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_tess_level_inner
:
102 case nir_intrinsic_load_tess_level_outer
:
103 info
->reads_tess_factors
= true;
105 case nir_intrinsic_image_store
:
106 case nir_intrinsic_image_atomic_add
:
107 case nir_intrinsic_image_atomic_min
:
108 case nir_intrinsic_image_atomic_max
:
109 case nir_intrinsic_image_atomic_and
:
110 case nir_intrinsic_image_atomic_or
:
111 case nir_intrinsic_image_atomic_xor
:
112 case nir_intrinsic_image_atomic_exchange
:
113 case nir_intrinsic_image_atomic_comp_swap
:
114 case nir_intrinsic_store_ssbo
:
115 case nir_intrinsic_ssbo_atomic_add
:
116 case nir_intrinsic_ssbo_atomic_imin
:
117 case nir_intrinsic_ssbo_atomic_umin
:
118 case nir_intrinsic_ssbo_atomic_imax
:
119 case nir_intrinsic_ssbo_atomic_umax
:
120 case nir_intrinsic_ssbo_atomic_and
:
121 case nir_intrinsic_ssbo_atomic_or
:
122 case nir_intrinsic_ssbo_atomic_xor
:
123 case nir_intrinsic_ssbo_atomic_exchange
:
124 case nir_intrinsic_ssbo_atomic_comp_swap
:
125 info
->writes_memory
= true;
133 void si_nir_scan_shader(const struct nir_shader
*nir
,
134 struct tgsi_shader_info
*info
)
139 assert(nir
->info
.stage
== MESA_SHADER_VERTEX
||
140 nir
->info
.stage
== MESA_SHADER_GEOMETRY
||
141 nir
->info
.stage
== MESA_SHADER_FRAGMENT
);
143 info
->processor
= pipe_shader_type_from_mesa(nir
->info
.stage
);
144 info
->num_tokens
= 2; /* indicate that the shader is non-empty */
145 info
->num_instructions
= 2;
147 if (nir
->info
.stage
== MESA_SHADER_TESS_CTRL
) {
148 info
->properties
[TGSI_PROPERTY_TCS_VERTICES_OUT
] =
149 nir
->info
.tess
.tcs_vertices_out
;
152 if (nir
->info
.stage
== MESA_SHADER_TESS_EVAL
) {
153 if (nir
->info
.tess
.primitive_mode
== GL_ISOLINES
)
154 info
->properties
[TGSI_PROPERTY_TES_PRIM_MODE
] = PIPE_PRIM_LINES
;
156 info
->properties
[TGSI_PROPERTY_TES_PRIM_MODE
] = nir
->info
.tess
.primitive_mode
;
158 STATIC_ASSERT((TESS_SPACING_EQUAL
+ 1) % 3 == PIPE_TESS_SPACING_EQUAL
);
159 STATIC_ASSERT((TESS_SPACING_FRACTIONAL_ODD
+ 1) % 3 ==
160 PIPE_TESS_SPACING_FRACTIONAL_ODD
);
161 STATIC_ASSERT((TESS_SPACING_FRACTIONAL_EVEN
+ 1) % 3 ==
162 PIPE_TESS_SPACING_FRACTIONAL_EVEN
);
164 info
->properties
[TGSI_PROPERTY_TES_SPACING
] = (nir
->info
.tess
.spacing
+ 1) % 3;
165 info
->properties
[TGSI_PROPERTY_TES_VERTEX_ORDER_CW
] = !nir
->info
.tess
.ccw
;
166 info
->properties
[TGSI_PROPERTY_TES_POINT_MODE
] = nir
->info
.tess
.point_mode
;
169 if (nir
->info
.stage
== MESA_SHADER_GEOMETRY
) {
170 info
->properties
[TGSI_PROPERTY_GS_INPUT_PRIM
] = nir
->info
.gs
.input_primitive
;
171 info
->properties
[TGSI_PROPERTY_GS_OUTPUT_PRIM
] = nir
->info
.gs
.output_primitive
;
172 info
->properties
[TGSI_PROPERTY_GS_MAX_OUTPUT_VERTICES
] = nir
->info
.gs
.vertices_out
;
173 info
->properties
[TGSI_PROPERTY_GS_INVOCATIONS
] = nir
->info
.gs
.invocations
;
177 uint64_t processed_inputs
= 0;
178 unsigned num_inputs
= 0;
179 nir_foreach_variable(variable
, &nir
->inputs
) {
180 unsigned semantic_name
, semantic_index
;
181 unsigned attrib_count
= glsl_count_attribute_slots(variable
->type
,
182 nir
->info
.stage
== MESA_SHADER_VERTEX
);
184 /* Vertex shader inputs don't have semantics. The state
185 * tracker has already mapped them to attributes via
186 * variable->data.driver_location.
188 if (nir
->info
.stage
== MESA_SHADER_VERTEX
)
191 assert(nir
->info
.stage
!= MESA_SHADER_FRAGMENT
||
192 (attrib_count
== 1 && "not implemented"));
194 /* Fragment shader position is a system value. */
195 if (nir
->info
.stage
== MESA_SHADER_FRAGMENT
&&
196 variable
->data
.location
== VARYING_SLOT_POS
) {
197 if (variable
->data
.pixel_center_integer
)
198 info
->properties
[TGSI_PROPERTY_FS_COORD_PIXEL_CENTER
] =
199 TGSI_FS_COORD_PIXEL_CENTER_INTEGER
;
205 i
= variable
->data
.driver_location
;
206 if (processed_inputs
& ((uint64_t)1 << i
))
209 processed_inputs
|= ((uint64_t)1 << i
);
212 tgsi_get_gl_varying_semantic(variable
->data
.location
, true,
213 &semantic_name
, &semantic_index
);
215 info
->input_semantic_name
[i
] = semantic_name
;
216 info
->input_semantic_index
[i
] = semantic_index
;
218 if (variable
->data
.sample
)
219 info
->input_interpolate_loc
[i
] = TGSI_INTERPOLATE_LOC_SAMPLE
;
220 else if (variable
->data
.centroid
)
221 info
->input_interpolate_loc
[i
] = TGSI_INTERPOLATE_LOC_CENTROID
;
223 info
->input_interpolate_loc
[i
] = TGSI_INTERPOLATE_LOC_CENTER
;
225 enum glsl_base_type base_type
=
226 glsl_get_base_type(glsl_without_array(variable
->type
));
228 switch (variable
->data
.interpolation
) {
229 case INTERP_MODE_NONE
:
230 if (glsl_base_type_is_integer(base_type
)) {
231 info
->input_interpolate
[i
] = TGSI_INTERPOLATE_CONSTANT
;
235 if (semantic_name
== TGSI_SEMANTIC_COLOR
) {
236 info
->input_interpolate
[i
] = TGSI_INTERPOLATE_COLOR
;
237 goto persp_locations
;
240 case INTERP_MODE_SMOOTH
:
241 assert(!glsl_base_type_is_integer(base_type
));
243 info
->input_interpolate
[i
] = TGSI_INTERPOLATE_PERSPECTIVE
;
246 if (variable
->data
.sample
)
247 info
->uses_persp_sample
= true;
248 else if (variable
->data
.centroid
)
249 info
->uses_persp_centroid
= true;
251 info
->uses_persp_center
= true;
254 case INTERP_MODE_NOPERSPECTIVE
:
255 assert(!glsl_base_type_is_integer(base_type
));
257 info
->input_interpolate
[i
] = TGSI_INTERPOLATE_LINEAR
;
259 if (variable
->data
.sample
)
260 info
->uses_linear_sample
= true;
261 else if (variable
->data
.centroid
)
262 info
->uses_linear_centroid
= true;
264 info
->uses_linear_center
= true;
267 case INTERP_MODE_FLAT
:
268 info
->input_interpolate
[i
] = TGSI_INTERPOLATE_CONSTANT
;
272 /* TODO make this more precise */
273 if (variable
->data
.location
== VARYING_SLOT_COL0
)
274 info
->colors_read
|= 0x0f;
275 else if (variable
->data
.location
== VARYING_SLOT_COL1
)
276 info
->colors_read
|= 0xf0;
279 if (nir
->info
.stage
!= MESA_SHADER_VERTEX
)
280 info
->num_inputs
= num_inputs
;
282 info
->num_inputs
= nir
->num_inputs
;
285 uint64_t processed_outputs
= 0;
286 unsigned num_outputs
= 0;
287 nir_foreach_variable(variable
, &nir
->outputs
) {
288 unsigned semantic_name
, semantic_index
;
290 if (nir
->info
.stage
== MESA_SHADER_FRAGMENT
) {
291 tgsi_get_gl_frag_result_semantic(variable
->data
.location
,
292 &semantic_name
, &semantic_index
);
294 tgsi_get_gl_varying_semantic(variable
->data
.location
, true,
295 &semantic_name
, &semantic_index
);
298 i
= variable
->data
.driver_location
;
299 if (processed_outputs
& ((uint64_t)1 << i
))
302 processed_outputs
|= ((uint64_t)1 << i
);
305 info
->output_semantic_name
[i
] = semantic_name
;
306 info
->output_semantic_index
[i
] = semantic_index
;
307 info
->output_usagemask
[i
] = TGSI_WRITEMASK_XYZW
;
309 unsigned num_components
= 4;
310 unsigned vector_elements
= glsl_get_vector_elements(glsl_without_array(variable
->type
));
312 num_components
= vector_elements
;
314 unsigned gs_out_streams
;
315 if (variable
->data
.stream
& (1u << 31)) {
316 gs_out_streams
= variable
->data
.stream
& ~(1u << 31);
318 assert(variable
->data
.stream
< 4);
320 for (unsigned j
= 0; j
< num_components
; ++j
)
321 gs_out_streams
|= variable
->data
.stream
<< (2 * (variable
->data
.location_frac
+ j
));
324 unsigned streamx
= gs_out_streams
& 3;
325 unsigned streamy
= (gs_out_streams
>> 2) & 3;
326 unsigned streamz
= (gs_out_streams
>> 4) & 3;
327 unsigned streamw
= (gs_out_streams
>> 6) & 3;
329 if (info
->output_usagemask
[i
] & TGSI_WRITEMASK_X
) {
330 info
->output_streams
[i
] |= streamx
;
331 info
->num_stream_output_components
[streamx
]++;
333 if (info
->output_usagemask
[i
] & TGSI_WRITEMASK_Y
) {
334 info
->output_streams
[i
] |= streamy
<< 2;
335 info
->num_stream_output_components
[streamy
]++;
337 if (info
->output_usagemask
[i
] & TGSI_WRITEMASK_Z
) {
338 info
->output_streams
[i
] |= streamz
<< 4;
339 info
->num_stream_output_components
[streamz
]++;
341 if (info
->output_usagemask
[i
] & TGSI_WRITEMASK_W
) {
342 info
->output_streams
[i
] |= streamw
<< 6;
343 info
->num_stream_output_components
[streamw
]++;
346 switch (semantic_name
) {
347 case TGSI_SEMANTIC_PRIMID
:
348 info
->writes_primid
= true;
350 case TGSI_SEMANTIC_VIEWPORT_INDEX
:
351 info
->writes_viewport_index
= true;
353 case TGSI_SEMANTIC_LAYER
:
354 info
->writes_layer
= true;
356 case TGSI_SEMANTIC_PSIZE
:
357 info
->writes_psize
= true;
359 case TGSI_SEMANTIC_CLIPVERTEX
:
360 info
->writes_clipvertex
= true;
362 case TGSI_SEMANTIC_COLOR
:
363 info
->colors_written
|= 1 << semantic_index
;
365 case TGSI_SEMANTIC_STENCIL
:
366 info
->writes_stencil
= true;
368 case TGSI_SEMANTIC_SAMPLEMASK
:
369 info
->writes_samplemask
= true;
371 case TGSI_SEMANTIC_EDGEFLAG
:
372 info
->writes_edgeflag
= true;
374 case TGSI_SEMANTIC_POSITION
:
375 if (info
->processor
== PIPE_SHADER_FRAGMENT
)
376 info
->writes_z
= true;
378 info
->writes_position
= true;
383 info
->num_outputs
= num_outputs
;
385 nir_foreach_variable(variable
, &nir
->uniforms
) {
386 const struct glsl_type
*type
= variable
->type
;
387 enum glsl_base_type base_type
=
388 glsl_get_base_type(glsl_without_array(type
));
389 unsigned aoa_size
= MAX2(1, glsl_get_aoa_size(type
));
391 /* We rely on the fact that nir_lower_samplers_as_deref has
392 * eliminated struct dereferences.
394 if (base_type
== GLSL_TYPE_SAMPLER
)
395 info
->samplers_declared
|=
396 u_bit_consecutive(variable
->data
.binding
, aoa_size
);
397 else if (base_type
== GLSL_TYPE_IMAGE
)
398 info
->images_declared
|=
399 u_bit_consecutive(variable
->data
.binding
, aoa_size
);
402 info
->num_written_clipdistance
= nir
->info
.clip_distance_array_size
;
403 info
->num_written_culldistance
= nir
->info
.cull_distance_array_size
;
404 info
->clipdist_writemask
= u_bit_consecutive(0, info
->num_written_clipdistance
);
405 info
->culldist_writemask
= u_bit_consecutive(0, info
->num_written_culldistance
);
407 if (info
->processor
== PIPE_SHADER_FRAGMENT
)
408 info
->uses_kill
= nir
->info
.fs
.uses_discard
;
410 /* TODO make this more accurate */
411 info
->const_buffers_declared
= u_bit_consecutive(0, SI_NUM_CONST_BUFFERS
);
412 info
->shader_buffers_declared
= u_bit_consecutive(0, SI_NUM_SHADER_BUFFERS
);
414 func
= (struct nir_function
*)exec_list_get_head_const(&nir
->functions
);
415 nir_foreach_block(block
, func
->impl
) {
416 nir_foreach_instr(instr
, block
)
417 scan_instruction(info
, instr
);
422 * Perform "lowering" operations on the NIR that are run once when the shader
423 * selector is created.
426 si_lower_nir(struct si_shader_selector
* sel
)
428 /* Adjust the driver location of inputs and outputs. The state tracker
429 * interprets them as slots, while the ac/nir backend interprets them
430 * as individual components.
432 nir_foreach_variable(variable
, &sel
->nir
->inputs
)
433 variable
->data
.driver_location
*= 4;
435 nir_foreach_variable(variable
, &sel
->nir
->outputs
) {
436 variable
->data
.driver_location
*= 4;
438 if (sel
->nir
->info
.stage
== MESA_SHADER_FRAGMENT
) {
439 if (variable
->data
.location
== FRAG_RESULT_DEPTH
)
440 variable
->data
.driver_location
+= 2;
441 else if (variable
->data
.location
== FRAG_RESULT_STENCIL
)
442 variable
->data
.driver_location
+= 1;
446 /* Perform lowerings (and optimizations) of code.
448 * Performance considerations aside, we must:
449 * - lower certain ALU operations
450 * - ensure constant offsets for texture instructions are folded
451 * and copy-propagated
453 NIR_PASS_V(sel
->nir
, nir_lower_io
, nir_var_uniform
, type_size
,
454 (nir_lower_io_options
)0);
455 NIR_PASS_V(sel
->nir
, nir_lower_uniforms_to_ubo
);
457 NIR_PASS_V(sel
->nir
, nir_lower_returns
);
458 NIR_PASS_V(sel
->nir
, nir_lower_vars_to_ssa
);
459 NIR_PASS_V(sel
->nir
, nir_lower_alu_to_scalar
);
460 NIR_PASS_V(sel
->nir
, nir_lower_phis_to_scalar
);
462 static const struct nir_lower_tex_options lower_tex_options
= {
465 NIR_PASS_V(sel
->nir
, nir_lower_tex
, &lower_tex_options
);
471 /* (Constant) copy propagation is needed for txf with offsets. */
472 NIR_PASS(progress
, sel
->nir
, nir_copy_prop
);
473 NIR_PASS(progress
, sel
->nir
, nir_opt_remove_phis
);
474 NIR_PASS(progress
, sel
->nir
, nir_opt_dce
);
475 if (nir_opt_trivial_continues(sel
->nir
)) {
477 NIR_PASS(progress
, sel
->nir
, nir_copy_prop
);
478 NIR_PASS(progress
, sel
->nir
, nir_opt_dce
);
480 NIR_PASS(progress
, sel
->nir
, nir_opt_if
);
481 NIR_PASS(progress
, sel
->nir
, nir_opt_dead_cf
);
482 NIR_PASS(progress
, sel
->nir
, nir_opt_cse
);
483 NIR_PASS(progress
, sel
->nir
, nir_opt_peephole_select
, 8);
485 /* Needed for algebraic lowering */
486 NIR_PASS(progress
, sel
->nir
, nir_opt_algebraic
);
487 NIR_PASS(progress
, sel
->nir
, nir_opt_constant_folding
);
489 NIR_PASS(progress
, sel
->nir
, nir_opt_undef
);
490 NIR_PASS(progress
, sel
->nir
, nir_opt_conditional_discard
);
491 if (sel
->nir
->options
->max_unroll_iterations
) {
492 NIR_PASS(progress
, sel
->nir
, nir_opt_loop_unroll
, 0);
497 static void declare_nir_input_vs(struct si_shader_context
*ctx
,
498 struct nir_variable
*variable
,
501 si_llvm_load_input_vs(ctx
, variable
->data
.driver_location
/ 4, out
);
504 static void declare_nir_input_fs(struct si_shader_context
*ctx
,
505 struct nir_variable
*variable
,
506 unsigned input_index
,
509 unsigned slot
= variable
->data
.location
;
510 if (slot
== VARYING_SLOT_POS
) {
511 out
[0] = LLVMGetParam(ctx
->main_fn
, SI_PARAM_POS_X_FLOAT
);
512 out
[1] = LLVMGetParam(ctx
->main_fn
, SI_PARAM_POS_Y_FLOAT
);
513 out
[2] = LLVMGetParam(ctx
->main_fn
, SI_PARAM_POS_Z_FLOAT
);
514 out
[3] = ac_build_fdiv(&ctx
->ac
, ctx
->ac
.f32_1
,
515 LLVMGetParam(ctx
->main_fn
, SI_PARAM_POS_W_FLOAT
));
519 si_llvm_load_input_fs(ctx
, input_index
, out
);
522 LLVMValueRef
si_nir_load_input_gs(struct ac_shader_abi
*abi
,
524 unsigned driver_location
,
526 unsigned num_components
,
527 unsigned vertex_index
,
528 unsigned const_index
,
531 struct si_shader_context
*ctx
= si_shader_context_from_abi(abi
);
533 LLVMValueRef value
[4];
534 for (unsigned i
= component
; i
< num_components
+ component
; i
++) {
535 value
[i
] = si_llvm_load_input_gs(&ctx
->abi
, driver_location
/ 4,
536 vertex_index
, type
, i
);
539 return ac_build_varying_gather_values(&ctx
->ac
, value
, num_components
, component
);
543 si_nir_load_sampler_desc(struct ac_shader_abi
*abi
,
544 unsigned descriptor_set
, unsigned base_index
,
545 unsigned constant_index
, LLVMValueRef dynamic_index
,
546 enum ac_descriptor_type desc_type
, bool image
,
549 struct si_shader_context
*ctx
= si_shader_context_from_abi(abi
);
550 LLVMBuilderRef builder
= ctx
->ac
.builder
;
551 LLVMValueRef list
= LLVMGetParam(ctx
->main_fn
, ctx
->param_samplers_and_images
);
552 LLVMValueRef index
= dynamic_index
;
554 assert(!descriptor_set
);
557 index
= ctx
->ac
.i32_0
;
559 index
= LLVMBuildAdd(builder
, index
,
560 LLVMConstInt(ctx
->ac
.i32
, base_index
+ constant_index
, false),
564 assert(desc_type
== AC_DESC_IMAGE
|| desc_type
== AC_DESC_BUFFER
);
565 assert(base_index
+ constant_index
< ctx
->num_images
);
568 index
= si_llvm_bound_index(ctx
, index
, ctx
->num_images
);
570 index
= LLVMBuildSub(ctx
->gallivm
.builder
,
571 LLVMConstInt(ctx
->i32
, SI_NUM_IMAGES
- 1, 0),
574 /* TODO: be smarter about when we use dcc_off */
575 return si_load_image_desc(ctx
, list
, index
, desc_type
, write
);
578 assert(base_index
+ constant_index
< ctx
->num_samplers
);
581 index
= si_llvm_bound_index(ctx
, index
, ctx
->num_samplers
);
583 index
= LLVMBuildAdd(ctx
->gallivm
.builder
, index
,
584 LLVMConstInt(ctx
->i32
, SI_NUM_IMAGES
/ 2, 0), "");
586 return si_load_sampler_desc(ctx
, list
, index
, desc_type
);
589 bool si_nir_build_llvm(struct si_shader_context
*ctx
, struct nir_shader
*nir
)
591 struct tgsi_shader_info
*info
= &ctx
->shader
->selector
->info
;
593 if (nir
->info
.stage
== MESA_SHADER_VERTEX
||
594 nir
->info
.stage
== MESA_SHADER_FRAGMENT
) {
595 uint64_t processed_inputs
= 0;
596 nir_foreach_variable(variable
, &nir
->inputs
) {
597 unsigned attrib_count
= glsl_count_attribute_slots(variable
->type
,
598 nir
->info
.stage
== MESA_SHADER_VERTEX
);
599 unsigned input_idx
= variable
->data
.driver_location
;
601 assert(attrib_count
== 1);
603 LLVMValueRef data
[4];
604 unsigned loc
= variable
->data
.location
;
606 /* Packed components share the same location so skip
607 * them if we have already processed the location.
609 if (processed_inputs
& ((uint64_t)1 << loc
))
612 if (nir
->info
.stage
== MESA_SHADER_VERTEX
)
613 declare_nir_input_vs(ctx
, variable
, data
);
614 else if (nir
->info
.stage
== MESA_SHADER_FRAGMENT
)
615 declare_nir_input_fs(ctx
, variable
, input_idx
/ 4, data
);
617 for (unsigned chan
= 0; chan
< 4; chan
++) {
618 ctx
->inputs
[input_idx
+ chan
] =
619 LLVMBuildBitCast(ctx
->ac
.builder
, data
[chan
], ctx
->ac
.i32
, "");
621 processed_inputs
|= ((uint64_t)1 << loc
);
625 ctx
->abi
.inputs
= &ctx
->inputs
[0];
626 ctx
->abi
.load_sampler_desc
= si_nir_load_sampler_desc
;
627 ctx
->abi
.clamp_shadow_reference
= true;
629 ctx
->num_samplers
= util_last_bit(info
->samplers_declared
);
630 ctx
->num_images
= util_last_bit(info
->images_declared
);
632 ac_nir_translate(&ctx
->ac
, &ctx
->abi
, nir
, NULL
);