2 * Copyright © 2015 Broadcom
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 * This lowering pass supports (as configured via nir_lower_tex_options)
26 * various texture related conversions:
27 * + texture projector lowering: converts the coordinate division for
28 * texture projection to be done in ALU instructions instead of
29 * asking the texture operation to do so.
30 * + lowering RECT: converts the un-normalized RECT texture coordinates
31 * to normalized coordinates with txs plus ALU instructions
32 * + saturate s/t/r coords: to emulate certain texture clamp/wrap modes,
33 * inserts instructions to clamp specified coordinates to [0.0, 1.0].
34 * Note that this automatically triggers texture projector lowering if
35 * needed, since clamping must happen after projector lowering.
39 #include "nir_builder.h"
42 project_src(nir_builder
*b
, nir_tex_instr
*tex
)
44 /* Find the projector in the srcs list, if present. */
46 for (proj_index
= 0; proj_index
< tex
->num_srcs
; proj_index
++) {
47 if (tex
->src
[proj_index
].src_type
== nir_tex_src_projector
)
50 if (proj_index
== tex
->num_srcs
)
53 b
->cursor
= nir_before_instr(&tex
->instr
);
55 nir_ssa_def
*inv_proj
=
56 nir_frcp(b
, nir_ssa_for_src(b
, tex
->src
[proj_index
].src
, 1));
58 /* Walk through the sources projecting the arguments. */
59 for (unsigned i
= 0; i
< tex
->num_srcs
; i
++) {
60 switch (tex
->src
[i
].src_type
) {
61 case nir_tex_src_coord
:
62 case nir_tex_src_comparitor
:
67 nir_ssa_def
*unprojected
=
68 nir_ssa_for_src(b
, tex
->src
[i
].src
, nir_tex_instr_src_size(tex
, i
));
69 nir_ssa_def
*projected
= nir_fmul(b
, unprojected
, inv_proj
);
71 /* Array indices don't get projected, so make an new vector with the
72 * coordinate's array index untouched.
74 if (tex
->is_array
&& tex
->src
[i
].src_type
== nir_tex_src_coord
) {
75 switch (tex
->coord_components
) {
77 projected
= nir_vec4(b
,
78 nir_channel(b
, projected
, 0),
79 nir_channel(b
, projected
, 1),
80 nir_channel(b
, projected
, 2),
81 nir_channel(b
, unprojected
, 3));
84 projected
= nir_vec3(b
,
85 nir_channel(b
, projected
, 0),
86 nir_channel(b
, projected
, 1),
87 nir_channel(b
, unprojected
, 2));
90 projected
= nir_vec2(b
,
91 nir_channel(b
, projected
, 0),
92 nir_channel(b
, unprojected
, 1));
95 unreachable("bad texture coord count for array");
100 nir_instr_rewrite_src(&tex
->instr
,
102 nir_src_for_ssa(projected
));
105 /* Now move the later tex sources down the array so that the projector
108 nir_instr_rewrite_src(&tex
->instr
, &tex
->src
[proj_index
].src
,
110 for (unsigned i
= proj_index
+ 1; i
< tex
->num_srcs
; i
++) {
111 tex
->src
[i
-1].src_type
= tex
->src
[i
].src_type
;
112 nir_instr_move_src(&tex
->instr
, &tex
->src
[i
-1].src
, &tex
->src
[i
].src
);
118 get_texture_size(nir_builder
*b
, nir_tex_instr
*tex
)
120 b
->cursor
= nir_before_instr(&tex
->instr
);
122 /* RECT textures should not be array: */
123 assert(!tex
->is_array
);
127 txs
= nir_tex_instr_create(b
->shader
, 1);
128 txs
->op
= nir_texop_txs
;
129 txs
->sampler_dim
= GLSL_SAMPLER_DIM_RECT
;
130 txs
->texture_index
= tex
->texture_index
;
131 txs
->dest_type
= nir_type_int
;
133 /* only single src, the lod: */
134 txs
->src
[0].src
= nir_src_for_ssa(nir_imm_int(b
, 0));
135 txs
->src
[0].src_type
= nir_tex_src_lod
;
137 nir_ssa_dest_init(&txs
->instr
, &txs
->dest
, 2, 32, NULL
);
138 nir_builder_instr_insert(b
, &txs
->instr
);
140 return nir_i2f(b
, &txs
->dest
.ssa
);
144 lower_rect(nir_builder
*b
, nir_tex_instr
*tex
)
146 nir_ssa_def
*txs
= get_texture_size(b
, tex
);
147 nir_ssa_def
*scale
= nir_frcp(b
, txs
);
149 /* Walk through the sources normalizing the requested arguments. */
150 for (unsigned i
= 0; i
< tex
->num_srcs
; i
++) {
151 if (tex
->src
[i
].src_type
!= nir_tex_src_coord
)
154 nir_ssa_def
*coords
=
155 nir_ssa_for_src(b
, tex
->src
[i
].src
, tex
->coord_components
);
156 nir_instr_rewrite_src(&tex
->instr
,
158 nir_src_for_ssa(nir_fmul(b
, coords
, scale
)));
161 tex
->sampler_dim
= GLSL_SAMPLER_DIM_2D
;
165 saturate_src(nir_builder
*b
, nir_tex_instr
*tex
, unsigned sat_mask
)
167 b
->cursor
= nir_before_instr(&tex
->instr
);
169 /* Walk through the sources saturating the requested arguments. */
170 for (unsigned i
= 0; i
< tex
->num_srcs
; i
++) {
171 if (tex
->src
[i
].src_type
!= nir_tex_src_coord
)
175 nir_ssa_for_src(b
, tex
->src
[i
].src
, tex
->coord_components
);
177 /* split src into components: */
178 nir_ssa_def
*comp
[4];
180 assume(tex
->coord_components
>= 1);
182 for (unsigned j
= 0; j
< tex
->coord_components
; j
++)
183 comp
[j
] = nir_channel(b
, src
, j
);
185 /* clamp requested components, array index does not get clamped: */
186 unsigned ncomp
= tex
->coord_components
;
190 for (unsigned j
= 0; j
< ncomp
; j
++) {
191 if ((1 << j
) & sat_mask
) {
192 if (tex
->sampler_dim
== GLSL_SAMPLER_DIM_RECT
) {
193 /* non-normalized texture coords, so clamp to texture
194 * size rather than [0.0, 1.0]
196 nir_ssa_def
*txs
= get_texture_size(b
, tex
);
197 comp
[j
] = nir_fmax(b
, comp
[j
], nir_imm_float(b
, 0.0));
198 comp
[j
] = nir_fmin(b
, comp
[j
], nir_channel(b
, txs
, j
));
200 comp
[j
] = nir_fsat(b
, comp
[j
]);
205 /* and move the result back into a single vecN: */
206 src
= nir_vec(b
, comp
, tex
->coord_components
);
208 nir_instr_rewrite_src(&tex
->instr
,
210 nir_src_for_ssa(src
));
215 get_zero_or_one(nir_builder
*b
, nir_alu_type type
, uint8_t swizzle_val
)
219 memset(&v
, 0, sizeof(v
));
221 if (swizzle_val
== 4) {
222 v
.u32
[0] = v
.u32
[1] = v
.u32
[2] = v
.u32
[3] = 0;
224 assert(swizzle_val
== 5);
225 if (type
== nir_type_float
)
226 v
.f32
[0] = v
.f32
[1] = v
.f32
[2] = v
.f32
[3] = 1.0;
228 v
.u32
[0] = v
.u32
[1] = v
.u32
[2] = v
.u32
[3] = 1;
231 return nir_build_imm(b
, 4, 32, v
);
235 swizzle_result(nir_builder
*b
, nir_tex_instr
*tex
, const uint8_t swizzle
[4])
237 assert(tex
->dest
.is_ssa
);
239 b
->cursor
= nir_after_instr(&tex
->instr
);
241 nir_ssa_def
*swizzled
;
242 if (tex
->op
== nir_texop_tg4
) {
243 if (swizzle
[tex
->component
] < 4) {
244 /* This one's easy */
245 tex
->component
= swizzle
[tex
->component
];
248 swizzled
= get_zero_or_one(b
, tex
->dest_type
, swizzle
[tex
->component
]);
251 assert(nir_tex_instr_dest_size(tex
) == 4);
252 if (swizzle
[0] < 4 && swizzle
[1] < 4 &&
253 swizzle
[2] < 4 && swizzle
[3] < 4) {
254 unsigned swiz
[4] = { swizzle
[0], swizzle
[1], swizzle
[2], swizzle
[3] };
255 /* We have no 0's or 1's, just emit a swizzling MOV */
256 swizzled
= nir_swizzle(b
, &tex
->dest
.ssa
, swiz
, 4, false);
258 nir_ssa_def
*srcs
[4];
259 for (unsigned i
= 0; i
< 4; i
++) {
260 if (swizzle
[i
] < 4) {
261 srcs
[i
] = nir_channel(b
, &tex
->dest
.ssa
, swizzle
[i
]);
263 srcs
[i
] = get_zero_or_one(b
, tex
->dest_type
, swizzle
[i
]);
266 swizzled
= nir_vec(b
, srcs
, 4);
270 nir_ssa_def_rewrite_uses_after(&tex
->dest
.ssa
, nir_src_for_ssa(swizzled
),
271 swizzled
->parent_instr
);
275 linearize_srgb_result(nir_builder
*b
, nir_tex_instr
*tex
)
277 assert(tex
->dest
.is_ssa
);
278 assert(nir_tex_instr_dest_size(tex
) == 4);
279 assert(nir_alu_type_get_base_type(tex
->dest_type
) == nir_type_float
);
281 b
->cursor
= nir_after_instr(&tex
->instr
);
283 static const unsigned swiz
[4] = {0, 1, 2, 0};
284 nir_ssa_def
*comp
= nir_swizzle(b
, &tex
->dest
.ssa
, swiz
, 3, true);
287 * (comp <= 0.04045) ?
289 * pow((comp + 0.055) / 1.055, 2.4)
291 nir_ssa_def
*low
= nir_fmul(b
, comp
, nir_imm_float(b
, 1.0 / 12.92));
292 nir_ssa_def
*high
= nir_fpow(b
,
296 nir_imm_float(b
, 0.055)),
297 nir_imm_float(b
, 1.0 / 1.055)),
298 nir_imm_float(b
, 2.4));
299 nir_ssa_def
*cond
= nir_fge(b
, nir_imm_float(b
, 0.04045), comp
);
300 nir_ssa_def
*rgb
= nir_bcsel(b
, cond
, low
, high
);
302 /* alpha is untouched: */
303 nir_ssa_def
*result
= nir_vec4(b
,
304 nir_channel(b
, rgb
, 0),
305 nir_channel(b
, rgb
, 1),
306 nir_channel(b
, rgb
, 2),
307 nir_channel(b
, &tex
->dest
.ssa
, 3));
309 nir_ssa_def_rewrite_uses_after(&tex
->dest
.ssa
, nir_src_for_ssa(result
),
310 result
->parent_instr
);
314 nir_lower_tex_block(nir_block
*block
, nir_builder
*b
,
315 const nir_lower_tex_options
*options
)
317 bool progress
= false;
319 nir_foreach_instr_safe(instr
, block
) {
320 if (instr
->type
!= nir_instr_type_tex
)
323 nir_tex_instr
*tex
= nir_instr_as_tex(instr
);
324 bool lower_txp
= !!(options
->lower_txp
& (1 << tex
->sampler_dim
));
326 /* mask of src coords to saturate (clamp): */
327 unsigned sat_mask
= 0;
329 if ((1 << tex
->sampler_index
) & options
->saturate_r
)
330 sat_mask
|= (1 << 2); /* .z */
331 if ((1 << tex
->sampler_index
) & options
->saturate_t
)
332 sat_mask
|= (1 << 1); /* .y */
333 if ((1 << tex
->sampler_index
) & options
->saturate_s
)
334 sat_mask
|= (1 << 0); /* .x */
336 /* If we are clamping any coords, we must lower projector first
337 * as clamping happens *after* projection:
339 if (lower_txp
|| sat_mask
) {
344 if ((tex
->sampler_dim
== GLSL_SAMPLER_DIM_RECT
) && options
->lower_rect
) {
350 saturate_src(b
, tex
, sat_mask
);
354 if (((1 << tex
->texture_index
) & options
->swizzle_result
) &&
355 !nir_tex_instr_is_query(tex
) &&
356 !(tex
->is_shadow
&& tex
->is_new_style_shadow
)) {
357 swizzle_result(b
, tex
, options
->swizzles
[tex
->texture_index
]);
361 /* should be after swizzle so we know which channels are rgb: */
362 if (((1 << tex
->texture_index
) & options
->lower_srgb
) &&
363 !nir_tex_instr_is_query(tex
) && !tex
->is_shadow
) {
364 linearize_srgb_result(b
, tex
);
373 nir_lower_tex_impl(nir_function_impl
*impl
,
374 const nir_lower_tex_options
*options
)
376 bool progress
= false;
378 nir_builder_init(&builder
, impl
);
380 nir_foreach_block(block
, impl
) {
381 progress
|= nir_lower_tex_block(block
, &builder
, options
);
384 nir_metadata_preserve(impl
, nir_metadata_block_index
|
385 nir_metadata_dominance
);
390 nir_lower_tex(nir_shader
*shader
, const nir_lower_tex_options
*options
)
392 bool progress
= false;
394 nir_foreach_function(function
, shader
) {
396 progress
|= nir_lower_tex_impl(function
->impl
, options
);