2 * Copyright © 2013 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
24 #include "util/ralloc.h"
26 #include "main/macros.h" /* Needed for MAX3 and MAX2 for format_rgb9e5 */
27 #include "util/format_rgb9e5.h"
28 #include "util/format_srgb.h"
30 #include "blorp_priv.h"
31 #include "compiler/brw_eu_defines.h"
33 #include "blorp_nir_builder.h"
35 #define FILE_DEBUG_FLAG DEBUG_BLORP
37 struct brw_blorp_const_color_prog_key
39 enum blorp_shader_type shader_type
; /* Must be BLORP_SHADER_TYPE_CLEAR */
40 bool use_simd16_replicated_data
;
41 bool clear_rgb_as_red
;
46 blorp_params_get_clear_kernel(struct blorp_batch
*batch
,
47 struct blorp_params
*params
,
48 bool use_replicated_data
,
49 bool clear_rgb_as_red
)
51 struct blorp_context
*blorp
= batch
->blorp
;
53 const struct brw_blorp_const_color_prog_key blorp_key
= {
54 .shader_type
= BLORP_SHADER_TYPE_CLEAR
,
55 .use_simd16_replicated_data
= use_replicated_data
,
56 .clear_rgb_as_red
= clear_rgb_as_red
,
59 if (blorp
->lookup_shader(batch
, &blorp_key
, sizeof(blorp_key
),
60 ¶ms
->wm_prog_kernel
, ¶ms
->wm_prog_data
))
63 void *mem_ctx
= ralloc_context(NULL
);
66 blorp_nir_init_shader(&b
, mem_ctx
, MESA_SHADER_FRAGMENT
, "BLORP-clear");
68 nir_variable
*v_color
=
69 BLORP_CREATE_NIR_INPUT(b
.shader
, clear_color
, glsl_vec4_type());
70 nir_ssa_def
*color
= nir_load_var(&b
, v_color
);
72 if (clear_rgb_as_red
) {
73 nir_variable
*frag_coord
=
74 nir_variable_create(b
.shader
, nir_var_shader_in
,
75 glsl_vec4_type(), "gl_FragCoord");
76 frag_coord
->data
.location
= VARYING_SLOT_POS
;
77 frag_coord
->data
.origin_upper_left
= true;
79 nir_ssa_def
*pos
= nir_f2i32(&b
, nir_load_var(&b
, frag_coord
));
80 nir_ssa_def
*comp
= nir_umod(&b
, nir_channel(&b
, pos
, 0),
82 nir_ssa_def
*color_component
=
83 nir_bcsel(&b
, nir_ieq(&b
, comp
, nir_imm_int(&b
, 0)),
84 nir_channel(&b
, color
, 0),
85 nir_bcsel(&b
, nir_ieq(&b
, comp
, nir_imm_int(&b
, 1)),
86 nir_channel(&b
, color
, 1),
87 nir_channel(&b
, color
, 2)));
89 nir_ssa_def
*u
= nir_ssa_undef(&b
, 1, 32);
90 color
= nir_vec4(&b
, color_component
, u
, u
, u
);
93 nir_variable
*frag_color
= nir_variable_create(b
.shader
, nir_var_shader_out
,
96 frag_color
->data
.location
= FRAG_RESULT_COLOR
;
97 nir_store_var(&b
, frag_color
, color
, 0xf);
99 struct brw_wm_prog_key wm_key
;
100 brw_blorp_init_wm_prog_key(&wm_key
);
102 struct brw_wm_prog_data prog_data
;
103 const unsigned *program
=
104 blorp_compile_fs(blorp
, mem_ctx
, b
.shader
, &wm_key
, use_replicated_data
,
108 blorp
->upload_shader(batch
, &blorp_key
, sizeof(blorp_key
),
109 program
, prog_data
.base
.program_size
,
110 &prog_data
.base
, sizeof(prog_data
),
111 ¶ms
->wm_prog_kernel
, ¶ms
->wm_prog_data
);
113 ralloc_free(mem_ctx
);
117 struct layer_offset_vs_key
{
118 enum blorp_shader_type shader_type
;
122 /* In the case of doing attachment clears, we are using a surface state that
123 * is handed to us so we can't set (and don't even know) the base array layer.
124 * In order to do a layered clear in this scenario, we need some way of adding
125 * the base array layer to the instance id. Unfortunately, our hardware has
126 * no real concept of "base instance", so we have to do it manually in a
130 blorp_params_get_layer_offset_vs(struct blorp_batch
*batch
,
131 struct blorp_params
*params
)
133 struct blorp_context
*blorp
= batch
->blorp
;
134 struct layer_offset_vs_key blorp_key
= {
135 .shader_type
= BLORP_SHADER_TYPE_LAYER_OFFSET_VS
,
138 if (params
->wm_prog_data
)
139 blorp_key
.num_inputs
= params
->wm_prog_data
->num_varying_inputs
;
141 if (blorp
->lookup_shader(batch
, &blorp_key
, sizeof(blorp_key
),
142 ¶ms
->vs_prog_kernel
, ¶ms
->vs_prog_data
))
145 void *mem_ctx
= ralloc_context(NULL
);
148 blorp_nir_init_shader(&b
, mem_ctx
, MESA_SHADER_VERTEX
, "BLORP-layer-offset-vs");
150 const struct glsl_type
*uvec4_type
= glsl_vector_type(GLSL_TYPE_UINT
, 4);
152 /* First we deal with the header which has instance and base instance */
153 nir_variable
*a_header
= nir_variable_create(b
.shader
, nir_var_shader_in
,
154 uvec4_type
, "header");
155 a_header
->data
.location
= VERT_ATTRIB_GENERIC0
;
157 nir_variable
*v_layer
= nir_variable_create(b
.shader
, nir_var_shader_out
,
158 glsl_int_type(), "layer_id");
159 v_layer
->data
.location
= VARYING_SLOT_LAYER
;
161 /* Compute the layer id */
162 nir_ssa_def
*header
= nir_load_var(&b
, a_header
);
163 nir_ssa_def
*base_layer
= nir_channel(&b
, header
, 0);
164 nir_ssa_def
*instance
= nir_channel(&b
, header
, 1);
165 nir_store_var(&b
, v_layer
, nir_iadd(&b
, instance
, base_layer
), 0x1);
167 /* Then we copy the vertex from the next slot to VARYING_SLOT_POS */
168 nir_variable
*a_vertex
= nir_variable_create(b
.shader
, nir_var_shader_in
,
169 glsl_vec4_type(), "a_vertex");
170 a_vertex
->data
.location
= VERT_ATTRIB_GENERIC1
;
172 nir_variable
*v_pos
= nir_variable_create(b
.shader
, nir_var_shader_out
,
173 glsl_vec4_type(), "v_pos");
174 v_pos
->data
.location
= VARYING_SLOT_POS
;
176 nir_copy_var(&b
, v_pos
, a_vertex
);
178 /* Then we copy everything else */
179 for (unsigned i
= 0; i
< blorp_key
.num_inputs
; i
++) {
180 nir_variable
*a_in
= nir_variable_create(b
.shader
, nir_var_shader_in
,
181 uvec4_type
, "input");
182 a_in
->data
.location
= VERT_ATTRIB_GENERIC2
+ i
;
184 nir_variable
*v_out
= nir_variable_create(b
.shader
, nir_var_shader_out
,
185 uvec4_type
, "output");
186 v_out
->data
.location
= VARYING_SLOT_VAR0
+ i
;
188 nir_copy_var(&b
, v_out
, a_in
);
191 struct brw_vs_prog_data vs_prog_data
;
192 memset(&vs_prog_data
, 0, sizeof(vs_prog_data
));
194 const unsigned *program
=
195 blorp_compile_vs(blorp
, mem_ctx
, b
.shader
, &vs_prog_data
);
198 blorp
->upload_shader(batch
, &blorp_key
, sizeof(blorp_key
),
199 program
, vs_prog_data
.base
.base
.program_size
,
200 &vs_prog_data
.base
.base
, sizeof(vs_prog_data
),
201 ¶ms
->vs_prog_kernel
, ¶ms
->vs_prog_data
);
203 ralloc_free(mem_ctx
);
207 /* The x0, y0, x1, and y1 parameters must already be populated with the render
208 * area of the framebuffer to be cleared.
211 get_fast_clear_rect(const struct isl_device
*dev
,
212 const struct isl_surf
*aux_surf
,
213 unsigned *x0
, unsigned *y0
,
214 unsigned *x1
, unsigned *y1
)
216 unsigned int x_align
, y_align
;
217 unsigned int x_scaledown
, y_scaledown
;
219 /* Only single sampled surfaces need to (and actually can) be resolved. */
220 if (aux_surf
->usage
== ISL_SURF_USAGE_CCS_BIT
) {
221 /* From the Ivy Bridge PRM, Vol2 Part1 11.7 "MCS Buffer for Render
222 * Target(s)", beneath the "Fast Color Clear" bullet (p327):
224 * Clear pass must have a clear rectangle that must follow
225 * alignment rules in terms of pixels and lines as shown in the
226 * table below. Further, the clear-rectangle height and width
227 * must be multiple of the following dimensions. If the height
228 * and width of the render target being cleared do not meet these
229 * requirements, an MCS buffer can be created such that it
230 * follows the requirement and covers the RT.
232 * The alignment size in the table that follows is related to the
233 * alignment size that is baked into the CCS surface format but with X
234 * alignment multiplied by 16 and Y alignment multiplied by 32.
236 x_align
= isl_format_get_layout(aux_surf
->format
)->bw
;
237 y_align
= isl_format_get_layout(aux_surf
->format
)->bh
;
241 /* SKL+ line alignment requirement for Y-tiled are half those of the prior
244 if (dev
->info
->gen
>= 9)
249 /* From the Ivy Bridge PRM, Vol2 Part1 11.7 "MCS Buffer for Render
250 * Target(s)", beneath the "Fast Color Clear" bullet (p327):
252 * In order to optimize the performance MCS buffer (when bound to
253 * 1X RT) clear similarly to MCS buffer clear for MSRT case,
254 * clear rect is required to be scaled by the following factors
255 * in the horizontal and vertical directions:
257 * The X and Y scale down factors in the table that follows are each
258 * equal to half the alignment value computed above.
260 x_scaledown
= x_align
/ 2;
261 y_scaledown
= y_align
/ 2;
263 /* From BSpec: 3D-Media-GPGPU Engine > 3D Pipeline > Pixel > Pixel
264 * Backend > MCS Buffer for Render Target(s) [DevIVB+] > Table "Color
265 * Clear of Non-MultiSampled Render Target Restrictions":
267 * Clear rectangle must be aligned to two times the number of
268 * pixels in the table shown below due to 16x16 hashing across the
274 assert(aux_surf
->usage
== ISL_SURF_USAGE_MCS_BIT
);
276 /* From the Ivy Bridge PRM, Vol2 Part1 11.7 "MCS Buffer for Render
277 * Target(s)", beneath the "MSAA Compression" bullet (p326):
279 * Clear pass for this case requires that scaled down primitive
280 * is sent down with upper left co-ordinate to coincide with
281 * actual rectangle being cleared. For MSAA, clear rectangle’s
282 * height and width need to as show in the following table in
283 * terms of (width,height) of the RT.
285 * MSAA Width of Clear Rect Height of Clear Rect
286 * 2X Ceil(1/8*width) Ceil(1/2*height)
287 * 4X Ceil(1/8*width) Ceil(1/2*height)
288 * 8X Ceil(1/2*width) Ceil(1/2*height)
289 * 16X width Ceil(1/2*height)
291 * The text "with upper left co-ordinate to coincide with actual
292 * rectangle being cleared" is a little confusing--it seems to imply
293 * that to clear a rectangle from (x,y) to (x+w,y+h), one needs to
294 * feed the pipeline using the rectangle (x,y) to
295 * (x+Ceil(w/N),y+Ceil(h/2)), where N is either 2 or 8 depending on
296 * the number of samples. Experiments indicate that this is not
297 * quite correct; actually, what the hardware appears to do is to
298 * align whatever rectangle is sent down the pipeline to the nearest
299 * multiple of 2x2 blocks, and then scale it up by a factor of N
300 * horizontally and 2 vertically. So the resulting alignment is 4
301 * vertically and either 4 or 16 horizontally, and the scaledown
302 * factor is 2 vertically and either 2 or 8 horizontally.
304 switch (aux_surf
->format
) {
305 case ISL_FORMAT_MCS_2X
:
306 case ISL_FORMAT_MCS_4X
:
309 case ISL_FORMAT_MCS_8X
:
312 case ISL_FORMAT_MCS_16X
:
316 unreachable("Unexpected MCS format for fast clear");
319 x_align
= x_scaledown
* 2;
320 y_align
= y_scaledown
* 2;
323 *x0
= ROUND_DOWN_TO(*x0
, x_align
) / x_scaledown
;
324 *y0
= ROUND_DOWN_TO(*y0
, y_align
) / y_scaledown
;
325 *x1
= ALIGN(*x1
, x_align
) / x_scaledown
;
326 *y1
= ALIGN(*y1
, y_align
) / y_scaledown
;
330 blorp_fast_clear(struct blorp_batch
*batch
,
331 const struct blorp_surf
*surf
, enum isl_format format
,
332 uint32_t level
, uint32_t start_layer
, uint32_t num_layers
,
333 uint32_t x0
, uint32_t y0
, uint32_t x1
, uint32_t y1
)
335 /* Ensure that all layers undergoing the clear have an auxiliary buffer. */
336 assert(start_layer
+ num_layers
<=
337 MAX2(surf
->aux_surf
->logical_level0_px
.depth
>> level
,
338 surf
->aux_surf
->logical_level0_px
.array_len
));
340 struct blorp_params params
;
341 blorp_params_init(¶ms
);
342 params
.num_layers
= num_layers
;
349 memset(¶ms
.wm_inputs
.clear_color
, 0xff, 4*sizeof(float));
350 params
.fast_clear_op
= ISL_AUX_OP_FAST_CLEAR
;
352 get_fast_clear_rect(batch
->blorp
->isl_dev
, surf
->aux_surf
,
353 ¶ms
.x0
, ¶ms
.y0
, ¶ms
.x1
, ¶ms
.y1
);
355 if (!blorp_params_get_clear_kernel(batch
, ¶ms
, true, false))
358 brw_blorp_surface_info_init(batch
->blorp
, ¶ms
.dst
, surf
, level
,
359 start_layer
, format
, true);
360 params
.num_samples
= params
.dst
.surf
.samples
;
362 batch
->blorp
->exec(batch
, ¶ms
);
365 static union isl_color_value
366 swizzle_color_value(union isl_color_value src
, struct isl_swizzle swizzle
)
368 union isl_color_value dst
= { .u32
= { 0, } };
370 /* We assign colors in ABGR order so that the first one will be taken in
371 * RGBA precedence order. According to the PRM docs for shader channel
372 * select, this matches Haswell hardware behavior.
374 if ((unsigned)(swizzle
.a
- ISL_CHANNEL_SELECT_RED
) < 4)
375 dst
.u32
[swizzle
.a
- ISL_CHANNEL_SELECT_RED
] = src
.u32
[3];
376 if ((unsigned)(swizzle
.b
- ISL_CHANNEL_SELECT_RED
) < 4)
377 dst
.u32
[swizzle
.b
- ISL_CHANNEL_SELECT_RED
] = src
.u32
[2];
378 if ((unsigned)(swizzle
.g
- ISL_CHANNEL_SELECT_RED
) < 4)
379 dst
.u32
[swizzle
.g
- ISL_CHANNEL_SELECT_RED
] = src
.u32
[1];
380 if ((unsigned)(swizzle
.r
- ISL_CHANNEL_SELECT_RED
) < 4)
381 dst
.u32
[swizzle
.r
- ISL_CHANNEL_SELECT_RED
] = src
.u32
[0];
387 blorp_clear(struct blorp_batch
*batch
,
388 const struct blorp_surf
*surf
,
389 enum isl_format format
, struct isl_swizzle swizzle
,
390 uint32_t level
, uint32_t start_layer
, uint32_t num_layers
,
391 uint32_t x0
, uint32_t y0
, uint32_t x1
, uint32_t y1
,
392 union isl_color_value clear_color
,
393 const bool color_write_disable
[4])
395 struct blorp_params params
;
396 blorp_params_init(¶ms
);
398 /* Manually apply the clear destination swizzle. This way swizzled clears
399 * will work for swizzles which we can't normally use for rendering and it
400 * also ensures that they work on pre-Haswell hardware which can't swizlle
403 clear_color
= swizzle_color_value(clear_color
, swizzle
);
404 swizzle
= ISL_SWIZZLE_IDENTITY
;
406 bool clear_rgb_as_red
= false;
407 if (format
== ISL_FORMAT_R9G9B9E5_SHAREDEXP
) {
408 clear_color
.u32
[0] = float3_to_rgb9e5(clear_color
.f32
);
409 format
= ISL_FORMAT_R32_UINT
;
410 } else if (format
== ISL_FORMAT_L8_UNORM_SRGB
) {
411 clear_color
.f32
[0] = util_format_linear_to_srgb_float(clear_color
.f32
[0]);
412 format
= ISL_FORMAT_R8_UNORM
;
413 } else if (format
== ISL_FORMAT_A4B4G4R4_UNORM
) {
414 /* Broadwell and earlier cannot render to this format so we need to work
415 * around it by swapping the colors around and using B4G4R4A4 instead.
417 const struct isl_swizzle ARGB
= ISL_SWIZZLE(ALPHA
, RED
, GREEN
, BLUE
);
418 clear_color
= swizzle_color_value(clear_color
, ARGB
);
419 format
= ISL_FORMAT_B4G4R4A4_UNORM
;
420 } else if (isl_format_get_layout(format
)->bpb
% 3 == 0) {
421 clear_rgb_as_red
= true;
422 if (format
== ISL_FORMAT_R8G8B8_UNORM_SRGB
) {
423 clear_color
.f32
[0] = util_format_linear_to_srgb_float(clear_color
.f32
[0]);
424 clear_color
.f32
[1] = util_format_linear_to_srgb_float(clear_color
.f32
[1]);
425 clear_color
.f32
[2] = util_format_linear_to_srgb_float(clear_color
.f32
[2]);
429 memcpy(¶ms
.wm_inputs
.clear_color
, clear_color
.f32
, sizeof(float) * 4);
431 bool use_simd16_replicated_data
= true;
433 /* From the SNB PRM (Vol4_Part1):
435 * "Replicated data (Message Type = 111) is only supported when
436 * accessing tiled memory. Using this Message Type to access linear
437 * (untiled) memory is UNDEFINED."
439 if (surf
->surf
->tiling
== ISL_TILING_LINEAR
)
440 use_simd16_replicated_data
= false;
442 /* Replicated clears don't work yet before gen6 */
443 if (batch
->blorp
->isl_dev
->info
->gen
< 6)
444 use_simd16_replicated_data
= false;
446 /* Constant color writes ignore everyting in blend and color calculator
447 * state. This is not documented.
449 if (color_write_disable
) {
450 for (unsigned i
= 0; i
< 4; i
++) {
451 params
.color_write_disable
[i
] = color_write_disable
[i
];
452 if (color_write_disable
[i
])
453 use_simd16_replicated_data
= false;
457 if (!blorp_params_get_clear_kernel(batch
, ¶ms
,
458 use_simd16_replicated_data
,
462 if (!blorp_ensure_sf_program(batch
, ¶ms
))
465 while (num_layers
> 0) {
466 brw_blorp_surface_info_init(batch
->blorp
, ¶ms
.dst
, surf
, level
,
467 start_layer
, format
, true);
468 params
.dst
.view
.swizzle
= swizzle
;
475 if (params
.dst
.tile_x_sa
|| params
.dst
.tile_y_sa
) {
476 assert(params
.dst
.surf
.samples
== 1);
477 assert(num_layers
== 1);
478 params
.x0
+= params
.dst
.tile_x_sa
;
479 params
.y0
+= params
.dst
.tile_y_sa
;
480 params
.x1
+= params
.dst
.tile_x_sa
;
481 params
.y1
+= params
.dst
.tile_y_sa
;
484 /* The MinLOD and MinimumArrayElement don't work properly for cube maps.
485 * Convert them to a single slice on gen4.
487 if (batch
->blorp
->isl_dev
->info
->gen
== 4 &&
488 (params
.dst
.surf
.usage
& ISL_SURF_USAGE_CUBE_BIT
)) {
489 blorp_surf_convert_to_single_slice(batch
->blorp
->isl_dev
, ¶ms
.dst
);
492 if (clear_rgb_as_red
) {
493 surf_fake_rgb_with_red(batch
->blorp
->isl_dev
, ¶ms
.dst
);
498 if (isl_format_is_compressed(params
.dst
.surf
.format
)) {
499 blorp_surf_convert_to_uncompressed(batch
->blorp
->isl_dev
, ¶ms
.dst
,
500 NULL
, NULL
, NULL
, NULL
);
501 //&dst_x, &dst_y, &dst_w, &dst_h);
504 if (params
.dst
.tile_x_sa
|| params
.dst
.tile_y_sa
) {
505 /* Either we're on gen4 where there is no multisampling or the
506 * surface is compressed which also implies no multisampling.
507 * Therefore, sa == px and we don't need to do a conversion.
509 assert(params
.dst
.surf
.samples
== 1);
510 params
.x0
+= params
.dst
.tile_x_sa
;
511 params
.y0
+= params
.dst
.tile_y_sa
;
512 params
.x1
+= params
.dst
.tile_x_sa
;
513 params
.y1
+= params
.dst
.tile_y_sa
;
516 params
.num_samples
= params
.dst
.surf
.samples
;
518 /* We may be restricted on the number of layers we can bind at any one
519 * time. In particular, Sandy Bridge has a maximum number of layers of
520 * 512 but a maximum 3D texture size is much larger.
522 params
.num_layers
= MIN2(params
.dst
.view
.array_len
, num_layers
);
524 const unsigned max_image_width
= 16 * 1024;
525 if (params
.dst
.surf
.logical_level0_px
.width
> max_image_width
) {
526 /* Clearing an RGB image as red multiplies the surface width by 3
527 * so it may now be too wide for the hardware surface limits. We
528 * have to break the clear up into pieces in order to clear wide
531 assert(clear_rgb_as_red
);
532 assert(params
.dst
.surf
.dim
== ISL_SURF_DIM_2D
);
533 assert(params
.dst
.surf
.tiling
== ISL_TILING_LINEAR
);
534 assert(params
.dst
.surf
.logical_level0_px
.depth
== 1);
535 assert(params
.dst
.surf
.logical_level0_px
.array_len
== 1);
536 assert(params
.dst
.surf
.levels
== 1);
537 assert(params
.dst
.surf
.samples
== 1);
538 assert(params
.dst
.tile_x_sa
== 0 || params
.dst
.tile_y_sa
== 0);
539 assert(params
.dst
.aux_usage
== ISL_AUX_USAGE_NONE
);
541 /* max_image_width rounded down to a multiple of 3 */
542 const unsigned max_fake_rgb_width
= (max_image_width
/ 3) * 3;
544 isl_format_get_layout(params
.dst
.surf
.format
)->bpb
/ 8;
546 params
.dst
.surf
.logical_level0_px
.width
= max_fake_rgb_width
;
547 params
.dst
.surf
.phys_level0_sa
.width
= max_fake_rgb_width
;
549 uint32_t orig_x0
= params
.x0
, orig_x1
= params
.x1
;
550 uint64_t orig_offset
= params
.dst
.addr
.offset
;
551 for (uint32_t x
= orig_x0
; x
< orig_x1
; x
+= max_fake_rgb_width
) {
552 /* Offset to the surface. It's easy because we're linear */
553 params
.dst
.addr
.offset
= orig_offset
+ x
* cpp
;
556 params
.x1
= MIN2(orig_x1
- x
, max_image_width
);
558 batch
->blorp
->exec(batch
, ¶ms
);
561 batch
->blorp
->exec(batch
, ¶ms
);
564 start_layer
+= params
.num_layers
;
565 num_layers
-= params
.num_layers
;
570 blorp_clear_depth_stencil(struct blorp_batch
*batch
,
571 const struct blorp_surf
*depth
,
572 const struct blorp_surf
*stencil
,
573 uint32_t level
, uint32_t start_layer
,
575 uint32_t x0
, uint32_t y0
, uint32_t x1
, uint32_t y1
,
576 bool clear_depth
, float depth_value
,
577 uint8_t stencil_mask
, uint8_t stencil_value
)
579 struct blorp_params params
;
580 blorp_params_init(¶ms
);
587 if (ISL_DEV_GEN(batch
->blorp
->isl_dev
) == 6) {
588 /* For some reason, Sandy Bridge gets occlusion queries wrong if we
589 * don't have a shader. In particular, it records samples even though
590 * we disable statistics in 3DSTATE_WM. Give it the usual clear shader
591 * to work around the issue.
593 if (!blorp_params_get_clear_kernel(batch
, ¶ms
, false, false))
597 while (num_layers
> 0) {
598 params
.num_layers
= num_layers
;
601 brw_blorp_surface_info_init(batch
->blorp
, ¶ms
.stencil
, stencil
,
603 ISL_FORMAT_UNSUPPORTED
, true);
604 params
.stencil_mask
= stencil_mask
;
605 params
.stencil_ref
= stencil_value
;
607 params
.dst
.surf
.samples
= params
.stencil
.surf
.samples
;
608 params
.dst
.surf
.logical_level0_px
=
609 params
.stencil
.surf
.logical_level0_px
;
610 params
.dst
.view
= params
.depth
.view
;
612 params
.num_samples
= params
.stencil
.surf
.samples
;
614 /* We may be restricted on the number of layers we can bind at any
615 * one time. In particular, Sandy Bridge has a maximum number of
616 * layers of 512 but a maximum 3D texture size is much larger.
618 if (params
.stencil
.view
.array_len
< params
.num_layers
)
619 params
.num_layers
= params
.stencil
.view
.array_len
;
623 brw_blorp_surface_info_init(batch
->blorp
, ¶ms
.depth
, depth
,
625 ISL_FORMAT_UNSUPPORTED
, true);
626 params
.z
= depth_value
;
627 params
.depth_format
=
628 isl_format_get_depth_format(depth
->surf
->format
, false);
630 params
.dst
.surf
.samples
= params
.depth
.surf
.samples
;
631 params
.dst
.surf
.logical_level0_px
=
632 params
.depth
.surf
.logical_level0_px
;
633 params
.dst
.view
= params
.depth
.view
;
635 params
.num_samples
= params
.depth
.surf
.samples
;
637 /* We may be restricted on the number of layers we can bind at any
638 * one time. In particular, Sandy Bridge has a maximum number of
639 * layers of 512 but a maximum 3D texture size is much larger.
641 if (params
.depth
.view
.array_len
< params
.num_layers
)
642 params
.num_layers
= params
.depth
.view
.array_len
;
645 batch
->blorp
->exec(batch
, ¶ms
);
647 start_layer
+= params
.num_layers
;
648 num_layers
-= params
.num_layers
;
653 blorp_can_hiz_clear_depth(uint8_t gen
, enum isl_format format
,
654 uint32_t num_samples
,
655 uint32_t x0
, uint32_t y0
, uint32_t x1
, uint32_t y1
)
657 /* This function currently doesn't support any gen prior to gen8 */
660 if (gen
== 8 && format
== ISL_FORMAT_R16_UNORM
) {
661 /* Apply the D16 alignment restrictions. On BDW, HiZ has an 8x4 sample
662 * block with the following property: as the number of samples increases,
663 * the number of pixels representable by this block decreases by a factor
664 * of the sample dimensions. Sample dimensions scale following the MSAA
665 * interleaved pattern.
667 * Sample|Sample|Pixel
669 * ===================
676 * Table: Pixel Dimensions in a HiZ Sample Block Pre-SKL
678 const struct isl_extent2d sa_block_dim
=
679 isl_get_interleaved_msaa_px_size_sa(num_samples
);
680 const uint8_t align_px_w
= 8 / sa_block_dim
.w
;
681 const uint8_t align_px_h
= 4 / sa_block_dim
.h
;
683 /* Fast depth clears clear an entire sample block at a time. As a result,
684 * the rectangle must be aligned to the dimensions of the encompassing
685 * pixel block for a successful operation.
687 * Fast clears can still work if the upper-left corner is aligned and the
688 * bottom-rigtht corner touches the edge of a depth buffer whose extent
689 * is unaligned. This is because each miplevel in the depth buffer is
690 * padded by the Pixel Dim (similar to a standard compressed texture).
691 * In this case, the clear rectangle could be padded by to match the full
692 * depth buffer extent but to support multiple clearing techniques, we
693 * chose to be unaware of the depth buffer's extent and thus don't handle
696 if (x0
% align_px_w
|| y0
% align_px_h
||
697 x1
% align_px_w
|| y1
% align_px_h
)
704 blorp_hiz_clear_depth_stencil(struct blorp_batch
*batch
,
705 const struct blorp_surf
*depth
,
706 const struct blorp_surf
*stencil
,
708 uint32_t start_layer
, uint32_t num_layers
,
709 uint32_t x0
, uint32_t y0
,
710 uint32_t x1
, uint32_t y1
,
711 bool clear_depth
, float depth_value
,
712 bool clear_stencil
, uint8_t stencil_value
)
714 struct blorp_params params
;
715 blorp_params_init(¶ms
);
717 /* This requires WM_HZ_OP which only exists on gen8+ */
718 assert(ISL_DEV_GEN(batch
->blorp
->isl_dev
) >= 8);
720 params
.hiz_op
= ISL_AUX_OP_FAST_CLEAR
;
721 params
.num_layers
= 1;
728 for (uint32_t l
= 0; l
< num_layers
; l
++) {
729 const uint32_t layer
= start_layer
+ l
;
731 brw_blorp_surface_info_init(batch
->blorp
, ¶ms
.stencil
, stencil
,
733 ISL_FORMAT_UNSUPPORTED
, true);
734 params
.stencil_mask
= 0xff;
735 params
.stencil_ref
= stencil_value
;
736 params
.num_samples
= params
.stencil
.surf
.samples
;
740 /* If we're clearing depth, we must have HiZ */
741 assert(depth
&& depth
->aux_usage
== ISL_AUX_USAGE_HIZ
);
743 brw_blorp_surface_info_init(batch
->blorp
, ¶ms
.depth
, depth
,
745 ISL_FORMAT_UNSUPPORTED
, true);
746 params
.depth
.clear_color
.f32
[0] = depth_value
;
747 params
.depth_format
=
748 isl_format_get_depth_format(depth
->surf
->format
, false);
749 params
.num_samples
= params
.depth
.surf
.samples
;
752 batch
->blorp
->exec(batch
, ¶ms
);
756 /* Given a depth stencil attachment, this function performs a fast depth clear
757 * on a depth portion and a regular clear on the stencil portion. When
758 * performing a fast depth clear on the depth portion, the HiZ buffer is simply
759 * tagged as cleared so the depth clear value is not actually needed.
762 blorp_gen8_hiz_clear_attachments(struct blorp_batch
*batch
,
763 uint32_t num_samples
,
764 uint32_t x0
, uint32_t y0
,
765 uint32_t x1
, uint32_t y1
,
766 bool clear_depth
, bool clear_stencil
,
767 uint8_t stencil_value
)
769 assert(batch
->flags
& BLORP_BATCH_NO_EMIT_DEPTH_STENCIL
);
771 struct blorp_params params
;
772 blorp_params_init(¶ms
);
773 params
.num_layers
= 1;
774 params
.hiz_op
= ISL_AUX_OP_FAST_CLEAR
;
779 params
.num_samples
= num_samples
;
780 params
.depth
.enabled
= clear_depth
;
781 params
.stencil
.enabled
= clear_stencil
;
782 params
.stencil_ref
= stencil_value
;
783 batch
->blorp
->exec(batch
, ¶ms
);
786 /** Clear active color/depth/stencili attachments
788 * This function performs a clear operation on the currently bound
789 * color/depth/stencil attachments. It is assumed that any information passed
790 * in here is valid, consistent, and in-bounds relative to the currently
791 * attached depth/stencil. The binding_table_offset parameter is the 32-bit
792 * offset relative to surface state base address where pre-baked binding table
793 * that we are to use lives. If clear_color is false, binding_table_offset
794 * must point to a binding table with one entry which is a valid null surface
795 * that matches the currently bound depth and stencil.
798 blorp_clear_attachments(struct blorp_batch
*batch
,
799 uint32_t binding_table_offset
,
800 enum isl_format depth_format
,
801 uint32_t num_samples
,
802 uint32_t start_layer
, uint32_t num_layers
,
803 uint32_t x0
, uint32_t y0
, uint32_t x1
, uint32_t y1
,
804 bool clear_color
, union isl_color_value color_value
,
805 bool clear_depth
, float depth_value
,
806 uint8_t stencil_mask
, uint8_t stencil_value
)
808 struct blorp_params params
;
809 blorp_params_init(¶ms
);
811 assert(batch
->flags
& BLORP_BATCH_NO_EMIT_DEPTH_STENCIL
);
818 params
.use_pre_baked_binding_table
= true;
819 params
.pre_baked_binding_table_offset
= binding_table_offset
;
821 params
.num_layers
= num_layers
;
822 params
.num_samples
= num_samples
;
825 params
.dst
.enabled
= true;
827 memcpy(¶ms
.wm_inputs
.clear_color
, color_value
.f32
, sizeof(float) * 4);
829 /* Unfortunately, without knowing whether or not our destination surface
830 * is tiled or not, we have to assume it may be linear. This means no
831 * SIMD16_REPDATA for us. :-(
833 if (!blorp_params_get_clear_kernel(batch
, ¶ms
, false, false))
838 params
.depth
.enabled
= true;
840 params
.z
= depth_value
;
841 params
.depth_format
= isl_format_get_depth_format(depth_format
, false);
845 params
.stencil
.enabled
= true;
847 params
.stencil_mask
= stencil_mask
;
848 params
.stencil_ref
= stencil_value
;
851 if (!blorp_params_get_layer_offset_vs(batch
, ¶ms
))
854 params
.vs_inputs
.base_layer
= start_layer
;
856 batch
->blorp
->exec(batch
, ¶ms
);
860 blorp_ccs_resolve(struct blorp_batch
*batch
,
861 struct blorp_surf
*surf
, uint32_t level
,
862 uint32_t start_layer
, uint32_t num_layers
,
863 enum isl_format format
,
864 enum isl_aux_op resolve_op
)
866 struct blorp_params params
;
868 blorp_params_init(¶ms
);
869 brw_blorp_surface_info_init(batch
->blorp
, ¶ms
.dst
, surf
,
870 level
, start_layer
, format
, true);
872 /* From the Ivy Bridge PRM, Vol2 Part1 11.9 "Render Target Resolve":
874 * A rectangle primitive must be scaled down by the following factors
875 * with respect to render target being resolved.
877 * The scaledown factors in the table that follows are related to the block
878 * size of the CCS format. For IVB and HSW, we divide by two, for BDW we
879 * multiply by 8 and 16. On Sky Lake, we multiply by 8.
881 const struct isl_format_layout
*aux_fmtl
=
882 isl_format_get_layout(params
.dst
.aux_surf
.format
);
883 assert(aux_fmtl
->txc
== ISL_TXC_CCS
);
885 unsigned x_scaledown
, y_scaledown
;
886 if (ISL_DEV_GEN(batch
->blorp
->isl_dev
) >= 9) {
887 x_scaledown
= aux_fmtl
->bw
* 8;
888 y_scaledown
= aux_fmtl
->bh
* 8;
889 } else if (ISL_DEV_GEN(batch
->blorp
->isl_dev
) >= 8) {
890 x_scaledown
= aux_fmtl
->bw
* 8;
891 y_scaledown
= aux_fmtl
->bh
* 16;
893 x_scaledown
= aux_fmtl
->bw
/ 2;
894 y_scaledown
= aux_fmtl
->bh
/ 2;
896 params
.x0
= params
.y0
= 0;
897 params
.x1
= minify(params
.dst
.aux_surf
.logical_level0_px
.width
, level
);
898 params
.y1
= minify(params
.dst
.aux_surf
.logical_level0_px
.height
, level
);
899 params
.x1
= ALIGN(params
.x1
, x_scaledown
) / x_scaledown
;
900 params
.y1
= ALIGN(params
.y1
, y_scaledown
) / y_scaledown
;
902 if (batch
->blorp
->isl_dev
->info
->gen
>= 9) {
903 assert(resolve_op
== ISL_AUX_OP_FULL_RESOLVE
||
904 resolve_op
== ISL_AUX_OP_PARTIAL_RESOLVE
);
906 /* Broadwell and earlier do not have a partial resolve */
907 assert(resolve_op
== ISL_AUX_OP_FULL_RESOLVE
);
909 params
.fast_clear_op
= resolve_op
;
910 params
.num_layers
= num_layers
;
912 /* Note: there is no need to initialize push constants because it doesn't
913 * matter what data gets dispatched to the render target. However, we must
914 * ensure that the fragment shader delivers the data using the "replicated
918 if (!blorp_params_get_clear_kernel(batch
, ¶ms
, true, false))
921 batch
->blorp
->exec(batch
, ¶ms
);
925 blorp_nir_bit(nir_builder
*b
, nir_ssa_def
*src
, unsigned bit
)
927 return nir_iand(b
, nir_ushr(b
, src
, nir_imm_int(b
, bit
)),
931 struct blorp_mcs_partial_resolve_key
933 enum blorp_shader_type shader_type
;
934 bool indirect_clear_color
;
936 uint32_t num_samples
;
940 blorp_params_get_mcs_partial_resolve_kernel(struct blorp_batch
*batch
,
941 struct blorp_params
*params
)
943 struct blorp_context
*blorp
= batch
->blorp
;
944 const struct blorp_mcs_partial_resolve_key blorp_key
= {
945 .shader_type
= BLORP_SHADER_TYPE_MCS_PARTIAL_RESOLVE
,
946 .indirect_clear_color
= params
->dst
.clear_color_addr
.buffer
!= NULL
,
947 .int_format
= isl_format_has_int_channel(params
->dst
.view
.format
),
948 .num_samples
= params
->num_samples
,
951 if (blorp
->lookup_shader(batch
, &blorp_key
, sizeof(blorp_key
),
952 ¶ms
->wm_prog_kernel
, ¶ms
->wm_prog_data
))
955 void *mem_ctx
= ralloc_context(NULL
);
958 blorp_nir_init_shader(&b
, mem_ctx
, MESA_SHADER_FRAGMENT
,
959 "BLORP-mcs-partial-resolve");
961 nir_variable
*v_color
=
962 BLORP_CREATE_NIR_INPUT(b
.shader
, clear_color
, glsl_vec4_type());
964 nir_variable
*frag_color
=
965 nir_variable_create(b
.shader
, nir_var_shader_out
,
966 glsl_vec4_type(), "gl_FragColor");
967 frag_color
->data
.location
= FRAG_RESULT_COLOR
;
969 /* Do an MCS fetch and check if it is equal to the magic clear value */
971 blorp_nir_txf_ms_mcs(&b
, nir_f2i32(&b
, blorp_nir_frag_coord(&b
)),
972 nir_load_layer_id(&b
));
973 nir_ssa_def
*is_clear
=
974 blorp_nir_mcs_is_clear_color(&b
, mcs
, blorp_key
.num_samples
);
976 /* If we aren't the clear value, discard. */
977 nir_intrinsic_instr
*discard
=
978 nir_intrinsic_instr_create(b
.shader
, nir_intrinsic_discard_if
);
979 discard
->src
[0] = nir_src_for_ssa(nir_inot(&b
, is_clear
));
980 nir_builder_instr_insert(&b
, &discard
->instr
);
982 nir_ssa_def
*clear_color
= nir_load_var(&b
, v_color
);
983 if (blorp_key
.indirect_clear_color
&& blorp
->isl_dev
->info
->gen
<= 8) {
984 /* Gen7-8 clear colors are stored as single 0/1 bits */
985 clear_color
= nir_vec4(&b
, blorp_nir_bit(&b
, clear_color
, 31),
986 blorp_nir_bit(&b
, clear_color
, 30),
987 blorp_nir_bit(&b
, clear_color
, 29),
988 blorp_nir_bit(&b
, clear_color
, 28));
990 if (!blorp_key
.int_format
)
991 clear_color
= nir_i2f32(&b
, clear_color
);
993 nir_store_var(&b
, frag_color
, clear_color
, 0xf);
995 struct brw_wm_prog_key wm_key
;
996 brw_blorp_init_wm_prog_key(&wm_key
);
997 wm_key
.tex
.compressed_multisample_layout_mask
= 1;
998 wm_key
.tex
.msaa_16
= blorp_key
.num_samples
== 16;
999 wm_key
.multisample_fbo
= true;
1001 struct brw_wm_prog_data prog_data
;
1002 const unsigned *program
=
1003 blorp_compile_fs(blorp
, mem_ctx
, b
.shader
, &wm_key
, false,
1007 blorp
->upload_shader(batch
, &blorp_key
, sizeof(blorp_key
),
1008 program
, prog_data
.base
.program_size
,
1009 &prog_data
.base
, sizeof(prog_data
),
1010 ¶ms
->wm_prog_kernel
, ¶ms
->wm_prog_data
);
1012 ralloc_free(mem_ctx
);
1017 blorp_mcs_partial_resolve(struct blorp_batch
*batch
,
1018 struct blorp_surf
*surf
,
1019 enum isl_format format
,
1020 uint32_t start_layer
, uint32_t num_layers
)
1022 struct blorp_params params
;
1023 blorp_params_init(¶ms
);
1025 assert(batch
->blorp
->isl_dev
->info
->gen
>= 7);
1029 params
.x1
= surf
->surf
->logical_level0_px
.width
;
1030 params
.y1
= surf
->surf
->logical_level0_px
.height
;
1032 brw_blorp_surface_info_init(batch
->blorp
, ¶ms
.src
, surf
, 0,
1033 start_layer
, format
, false);
1034 brw_blorp_surface_info_init(batch
->blorp
, ¶ms
.dst
, surf
, 0,
1035 start_layer
, format
, true);
1037 params
.num_samples
= params
.dst
.surf
.samples
;
1038 params
.num_layers
= num_layers
;
1039 params
.dst_clear_color_as_input
= surf
->clear_color_addr
.buffer
!= NULL
;
1041 memcpy(¶ms
.wm_inputs
.clear_color
,
1042 surf
->clear_color
.f32
, sizeof(float) * 4);
1044 if (!blorp_params_get_mcs_partial_resolve_kernel(batch
, ¶ms
))
1047 batch
->blorp
->exec(batch
, ¶ms
);
1050 /** Clear a CCS to the "uncompressed" state
1052 * This pass is the CCS equivalent of a "HiZ resolve". It sets the CCS values
1053 * for a given layer/level of a surface to 0x0 which is the "uncompressed"
1054 * state which tells the sampler to go look at the main surface.
1057 blorp_ccs_ambiguate(struct blorp_batch
*batch
,
1058 struct blorp_surf
*surf
,
1059 uint32_t level
, uint32_t layer
)
1061 struct blorp_params params
;
1062 blorp_params_init(¶ms
);
1064 assert(ISL_DEV_GEN(batch
->blorp
->isl_dev
) >= 7);
1066 const struct isl_format_layout
*aux_fmtl
=
1067 isl_format_get_layout(surf
->aux_surf
->format
);
1068 assert(aux_fmtl
->txc
== ISL_TXC_CCS
);
1070 params
.dst
= (struct brw_blorp_surface_info
) {
1072 .addr
= surf
->aux_addr
,
1074 .usage
= ISL_SURF_USAGE_RENDER_TARGET_BIT
,
1075 .format
= ISL_FORMAT_R32G32B32A32_UINT
,
1077 .base_array_layer
= 0,
1080 .swizzle
= ISL_SWIZZLE_IDENTITY
,
1085 if (surf
->surf
->dim
== ISL_SURF_DIM_3D
) {
1090 uint32_t offset_B
, x_offset_el
, y_offset_el
;
1091 isl_surf_get_image_offset_el(surf
->aux_surf
, level
, layer
, z
,
1092 &x_offset_el
, &y_offset_el
);
1093 isl_tiling_get_intratile_offset_el(surf
->aux_surf
->tiling
, aux_fmtl
->bpb
,
1094 surf
->aux_surf
->row_pitch_B
,
1095 x_offset_el
, y_offset_el
,
1096 &offset_B
, &x_offset_el
, &y_offset_el
);
1097 params
.dst
.addr
.offset
+= offset_B
;
1099 const uint32_t width_px
=
1100 minify(surf
->aux_surf
->logical_level0_px
.width
, level
);
1101 const uint32_t height_px
=
1102 minify(surf
->aux_surf
->logical_level0_px
.height
, level
);
1103 const uint32_t width_el
= DIV_ROUND_UP(width_px
, aux_fmtl
->bw
);
1104 const uint32_t height_el
= DIV_ROUND_UP(height_px
, aux_fmtl
->bh
);
1106 struct isl_tile_info ccs_tile_info
;
1107 isl_surf_get_tile_info(surf
->aux_surf
, &ccs_tile_info
);
1109 /* We're going to map it as a regular RGBA32_UINT surface. We need to
1110 * downscale a good deal. We start by computing the area on the CCS to
1111 * clear in units of Y-tiled cache lines.
1113 uint32_t x_offset_cl
, y_offset_cl
, width_cl
, height_cl
;
1114 if (ISL_DEV_GEN(batch
->blorp
->isl_dev
) >= 8) {
1115 /* From the Sky Lake PRM Vol. 12 in the section on planes:
1117 * "The Color Control Surface (CCS) contains the compression status
1118 * of the cache-line pairs. The compression state of the cache-line
1119 * pair is specified by 2 bits in the CCS. Each CCS cache-line
1120 * represents an area on the main surface of 16x16 sets of 128 byte
1121 * Y-tiled cache-line-pairs. CCS is always Y tiled."
1123 * Each 2-bit surface element in the CCS corresponds to a single
1124 * cache-line pair in the main surface. This means that 16x16 el block
1125 * in the CCS maps to a Y-tiled cache line. Fortunately, CCS layouts
1126 * are calculated with a very large alignment so we can round up to a
1127 * whole cache line without worrying about overdraw.
1130 /* On Broadwell and above, a CCS tile is the same as a Y tile when
1131 * viewed at the cache-line granularity. Fortunately, the horizontal
1132 * and vertical alignment requirements of the CCS are such that we can
1133 * align to an entire cache line without worrying about crossing over
1134 * from one LOD to another.
1136 const uint32_t x_el_per_cl
= ccs_tile_info
.logical_extent_el
.w
/ 8;
1137 const uint32_t y_el_per_cl
= ccs_tile_info
.logical_extent_el
.h
/ 8;
1138 assert(surf
->aux_surf
->image_alignment_el
.w
% x_el_per_cl
== 0);
1139 assert(surf
->aux_surf
->image_alignment_el
.h
% y_el_per_cl
== 0);
1141 assert(x_offset_el
% x_el_per_cl
== 0);
1142 assert(y_offset_el
% y_el_per_cl
== 0);
1143 x_offset_cl
= x_offset_el
/ x_el_per_cl
;
1144 y_offset_cl
= y_offset_el
/ y_el_per_cl
;
1145 width_cl
= DIV_ROUND_UP(width_el
, x_el_per_cl
);
1146 height_cl
= DIV_ROUND_UP(height_el
, y_el_per_cl
);
1148 /* On gen7, the CCS tiling is not so nice. However, there we are
1149 * guaranteed that we only have a single level and slice so we don't
1150 * have to worry about it and can just align to a whole tile.
1152 assert(surf
->aux_surf
->logical_level0_px
.depth
== 1);
1153 assert(surf
->aux_surf
->logical_level0_px
.array_len
== 1);
1154 assert(x_offset_el
== 0 && y_offset_el
== 0);
1155 const uint32_t width_tl
=
1156 DIV_ROUND_UP(width_el
, ccs_tile_info
.logical_extent_el
.w
);
1157 const uint32_t height_tl
=
1158 DIV_ROUND_UP(height_el
, ccs_tile_info
.logical_extent_el
.h
);
1161 width_cl
= width_tl
* 8;
1162 height_cl
= height_tl
* 8;
1165 /* We're going to use a RGBA32 format so as to write data as quickly as
1166 * possible. A y-tiled cache line will then be 1x4 px.
1168 const uint32_t x_offset_rgba_px
= x_offset_cl
;
1169 const uint32_t y_offset_rgba_px
= y_offset_cl
* 4;
1170 const uint32_t width_rgba_px
= width_cl
;
1171 const uint32_t height_rgba_px
= height_cl
* 4;
1173 MAYBE_UNUSED
bool ok
=
1174 isl_surf_init(batch
->blorp
->isl_dev
, ¶ms
.dst
.surf
,
1175 .dim
= ISL_SURF_DIM_2D
,
1176 .format
= ISL_FORMAT_R32G32B32A32_UINT
,
1177 .width
= width_rgba_px
+ x_offset_rgba_px
,
1178 .height
= height_rgba_px
+ y_offset_rgba_px
,
1183 .row_pitch_B
= surf
->aux_surf
->row_pitch_B
,
1184 .usage
= ISL_SURF_USAGE_RENDER_TARGET_BIT
,
1185 .tiling_flags
= ISL_TILING_Y0_BIT
);
1188 params
.x0
= x_offset_rgba_px
;
1189 params
.y0
= y_offset_rgba_px
;
1190 params
.x1
= x_offset_rgba_px
+ width_rgba_px
;
1191 params
.y1
= y_offset_rgba_px
+ height_rgba_px
;
1193 /* A CCS value of 0 means "uncompressed." */
1194 memset(¶ms
.wm_inputs
.clear_color
, 0,
1195 sizeof(params
.wm_inputs
.clear_color
));
1197 if (!blorp_params_get_clear_kernel(batch
, ¶ms
, true, false))
1200 batch
->blorp
->exec(batch
, ¶ms
);