2 * Copyright © 2014 Intel Corporation
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:
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 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
20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
25 #include "brw_shader.h"
26 #include "common/gen_debug.h"
27 #include "compiler/glsl_types.h"
28 #include "compiler/nir/nir_builder.h"
31 is_input(nir_intrinsic_instr
*intrin
)
33 return intrin
->intrinsic
== nir_intrinsic_load_input
||
34 intrin
->intrinsic
== nir_intrinsic_load_per_vertex_input
||
35 intrin
->intrinsic
== nir_intrinsic_load_interpolated_input
;
39 is_output(nir_intrinsic_instr
*intrin
)
41 return intrin
->intrinsic
== nir_intrinsic_load_output
||
42 intrin
->intrinsic
== nir_intrinsic_load_per_vertex_output
||
43 intrin
->intrinsic
== nir_intrinsic_store_output
||
44 intrin
->intrinsic
== nir_intrinsic_store_per_vertex_output
;
48 * In many cases, we just add the base and offset together, so there's no
49 * reason to keep them separate. Sometimes, combining them is essential:
50 * if a shader only accesses part of a compound variable (such as a matrix
51 * or array), the variable's base may not actually exist in the VUE map.
53 * This pass adds constant offsets to instr->const_index[0], and resets
54 * the offset source to 0. Non-constant offsets remain unchanged - since
55 * we don't know what part of a compound variable is accessed, we allocate
56 * storage for the entire thing.
60 add_const_offset_to_base_block(nir_block
*block
, nir_builder
*b
,
61 nir_variable_mode mode
)
63 nir_foreach_instr_safe(instr
, block
) {
64 if (instr
->type
!= nir_instr_type_intrinsic
)
67 nir_intrinsic_instr
*intrin
= nir_instr_as_intrinsic(instr
);
69 if ((mode
== nir_var_shader_in
&& is_input(intrin
)) ||
70 (mode
== nir_var_shader_out
&& is_output(intrin
))) {
71 nir_src
*offset
= nir_get_io_offset_src(intrin
);
72 nir_const_value
*const_offset
= nir_src_as_const_value(*offset
);
75 intrin
->const_index
[0] += const_offset
->u32
[0];
76 b
->cursor
= nir_before_instr(&intrin
->instr
);
77 nir_instr_rewrite_src(&intrin
->instr
, offset
,
78 nir_src_for_ssa(nir_imm_int(b
, 0)));
86 add_const_offset_to_base(nir_shader
*nir
, nir_variable_mode mode
)
88 nir_foreach_function(f
, nir
) {
91 nir_builder_init(&b
, f
->impl
);
92 nir_foreach_block(block
, f
->impl
) {
93 add_const_offset_to_base_block(block
, &b
, mode
);
100 remap_vs_attrs(nir_block
*block
, shader_info
*nir_info
)
102 nir_foreach_instr(instr
, block
) {
103 if (instr
->type
!= nir_instr_type_intrinsic
)
106 nir_intrinsic_instr
*intrin
= nir_instr_as_intrinsic(instr
);
108 if (intrin
->intrinsic
== nir_intrinsic_load_input
) {
109 /* Attributes come in a contiguous block, ordered by their
110 * gl_vert_attrib value. That means we can compute the slot
111 * number for an attribute by masking out the enabled attributes
112 * before it and counting the bits.
114 int attr
= intrin
->const_index
[0];
115 int slot
= _mesa_bitcount_64(nir_info
->inputs_read
&
116 BITFIELD64_MASK(attr
));
117 intrin
->const_index
[0] = 4 * slot
;
124 remap_inputs_with_vue_map(nir_block
*block
, const struct brw_vue_map
*vue_map
)
126 nir_foreach_instr(instr
, block
) {
127 if (instr
->type
!= nir_instr_type_intrinsic
)
130 nir_intrinsic_instr
*intrin
= nir_instr_as_intrinsic(instr
);
132 if (intrin
->intrinsic
== nir_intrinsic_load_input
||
133 intrin
->intrinsic
== nir_intrinsic_load_per_vertex_input
) {
134 int vue_slot
= vue_map
->varying_to_slot
[intrin
->const_index
[0]];
135 assert(vue_slot
!= -1);
136 intrin
->const_index
[0] = vue_slot
;
143 remap_tess_levels(nir_builder
*b
, nir_intrinsic_instr
*intr
,
144 GLenum primitive_mode
)
146 const int location
= nir_intrinsic_base(intr
);
147 const unsigned component
= nir_intrinsic_component(intr
);
150 if (location
== VARYING_SLOT_TESS_LEVEL_INNER
) {
151 switch (primitive_mode
) {
153 /* gl_TessLevelInner[0..1] lives at DWords 3-2 (reversed). */
154 nir_intrinsic_set_base(intr
, 0);
155 nir_intrinsic_set_component(intr
, 3 - component
);
156 out_of_bounds
= false;
159 /* gl_TessLevelInner[0] lives at DWord 4. */
160 nir_intrinsic_set_base(intr
, 1);
161 out_of_bounds
= component
> 0;
164 out_of_bounds
= true;
167 unreachable("Bogus tessellation domain");
169 } else if (location
== VARYING_SLOT_TESS_LEVEL_OUTER
) {
170 if (primitive_mode
== GL_ISOLINES
) {
171 /* gl_TessLevelOuter[0..1] lives at DWords 6-7 (in order). */
172 nir_intrinsic_set_base(intr
, 1);
173 nir_intrinsic_set_component(intr
, 2 + nir_intrinsic_component(intr
));
174 out_of_bounds
= component
> 1;
176 /* Triangles use DWords 7-5 (reversed); Quads use 7-4 (reversed) */
177 nir_intrinsic_set_base(intr
, 1);
178 nir_intrinsic_set_component(intr
, 3 - nir_intrinsic_component(intr
));
179 out_of_bounds
= component
== 3 && primitive_mode
== GL_TRIANGLES
;
186 if (nir_intrinsic_infos
[intr
->intrinsic
].has_dest
) {
187 b
->cursor
= nir_before_instr(&intr
->instr
);
188 nir_ssa_def
*undef
= nir_ssa_undef(b
, 1, 32);
189 nir_ssa_def_rewrite_uses(&intr
->dest
.ssa
, nir_src_for_ssa(undef
));
191 nir_instr_remove(&intr
->instr
);
198 remap_patch_urb_offsets(nir_block
*block
, nir_builder
*b
,
199 const struct brw_vue_map
*vue_map
,
200 GLenum tes_primitive_mode
)
202 const bool is_passthrough_tcs
= b
->shader
->info
->name
&&
203 strcmp(b
->shader
->info
->name
, "passthrough") == 0;
205 nir_foreach_instr_safe(instr
, block
) {
206 if (instr
->type
!= nir_instr_type_intrinsic
)
209 nir_intrinsic_instr
*intrin
= nir_instr_as_intrinsic(instr
);
211 gl_shader_stage stage
= b
->shader
->stage
;
213 if ((stage
== MESA_SHADER_TESS_CTRL
&& is_output(intrin
)) ||
214 (stage
== MESA_SHADER_TESS_EVAL
&& is_input(intrin
))) {
216 if (!is_passthrough_tcs
&&
217 remap_tess_levels(b
, intrin
, tes_primitive_mode
))
220 int vue_slot
= vue_map
->varying_to_slot
[intrin
->const_index
[0]];
221 assert(vue_slot
!= -1);
222 intrin
->const_index
[0] = vue_slot
;
224 nir_src
*vertex
= nir_get_io_vertex_index_src(intrin
);
226 nir_const_value
*const_vertex
= nir_src_as_const_value(*vertex
);
228 intrin
->const_index
[0] += const_vertex
->u32
[0] *
229 vue_map
->num_per_vertex_slots
;
231 b
->cursor
= nir_before_instr(&intrin
->instr
);
233 /* Multiply by the number of per-vertex slots. */
234 nir_ssa_def
*vertex_offset
=
236 nir_ssa_for_src(b
, *vertex
, 1),
238 vue_map
->num_per_vertex_slots
));
240 /* Add it to the existing offset */
241 nir_src
*offset
= nir_get_io_offset_src(intrin
);
242 nir_ssa_def
*total_offset
=
243 nir_iadd(b
, vertex_offset
,
244 nir_ssa_for_src(b
, *offset
, 1));
246 nir_instr_rewrite_src(&intrin
->instr
, offset
,
247 nir_src_for_ssa(total_offset
));
256 brw_nir_lower_vs_inputs(nir_shader
*nir
,
258 bool use_legacy_snorm_formula
,
259 const uint8_t *vs_attrib_wa_flags
)
261 /* Start with the location of the variable's base. */
262 foreach_list_typed(nir_variable
, var
, node
, &nir
->inputs
) {
263 var
->data
.driver_location
= var
->data
.location
;
266 /* Now use nir_lower_io to walk dereference chains. Attribute arrays are
267 * loaded as one vec4 or dvec4 per element (or matrix column), depending on
268 * whether it is a double-precision type or not.
270 nir_lower_io(nir
, nir_var_shader_in
, type_size_vec4
, 0);
272 /* This pass needs actual constants */
273 nir_opt_constant_folding(nir
);
275 add_const_offset_to_base(nir
, nir_var_shader_in
);
277 brw_nir_apply_attribute_workarounds(nir
, use_legacy_snorm_formula
,
281 /* Finally, translate VERT_ATTRIB_* values into the actual registers. */
283 nir_foreach_function(function
, nir
) {
284 if (function
->impl
) {
285 nir_foreach_block(block
, function
->impl
) {
286 remap_vs_attrs(block
, nir
->info
);
294 brw_nir_lower_vue_inputs(nir_shader
*nir
, bool is_scalar
,
295 const struct brw_vue_map
*vue_map
)
297 foreach_list_typed(nir_variable
, var
, node
, &nir
->inputs
) {
298 var
->data
.driver_location
= var
->data
.location
;
301 /* Inputs are stored in vec4 slots, so use type_size_vec4(). */
302 nir_lower_io(nir
, nir_var_shader_in
, type_size_vec4
, 0);
304 if (is_scalar
|| nir
->stage
!= MESA_SHADER_GEOMETRY
) {
305 /* This pass needs actual constants */
306 nir_opt_constant_folding(nir
);
308 add_const_offset_to_base(nir
, nir_var_shader_in
);
310 nir_foreach_function(function
, nir
) {
311 if (function
->impl
) {
312 nir_foreach_block(block
, function
->impl
) {
313 remap_inputs_with_vue_map(block
, vue_map
);
321 brw_nir_lower_tes_inputs(nir_shader
*nir
, const struct brw_vue_map
*vue_map
)
323 foreach_list_typed(nir_variable
, var
, node
, &nir
->inputs
) {
324 var
->data
.driver_location
= var
->data
.location
;
327 nir_lower_io(nir
, nir_var_shader_in
, type_size_vec4
, 0);
329 /* This pass needs actual constants */
330 nir_opt_constant_folding(nir
);
332 add_const_offset_to_base(nir
, nir_var_shader_in
);
334 nir_foreach_function(function
, nir
) {
335 if (function
->impl
) {
337 nir_builder_init(&b
, function
->impl
);
338 nir_foreach_block(block
, function
->impl
) {
339 remap_patch_urb_offsets(block
, &b
, vue_map
,
340 nir
->info
->tess
.primitive_mode
);
347 brw_nir_lower_fs_inputs(nir_shader
*nir
,
348 const struct gen_device_info
*devinfo
,
349 const struct brw_wm_prog_key
*key
)
351 foreach_list_typed(nir_variable
, var
, node
, &nir
->inputs
) {
352 var
->data
.driver_location
= var
->data
.location
;
354 /* Apply default interpolation mode.
356 * Everything defaults to smooth except for the legacy GL color
357 * built-in variables, which might be flat depending on API state.
359 if (var
->data
.interpolation
== INTERP_MODE_NONE
) {
360 const bool flat
= key
->flat_shade
&&
361 (var
->data
.location
== VARYING_SLOT_COL0
||
362 var
->data
.location
== VARYING_SLOT_COL1
);
364 var
->data
.interpolation
= flat
? INTERP_MODE_FLAT
365 : INTERP_MODE_SMOOTH
;
368 /* On Ironlake and below, there is only one interpolation mode.
369 * Centroid interpolation doesn't mean anything on this hardware --
370 * there is no multisampling.
372 if (devinfo
->gen
< 6) {
373 var
->data
.centroid
= false;
374 var
->data
.sample
= false;
378 nir_lower_io_options lower_io_options
= 0;
379 if (key
->persample_interp
)
380 lower_io_options
|= nir_lower_io_force_sample_interpolation
;
382 nir_lower_io(nir
, nir_var_shader_in
, type_size_vec4
, lower_io_options
);
384 /* This pass needs actual constants */
385 nir_opt_constant_folding(nir
);
387 add_const_offset_to_base(nir
, nir_var_shader_in
);
391 brw_nir_lower_vue_outputs(nir_shader
*nir
,
394 nir_foreach_variable(var
, &nir
->outputs
) {
395 var
->data
.driver_location
= var
->data
.location
;
398 nir_lower_io(nir
, nir_var_shader_out
, type_size_vec4
, 0);
402 brw_nir_lower_tcs_outputs(nir_shader
*nir
, const struct brw_vue_map
*vue_map
,
403 GLenum tes_primitive_mode
)
405 nir_foreach_variable(var
, &nir
->outputs
) {
406 var
->data
.driver_location
= var
->data
.location
;
409 nir_lower_io(nir
, nir_var_shader_out
, type_size_vec4
, 0);
411 /* This pass needs actual constants */
412 nir_opt_constant_folding(nir
);
414 add_const_offset_to_base(nir
, nir_var_shader_out
);
416 nir_foreach_function(function
, nir
) {
417 if (function
->impl
) {
419 nir_builder_init(&b
, function
->impl
);
420 nir_foreach_block(block
, function
->impl
) {
421 remap_patch_urb_offsets(block
, &b
, vue_map
, tes_primitive_mode
);
428 brw_nir_lower_fs_outputs(nir_shader
*nir
)
430 nir_foreach_variable(var
, &nir
->outputs
) {
431 var
->data
.driver_location
=
432 SET_FIELD(var
->data
.index
, BRW_NIR_FRAG_OUTPUT_INDEX
) |
433 SET_FIELD(var
->data
.location
, BRW_NIR_FRAG_OUTPUT_LOCATION
);
436 nir_lower_io(nir
, nir_var_shader_out
, type_size_dvec4
, 0);
440 brw_nir_lower_cs_shared(nir_shader
*nir
)
442 nir_assign_var_locations(&nir
->shared
, &nir
->num_shared
,
443 type_size_scalar_bytes
);
444 nir_lower_io(nir
, nir_var_shared
, type_size_scalar_bytes
, 0);
447 #define OPT(pass, ...) ({ \
448 bool this_progress = false; \
449 NIR_PASS(this_progress, nir, pass, ##__VA_ARGS__); \
455 #define OPT_V(pass, ...) NIR_PASS_V(nir, pass, ##__VA_ARGS__)
458 nir_optimize(nir_shader
*nir
, const struct brw_compiler
*compiler
,
461 nir_variable_mode indirect_mask
= 0;
462 if (compiler
->glsl_compiler_options
[nir
->stage
].EmitNoIndirectInput
)
463 indirect_mask
|= nir_var_shader_in
;
464 if (compiler
->glsl_compiler_options
[nir
->stage
].EmitNoIndirectOutput
)
465 indirect_mask
|= nir_var_shader_out
;
466 if (compiler
->glsl_compiler_options
[nir
->stage
].EmitNoIndirectTemp
)
467 indirect_mask
|= nir_var_local
;
472 OPT_V(nir_lower_vars_to_ssa
);
473 OPT(nir_opt_copy_prop_vars
);
476 OPT(nir_lower_alu_to_scalar
);
482 OPT(nir_lower_phis_to_scalar
);
488 OPT(nir_opt_peephole_select
, 0);
489 OPT(nir_opt_algebraic
);
490 OPT(nir_opt_constant_folding
);
491 OPT(nir_opt_dead_cf
);
492 if (OPT(nir_opt_trivial_continues
)) {
493 /* If nir_opt_trivial_continues makes progress, then we need to clean
494 * things up if we want any hope of nir_opt_if or nir_opt_loop_unroll
501 if (nir
->options
->max_unroll_iterations
!= 0) {
502 OPT(nir_opt_loop_unroll
, indirect_mask
);
504 OPT(nir_opt_remove_phis
);
506 OPT_V(nir_lower_doubles
, nir_lower_drcp
|
513 nir_lower_dround_even
|
515 OPT_V(nir_lower_64bit_pack
);
521 /* Does some simple lowering and runs the standard suite of optimizations
523 * This is intended to be called more-or-less directly after you get the
524 * shader out of GLSL or some other source. While it is geared towards i965,
525 * it is not at all generator-specific except for the is_scalar flag. Even
526 * there, it is safe to call with is_scalar = false for a shader that is
527 * intended for the FS backend as long as nir_optimize is called again with
528 * is_scalar = true to scalarize everything prior to code gen.
531 brw_preprocess_nir(const struct brw_compiler
*compiler
, nir_shader
*nir
)
533 const struct gen_device_info
*devinfo
= compiler
->devinfo
;
534 bool progress
; /* Written by OPT and OPT_V */
537 const bool is_scalar
= compiler
->scalar_stage
[nir
->stage
];
539 if (nir
->stage
== MESA_SHADER_GEOMETRY
)
540 OPT(nir_lower_gs_intrinsics
);
542 /* See also brw_nir_trig_workarounds.py */
543 if (compiler
->precise_trig
&&
544 !(devinfo
->gen
>= 10 || devinfo
->is_kabylake
))
545 OPT(brw_nir_apply_trig_workarounds
);
547 static const nir_lower_tex_options tex_options
= {
549 .lower_txf_offset
= true,
550 .lower_rect_offset
= true,
551 .lower_txd_cube_map
= true,
554 OPT(nir_lower_tex
, &tex_options
);
555 OPT(nir_normalize_cubemap_coords
);
557 OPT(nir_lower_global_vars_to_local
);
559 OPT(nir_split_var_copies
);
561 nir
= nir_optimize(nir
, compiler
, is_scalar
);
564 OPT_V(nir_lower_load_const_to_scalar
);
567 /* Lower a bunch of stuff */
568 OPT_V(nir_lower_var_copies
);
570 OPT_V(nir_lower_clip_cull_distance_arrays
);
572 nir_variable_mode indirect_mask
= 0;
573 if (compiler
->glsl_compiler_options
[nir
->stage
].EmitNoIndirectInput
)
574 indirect_mask
|= nir_var_shader_in
;
575 if (compiler
->glsl_compiler_options
[nir
->stage
].EmitNoIndirectOutput
)
576 indirect_mask
|= nir_var_shader_out
;
577 if (compiler
->glsl_compiler_options
[nir
->stage
].EmitNoIndirectTemp
)
578 indirect_mask
|= nir_var_local
;
580 nir_lower_indirect_derefs(nir
, indirect_mask
);
582 nir_lower_int64(nir
, nir_lower_imul64
|
586 /* Get rid of split copies */
587 nir
= nir_optimize(nir
, compiler
, is_scalar
);
589 OPT(nir_remove_dead_variables
, nir_var_local
);
594 /* Prepare the given shader for codegen
596 * This function is intended to be called right before going into the actual
597 * backend and is highly backend-specific. Also, once this function has been
598 * called on a shader, it will no longer be in SSA form so most optimizations
602 brw_postprocess_nir(nir_shader
*nir
, const struct brw_compiler
*compiler
,
605 const struct gen_device_info
*devinfo
= compiler
->devinfo
;
607 (INTEL_DEBUG
& intel_debug_flag_for_shader_stage(nir
->stage
));
609 bool progress
; /* Written by OPT and OPT_V */
612 nir
= nir_optimize(nir
, compiler
, is_scalar
);
614 if (devinfo
->gen
>= 6) {
615 /* Try and fuse multiply-adds */
616 OPT(brw_nir_opt_peephole_ffma
);
619 OPT(nir_opt_algebraic_late
);
621 OPT_V(nir_lower_to_source_mods
);
624 OPT(nir_opt_move_comparisons
);
626 OPT(nir_lower_locals_to_regs
);
628 if (unlikely(debug_enabled
)) {
629 /* Re-index SSA defs so we print more sensible numbers. */
630 nir_foreach_function(function
, nir
) {
632 nir_index_ssa_defs(function
->impl
);
635 fprintf(stderr
, "NIR (SSA form) for %s shader:\n",
636 _mesa_shader_stage_to_string(nir
->stage
));
637 nir_print_shader(nir
, stderr
);
640 OPT_V(nir_convert_from_ssa
, true);
643 OPT_V(nir_move_vec_src_uses_to_dest
);
644 OPT(nir_lower_vec_to_movs
);
647 /* This is the last pass we run before we start emitting stuff. It
648 * determines when we need to insert boolean resolves on Gen <= 5. We
649 * run it last because it stashes data in instr->pass_flags and we don't
650 * want that to be squashed by other NIR passes.
652 if (devinfo
->gen
<= 5)
653 brw_nir_analyze_boolean_resolves(nir
);
657 if (unlikely(debug_enabled
)) {
658 fprintf(stderr
, "NIR (final form) for %s shader:\n",
659 _mesa_shader_stage_to_string(nir
->stage
));
660 nir_print_shader(nir
, stderr
);
667 brw_nir_apply_sampler_key(nir_shader
*nir
,
668 const struct brw_compiler
*compiler
,
669 const struct brw_sampler_prog_key_data
*key_tex
,
672 const struct gen_device_info
*devinfo
= compiler
->devinfo
;
673 nir_lower_tex_options tex_options
= { 0 };
675 /* Iron Lake and prior require lowering of all rectangle textures */
676 if (devinfo
->gen
< 6)
677 tex_options
.lower_rect
= true;
679 /* Prior to Broadwell, our hardware can't actually do GL_CLAMP */
680 if (devinfo
->gen
< 8) {
681 tex_options
.saturate_s
= key_tex
->gl_clamp_mask
[0];
682 tex_options
.saturate_t
= key_tex
->gl_clamp_mask
[1];
683 tex_options
.saturate_r
= key_tex
->gl_clamp_mask
[2];
686 /* Prior to Haswell, we have to fake texture swizzle */
687 for (unsigned s
= 0; s
< MAX_SAMPLERS
; s
++) {
688 if (key_tex
->swizzles
[s
] == SWIZZLE_NOOP
)
691 tex_options
.swizzle_result
|= (1 << s
);
692 for (unsigned c
= 0; c
< 4; c
++)
693 tex_options
.swizzles
[s
][c
] = GET_SWZ(key_tex
->swizzles
[s
], c
);
696 /* Prior to Haswell, we have to lower gradients on shadow samplers */
697 tex_options
.lower_txd_shadow
= devinfo
->gen
< 8 && !devinfo
->is_haswell
;
699 tex_options
.lower_y_uv_external
= key_tex
->y_uv_image_mask
;
700 tex_options
.lower_y_u_v_external
= key_tex
->y_u_v_image_mask
;
701 tex_options
.lower_yx_xuxv_external
= key_tex
->yx_xuxv_image_mask
;
703 if (nir_lower_tex(nir
, &tex_options
)) {
704 nir_validate_shader(nir
);
705 nir
= nir_optimize(nir
, compiler
, is_scalar
);
712 brw_type_for_nir_type(const struct gen_device_info
*devinfo
, nir_alu_type type
)
716 case nir_type_uint32
:
717 return BRW_REGISTER_TYPE_UD
;
720 case nir_type_bool32
:
722 return BRW_REGISTER_TYPE_D
;
724 case nir_type_float32
:
725 return BRW_REGISTER_TYPE_F
;
726 case nir_type_float64
:
727 return BRW_REGISTER_TYPE_DF
;
729 return devinfo
->gen
< 8 ? BRW_REGISTER_TYPE_DF
: BRW_REGISTER_TYPE_Q
;
730 case nir_type_uint64
:
731 return devinfo
->gen
< 8 ? BRW_REGISTER_TYPE_DF
: BRW_REGISTER_TYPE_UQ
;
733 unreachable("unknown type");
736 return BRW_REGISTER_TYPE_F
;
739 /* Returns the glsl_base_type corresponding to a nir_alu_type.
740 * This is used by both brw_vec4_nir and brw_fs_nir.
743 brw_glsl_base_type_for_nir_type(nir_alu_type type
)
747 case nir_type_float32
:
748 return GLSL_TYPE_FLOAT
;
750 case nir_type_float64
:
751 return GLSL_TYPE_DOUBLE
;
755 return GLSL_TYPE_INT
;
758 case nir_type_uint32
:
759 return GLSL_TYPE_UINT
;
762 unreachable("bad type");