info
->input_interpolate
[i
] = TGSI_INTERPOLATE_CONSTANT
;
259 if (semantic_name
== TGSI_SEMANTIC_COLOR
) {
260 info
->input_interpolate
[i
] = TGSI_INTERPOLATE_COLOR
;
261 goto persp_locations
;
264 case INTERP_MODE_SMOOTH
:
265 assert(!glsl_base_type_is_integer(base_type
));
267 info
->input_interpolate
[i
] = TGSI_INTERPOLATE_PERSPECTIVE
;
270 if (variable
->data
.sample
)
271 info
->uses_persp_sample
= true;
272 else if (variable
->data
.centroid
)
273 info
->uses_persp_centroid
= true;
275 info
->uses_persp_center
= true;
278 case INTERP_MODE_NOPERSPECTIVE
:
279 assert(!glsl_base_type_is_integer(base_type
));
281 info
->input_interpolate
[i
] = TGSI_INTERPOLATE_LINEAR
;
283 if (variable
->data
.sample
)
284 info
->uses_linear_sample
= true;
285 else if (variable
->data
.centroid
)
286 info
->uses_linear_centroid
= true;
288 info
->uses_linear_center
= true;
291 case INTERP_MODE_FLAT
:
292 info
->input_interpolate
[i
] = TGSI_INTERPOLATE_CONSTANT
;
296 /* TODO make this more precise */
297 if (variable
->data
.location
== VARYING_SLOT_COL0
)
298 info
->colors_read
|= 0x0f;
299 else if (variable
->data
.location
== VARYING_SLOT_COL1
)
300 info
->colors_read
|= 0xf0;
303 if (nir
->info
.stage
!= MESA_SHADER_VERTEX
)
304 info
->num_inputs
= num_inputs
;
306 info
->num_inputs
= nir
->num_inputs
;
309 uint64_t processed_outputs
= 0;
310 unsigned num_outputs
= 0;
311 nir_foreach_variable(variable
, &nir
->outputs
) {
312 unsigned semantic_name
, semantic_index
;
314 if (nir
->info
.stage
== MESA_SHADER_FRAGMENT
) {
315 tgsi_get_gl_frag_result_semantic(variable
->data
.location
,
316 &semantic_name
, &semantic_index
);
318 tgsi_get_gl_varying_semantic(variable
->data
.location
, true,
319 &semantic_name
, &semantic_index
);
322 i
= variable
->data
.driver_location
;
323 if (processed_outputs
& ((uint64_t)1 << i
))
326 processed_outputs
|= ((uint64_t)1 << i
);
329 info
->output_semantic_name
[i
] = semantic_name
;
330 info
->output_semantic_index
[i
] = semantic_index
;
331 info
->output_usagemask
[i
] = TGSI_WRITEMASK_XYZW
;
333 unsigned num_components
= 4;
334 unsigned vector_elements
= glsl_get_vector_elements(glsl_without_array(variable
->type
));
336 num_components
= vector_elements
;
338 unsigned gs_out_streams
;
339 if (variable
->data
.stream
& (1u << 31)) {
340 gs_out_streams
= variable
->data
.stream
& ~(1u << 31);
342 assert(variable
->data
.stream
< 4);
344 for (unsigned j
= 0; j
< num_components
; ++j
)
345 gs_out_streams
|= variable
->data
.stream
<< (2 * (variable
->data
.location_frac
+ j
));
348 unsigned streamx
= gs_out_streams
& 3;
349 unsigned streamy
= (gs_out_streams
>> 2) & 3;
350 unsigned streamz
= (gs_out_streams
>> 4) & 3;
351 unsigned streamw
= (gs_out_streams
>> 6) & 3;
353 if (info
->output_usagemask
[i
] & TGSI_WRITEMASK_X
) {
354 info
->output_streams
[i
] |= streamx
;
355 info
->num_stream_output_components
[streamx
]++;
357 if (info
->output_usagemask
[i
] & TGSI_WRITEMASK_Y
) {
358 info
->output_streams
[i
] |= streamy
<< 2;
359 info
->num_stream_output_components
[streamy
]++;
361 if (info
->output_usagemask
[i
] & TGSI_WRITEMASK_Z
) {
362 info
->output_streams
[i
] |= streamz
<< 4;
363 info
->num_stream_output_components
[streamz
]++;
365 if (info
->output_usagemask
[i
] & TGSI_WRITEMASK_W
) {
366 info
->output_streams
[i
] |= streamw
<< 6;
367 info
->num_stream_output_components
[streamw
]++;
370 switch (semantic_name
) {
371 case TGSI_SEMANTIC_PRIMID
:
372 info
->writes_primid
= true;
374 case TGSI_SEMANTIC_VIEWPORT_INDEX
:
375 info
->writes_viewport_index
= true;
377 case TGSI_SEMANTIC_LAYER
:
378 info
->writes_layer
= true;
380 case TGSI_SEMANTIC_PSIZE
:
381 info
->writes_psize
= true;
383 case TGSI_SEMANTIC_CLIPVERTEX
:
384 info
->writes_clipvertex
= true;
386 case TGSI_SEMANTIC_COLOR
:
387 info
->colors_written
|= 1 << semantic_index
;
389 case TGSI_SEMANTIC_STENCIL
:
390 info
->writes_stencil
= true;
392 case TGSI_SEMANTIC_SAMPLEMASK
:
393 info
->writes_samplemask
= true;
395 case TGSI_SEMANTIC_EDGEFLAG
:
396 info
->writes_edgeflag
= true;
398 case TGSI_SEMANTIC_POSITION
:
399 if (info
->processor
== PIPE_SHADER_FRAGMENT
)
400 info
->writes_z
= true;
402 info
->writes_position
= true;
406 if (nir
->info
.stage
== MESA_SHADER_TESS_CTRL
) {
407 switch (semantic_name
) {
408 case TGSI_SEMANTIC_PATCH
:
409 info
->reads_perpatch_outputs
= true;
411 case TGSI_SEMANTIC_TESSINNER
:
412 case TGSI_SEMANTIC_TESSOUTER
:
413 info
->reads_tessfactor_outputs
= true;
416 info
->reads_pervertex_outputs
= true;
421 info
->num_outputs
= num_outputs
;
423 nir_foreach_variable(variable
, &nir
->uniforms
) {
424 const struct glsl_type
*type
= variable
->type
;
425 enum glsl_base_type base_type
=
426 glsl_get_base_type(glsl_without_array(type
));
427 unsigned aoa_size
= MAX2(1, glsl_get_aoa_size(type
));
429 /* We rely on the fact that nir_lower_samplers_as_deref has
430 * eliminated struct dereferences.
432 if (base_type
== GLSL_TYPE_SAMPLER
)
433 info
->samplers_declared
|=
434 u_bit_consecutive(variable
->data
.binding
, aoa_size
);
435 else if (base_type
== GLSL_TYPE_IMAGE
)
436 info
->images_declared
|=
437 u_bit_consecutive(variable
->data
.binding
, aoa_size
);
440 info
->num_written_clipdistance
= nir
->info
.clip_distance_array_size
;
441 info
->num_written_culldistance
= nir
->info
.cull_distance_array_size
;
442 info
->clipdist_writemask
= u_bit_consecutive(0, info
->num_written_clipdistance
);
443 info
->culldist_writemask
= u_bit_consecutive(0, info
->num_written_culldistance
);
445 if (info
->processor
== PIPE_SHADER_FRAGMENT
)
446 info
->uses_kill
= nir
->info
.fs
.uses_discard
;
448 /* TODO make this more accurate */
449 info
->const_buffers_declared
= u_bit_consecutive(0, SI_NUM_CONST_BUFFERS
);
450 info
->shader_buffers_declared
= u_bit_consecutive(0, SI_NUM_SHADER_BUFFERS
);
452 func
= (struct nir_function
*)exec_list_get_head_const(&nir
->functions
);
453 nir_foreach_block(block
, func
->impl
) {
454 nir_foreach_instr(instr
, block
)
455 scan_instruction(info
, instr
);
460 * Perform "lowering" operations on the NIR that are run once when the shader
461 * selector is created.
464 si_lower_nir(struct si_shader_selector
* sel
)
466 /* Adjust the driver location of inputs and outputs. The state tracker
467 * interprets them as slots, while the ac/nir backend interprets them
468 * as individual components.
470 nir_foreach_variable(variable
, &sel
->nir
->inputs
)
471 variable
->data
.driver_location
*= 4;
473 nir_foreach_variable(variable
, &sel
->nir
->outputs
) {
474 variable
->data
.driver_location
*= 4;
476 if (sel
->nir
->info
.stage
== MESA_SHADER_FRAGMENT
) {
477 if (variable
->data
.location
== FRAG_RESULT_DEPTH
)
478 variable
->data
.driver_location
+= 2;
479 else if (variable
->data
.location
== FRAG_RESULT_STENCIL
)
480 variable
->data
.driver_location
+= 1;
484 /* Perform lowerings (and optimizations) of code.
486 * Performance considerations aside, we must:
487 * - lower certain ALU operations
488 * - ensure constant offsets for texture instructions are folded
489 * and copy-propagated
491 NIR_PASS_V(sel
->nir
, nir_lower_io
, nir_var_uniform
, type_size
,
492 (nir_lower_io_options
)0);
493 NIR_PASS_V(sel
->nir
, nir_lower_uniforms_to_ubo
);
495 NIR_PASS_V(sel
->nir
, nir_lower_returns
);
496 NIR_PASS_V(sel
->nir
, nir_lower_vars_to_ssa
);
497 NIR_PASS_V(sel
->nir
, nir_lower_alu_to_scalar
);
498 NIR_PASS_V(sel
->nir
, nir_lower_phis_to_scalar
);
500 static const struct nir_lower_tex_options lower_tex_options
= {
503 NIR_PASS_V(sel
->nir
, nir_lower_tex
, &lower_tex_options
);
509 /* (Constant) copy propagation is needed for txf with offsets. */
510 NIR_PASS(progress
, sel
->nir
, nir_copy_prop
);
511 NIR_PASS(progress
, sel
->nir
, nir_opt_remove_phis
);
512 NIR_PASS(progress
, sel
->nir
, nir_opt_dce
);
513 if (nir_opt_trivial_continues(sel
->nir
)) {
515 NIR_PASS(progress
, sel
->nir
, nir_copy_prop
);
516 NIR_PASS(progress
, sel
->nir
, nir_opt_dce
);
518 NIR_PASS(progress
, sel
->nir
, nir_opt_if
);
519 NIR_PASS(progress
, sel
->nir
, nir_opt_dead_cf
);
520 NIR_PASS(progress
, sel
->nir
, nir_opt_cse
);
521 NIR_PASS(progress
, sel
->nir
, nir_opt_peephole_select
, 8);
523 /* Needed for algebraic lowering */
524 NIR_PASS(progress
, sel
->nir
, nir_opt_algebraic
);
525 NIR_PASS(progress
, sel
->nir
, nir_opt_constant_folding
);
527 NIR_PASS(progress
, sel
->nir
, nir_opt_undef
);
528 NIR_PASS(progress
, sel
->nir
, nir_opt_conditional_discard
);
529 if (sel
->nir
->options
->max_unroll_iterations
) {
530 NIR_PASS(progress
, sel
->nir
, nir_opt_loop_unroll
, 0);
535 static void declare_nir_input_vs(struct si_shader_context
*ctx
,
536 struct nir_variable
*variable
,
539 si_llvm_load_input_vs(ctx
, variable
->data
.driver_location
/ 4, out
);
542 static void declare_nir_input_fs(struct si_shader_context
*ctx
,
543 struct nir_variable
*variable
,
544 unsigned input_index
,
547 unsigned slot
= variable
->data
.location
;
548 if (slot
== VARYING_SLOT_POS
) {
549 out
[0] = LLVMGetParam(ctx
->main_fn
, SI_PARAM_POS_X_FLOAT
);
550 out
[1] = LLVMGetParam(ctx
->main_fn
, SI_PARAM_POS_Y_FLOAT
);
551 out
[2] = LLVMGetParam(ctx
->main_fn
, SI_PARAM_POS_Z_FLOAT
);
552 out
[3] = ac_build_fdiv(&ctx
->ac
, ctx
->ac
.f32_1
,
553 LLVMGetParam(ctx
->main_fn
, SI_PARAM_POS_W_FLOAT
));
557 si_llvm_load_input_fs(ctx
, input_index
, out
);
560 LLVMValueRef
si_nir_load_input_gs(struct ac_shader_abi
*abi
,
562 unsigned driver_location
,
564 unsigned num_components
,
565 unsigned vertex_index
,
566 unsigned const_index
,
569 struct si_shader_context
*ctx
= si_shader_context_from_abi(abi
);
571 LLVMValueRef value
[4];
572 for (unsigned i
= component
; i
< num_components
+ component
; i
++) {
573 value
[i
] = si_llvm_load_input_gs(&ctx
->abi
, driver_location
/ 4,
574 vertex_index
, type
, i
);
577 return ac_build_varying_gather_values(&ctx
->ac
, value
, num_components
, component
);
581 si_nir_load_sampler_desc(struct ac_shader_abi
*abi
,
582 unsigned descriptor_set
, unsigned base_index
,
583 unsigned constant_index
, LLVMValueRef dynamic_index
,
584 enum ac_descriptor_type desc_type
, bool image
,
587 struct si_shader_context
*ctx
= si_shader_context_from_abi(abi
);
588 LLVMBuilderRef builder
= ctx
->ac
.builder
;
589 LLVMValueRef list
= LLVMGetParam(ctx
->main_fn
, ctx
->param_samplers_and_images
);
590 LLVMValueRef index
= dynamic_index
;
592 assert(!descriptor_set
);
595 index
= ctx
->ac
.i32_0
;
597 index
= LLVMBuildAdd(builder
, index
,
598 LLVMConstInt(ctx
->ac
.i32
, base_index
+ constant_index
, false),
602 assert(desc_type
== AC_DESC_IMAGE
|| desc_type
== AC_DESC_BUFFER
);
603 assert(base_index
+ constant_index
< ctx
->num_images
);
606 index
= si_llvm_bound_index(ctx
, index
, ctx
->num_images
);
608 index
= LLVMBuildSub(ctx
->gallivm
.builder
,
609 LLVMConstInt(ctx
->i32
, SI_NUM_IMAGES
- 1, 0),
612 /* TODO: be smarter about when we use dcc_off */
613 return si_load_image_desc(ctx
, list
, index
, desc_type
, write
);
616 assert(base_index
+ constant_index
< ctx
->num_samplers
);
619 index
= si_llvm_bound_index(ctx
, index
, ctx
->num_samplers
);
621 index
= LLVMBuildAdd(ctx
->gallivm
.builder
, index
,
622 LLVMConstInt(ctx
->i32
, SI_NUM_IMAGES
/ 2, 0), "");
624 return si_load_sampler_desc(ctx
, list
, index
, desc_type
);
627 bool si_nir_build_llvm(struct si_shader_context
*ctx
, struct nir_shader
*nir
)
629 struct tgsi_shader_info
*info
= &ctx
->shader
->selector
->info
;
631 if (nir
->info
.stage
== MESA_SHADER_VERTEX
||
632 nir
->info
.stage
== MESA_SHADER_FRAGMENT
) {
633 uint64_t processed_inputs
= 0;
634 nir_foreach_variable(variable
, &nir
->inputs
) {
635 unsigned attrib_count
= glsl_count_attribute_slots(variable
->type
,
636 nir
->info
.stage
== MESA_SHADER_VERTEX
);
637 unsigned input_idx
= variable
->data
.driver_location
;
639 assert(attrib_count
== 1);
641 LLVMValueRef data
[4];
642 unsigned loc
= variable
->data
.location
;
644 /* Packed components share the same location so skip
645 * them if we have already processed the location.
647 if (processed_inputs
& ((uint64_t)1 << loc
))
650 if (nir
->info
.stage
== MESA_SHADER_VERTEX
)
651 declare_nir_input_vs(ctx
, variable
, data
);
652 else if (nir
->info
.stage
== MESA_SHADER_FRAGMENT
)
653 declare_nir_input_fs(ctx
, variable
, input_idx
/ 4, data
);
655 for (unsigned chan
= 0; chan
< 4; chan
++) {
656 ctx
->inputs
[input_idx
+ chan
] =
657 LLVMBuildBitCast(ctx
->ac
.builder
, data
[chan
], ctx
->ac
.i32
, "");
659 processed_inputs
|= ((uint64_t)1 << loc
);
663 ctx
->abi
.inputs
= &ctx
->inputs
[0];
664 ctx
->abi
.load_sampler_desc
= si_nir_load_sampler_desc
;
665 ctx
->abi
.clamp_shadow_reference
= true;
667 ctx
->num_samplers
= util_last_bit(info
->samplers_declared
);
668 ctx
->num_images
= util_last_bit(info
->images_declared
);
670 ac_nir_translate(&ctx
->ac
, &ctx
->abi
, nir
, NULL
);