1 /**************************************************************************
3 * Copyright 2009 VMware, Inc.
6 * Permission is hereby granted, free of charge, to any person obtaining a
7 * copy of this software and associated documentation files (the
8 * "Software"), to deal in the Software without restriction, including
9 * without limitation the rights to use, copy, modify, merge, publish,
10 * distribute, sub license, and/or sell copies of the Software, and to
11 * permit persons to whom the Software is furnished to do so, subject to
12 * the following conditions:
14 * The above copyright notice and this permission notice (including the
15 * next paragraph) shall be included in all copies or substantial portions
18 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
19 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
20 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
21 * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR
22 * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
23 * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
24 * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
26 **************************************************************************/
31 * Blend LLVM IR generation -- AoS layout.
33 * AoS blending is in general much slower than SoA, but there are some cases
34 * where it might be faster. In particular, if a pixel is rendered only once
35 * then the overhead of tiling and untiling will dominate over the speedup that
36 * SoA gives. So we might want to detect such cases and fallback to AoS in the
37 * future, but for now this function is here for historical/benchmarking
40 * Run lp_blend_test after any change to this file.
42 * @author Jose Fonseca <jfonseca@vmware.com>
46 #include "pipe/p_state.h"
47 #include "util/u_debug.h"
48 #include "util/u_format.h"
50 #include "gallivm/lp_bld_type.h"
51 #include "gallivm/lp_bld_const.h"
52 #include "gallivm/lp_bld_arit.h"
53 #include "gallivm/lp_bld_logic.h"
54 #include "gallivm/lp_bld_swizzle.h"
55 #include "gallivm/lp_bld_bitarit.h"
56 #include "gallivm/lp_bld_debug.h"
58 #include "lp_bld_blend.h"
62 * We may the same values several times, so we keep them here to avoid
63 * recomputing them. Also reusing the values allows us to do simplifications
64 * that LLVM optimization passes wouldn't normally be able to do.
66 struct lp_build_blend_aos_context
68 struct lp_build_context base
;
71 LLVMValueRef src_alpha
;
74 LLVMValueRef const_alpha
;
77 LLVMValueRef inv_src_alpha
;
79 LLVMValueRef inv_const
;
80 LLVMValueRef inv_const_alpha
;
81 LLVMValueRef saturate
;
83 LLVMValueRef rgb_src_factor
;
84 LLVMValueRef alpha_src_factor
;
85 LLVMValueRef rgb_dst_factor
;
86 LLVMValueRef alpha_dst_factor
;
91 lp_build_blend_factor_unswizzled(struct lp_build_blend_aos_context
*bld
,
95 LLVMValueRef src_alpha
= bld
->src_alpha
? bld
->src_alpha
: bld
->src
;
96 LLVMValueRef const_alpha
= bld
->const_alpha
? bld
->const_alpha
: bld
->const_
;
99 case PIPE_BLENDFACTOR_ZERO
:
100 return bld
->base
.zero
;
101 case PIPE_BLENDFACTOR_ONE
:
102 return bld
->base
.one
;
103 case PIPE_BLENDFACTOR_SRC_COLOR
:
105 case PIPE_BLENDFACTOR_SRC_ALPHA
:
107 case PIPE_BLENDFACTOR_DST_COLOR
:
108 case PIPE_BLENDFACTOR_DST_ALPHA
:
110 case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE
:
112 return bld
->base
.one
;
115 bld
->inv_dst
= lp_build_comp(&bld
->base
, bld
->dst
);
117 bld
->saturate
= lp_build_min(&bld
->base
, src_alpha
, bld
->inv_dst
);
118 return bld
->saturate
;
120 case PIPE_BLENDFACTOR_CONST_COLOR
:
122 case PIPE_BLENDFACTOR_CONST_ALPHA
:
124 case PIPE_BLENDFACTOR_SRC1_COLOR
:
125 case PIPE_BLENDFACTOR_SRC1_ALPHA
:
128 return bld
->base
.zero
;
129 case PIPE_BLENDFACTOR_INV_SRC_COLOR
:
131 bld
->inv_src
= lp_build_comp(&bld
->base
, bld
->src
);
133 case PIPE_BLENDFACTOR_INV_SRC_ALPHA
:
134 if(!bld
->inv_src_alpha
)
135 bld
->inv_src_alpha
= lp_build_comp(&bld
->base
, src_alpha
);
136 return bld
->inv_src_alpha
;
137 case PIPE_BLENDFACTOR_INV_DST_COLOR
:
138 case PIPE_BLENDFACTOR_INV_DST_ALPHA
:
140 bld
->inv_dst
= lp_build_comp(&bld
->base
, bld
->dst
);
142 case PIPE_BLENDFACTOR_INV_CONST_COLOR
:
144 bld
->inv_const
= lp_build_comp(&bld
->base
, bld
->const_
);
145 return bld
->inv_const
;
146 case PIPE_BLENDFACTOR_INV_CONST_ALPHA
:
147 if(!bld
->inv_const_alpha
)
148 bld
->inv_const_alpha
= lp_build_comp(&bld
->base
, const_alpha
);
149 return bld
->inv_const_alpha
;
150 case PIPE_BLENDFACTOR_INV_SRC1_COLOR
:
151 case PIPE_BLENDFACTOR_INV_SRC1_ALPHA
:
154 return bld
->base
.zero
;
157 return bld
->base
.zero
;
162 enum lp_build_blend_swizzle
{
163 LP_BUILD_BLEND_SWIZZLE_RGBA
= 0,
164 LP_BUILD_BLEND_SWIZZLE_AAAA
= 1
169 * How should we shuffle the base factor.
171 static enum lp_build_blend_swizzle
172 lp_build_blend_factor_swizzle(unsigned factor
)
175 case PIPE_BLENDFACTOR_ONE
:
176 case PIPE_BLENDFACTOR_ZERO
:
177 case PIPE_BLENDFACTOR_SRC_COLOR
:
178 case PIPE_BLENDFACTOR_DST_COLOR
:
179 case PIPE_BLENDFACTOR_CONST_COLOR
:
180 case PIPE_BLENDFACTOR_SRC1_COLOR
:
181 case PIPE_BLENDFACTOR_INV_SRC_COLOR
:
182 case PIPE_BLENDFACTOR_INV_DST_COLOR
:
183 case PIPE_BLENDFACTOR_INV_CONST_COLOR
:
184 case PIPE_BLENDFACTOR_INV_SRC1_COLOR
:
185 return LP_BUILD_BLEND_SWIZZLE_RGBA
;
186 case PIPE_BLENDFACTOR_SRC_ALPHA
:
187 case PIPE_BLENDFACTOR_DST_ALPHA
:
188 case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE
:
189 case PIPE_BLENDFACTOR_SRC1_ALPHA
:
190 case PIPE_BLENDFACTOR_CONST_ALPHA
:
191 case PIPE_BLENDFACTOR_INV_SRC_ALPHA
:
192 case PIPE_BLENDFACTOR_INV_DST_ALPHA
:
193 case PIPE_BLENDFACTOR_INV_CONST_ALPHA
:
194 case PIPE_BLENDFACTOR_INV_SRC1_ALPHA
:
195 return LP_BUILD_BLEND_SWIZZLE_AAAA
;
198 return LP_BUILD_BLEND_SWIZZLE_RGBA
;
204 lp_build_blend_swizzle(struct lp_build_blend_aos_context
*bld
,
207 enum lp_build_blend_swizzle rgb_swizzle
,
208 unsigned alpha_swizzle
,
209 unsigned num_channels
)
211 LLVMValueRef swizzled_rgb
;
213 switch (rgb_swizzle
) {
214 case LP_BUILD_BLEND_SWIZZLE_RGBA
:
217 case LP_BUILD_BLEND_SWIZZLE_AAAA
:
218 swizzled_rgb
= lp_build_swizzle_scalar_aos(&bld
->base
, rgb
, alpha_swizzle
, num_channels
);
222 swizzled_rgb
= bld
->base
.undef
;
226 swizzled_rgb
= lp_build_select_aos(&bld
->base
, 1 << alpha_swizzle
,
235 * @sa http://www.opengl.org/sdk/docs/man/xhtml/glBlendFuncSeparate.xml
238 lp_build_blend_factor(struct lp_build_blend_aos_context
*bld
,
240 unsigned alpha_factor
,
241 unsigned alpha_swizzle
,
242 unsigned num_channels
)
244 LLVMValueRef rgb_factor_
, alpha_factor_
;
245 enum lp_build_blend_swizzle rgb_swizzle
;
247 if (alpha_swizzle
== 0) {
248 return lp_build_blend_factor_unswizzled(bld
, alpha_factor
, TRUE
);
251 rgb_factor_
= lp_build_blend_factor_unswizzled(bld
, rgb_factor
, FALSE
);
253 if (alpha_swizzle
!= UTIL_FORMAT_SWIZZLE_NONE
) {
254 rgb_swizzle
= lp_build_blend_factor_swizzle(rgb_factor
);
255 alpha_factor_
= lp_build_blend_factor_unswizzled(bld
, alpha_factor
, TRUE
);
256 return lp_build_blend_swizzle(bld
, rgb_factor_
, alpha_factor_
, rgb_swizzle
, alpha_swizzle
, num_channels
);
264 * Performs blending of src and dst pixels
266 * @param blend the blend state of the shader variant
267 * @param cbuf_format format of the colour buffer
268 * @param type data type of the pixel vector
269 * @param rt render target index
270 * @param src blend src
271 * @param dst blend dst
272 * @param mask optional mask to apply to the blending result
273 * @param const_ const blend color
274 * @param swizzle swizzle values for RGBA
276 * @return the result of blending src and dst
279 lp_build_blend_aos(struct gallivm_state
*gallivm
,
280 const struct pipe_blend_state
*blend
,
281 enum pipe_format cbuf_format
,
285 LLVMValueRef src_alpha
,
289 LLVMValueRef const_alpha
,
290 const unsigned char swizzle
[4],
293 const struct pipe_rt_blend_state
* state
= &blend
->rt
[rt
];
294 const struct util_format_description
* desc
;
295 struct lp_build_blend_aos_context bld
;
296 LLVMValueRef src_factor
, dst_factor
;
298 unsigned alpha_swizzle
= UTIL_FORMAT_SWIZZLE_NONE
;
301 desc
= util_format_description(cbuf_format
);
303 /* Setup build context */
304 memset(&bld
, 0, sizeof bld
);
305 lp_build_context_init(&bld
.base
, gallivm
, type
);
309 bld
.src_alpha
= src_alpha
;
310 bld
.const_alpha
= const_alpha
;
312 /* Find the alpha channel if not provided seperately */
314 for (i
= 0; i
< 4; ++i
) {
315 if (swizzle
[i
] == 3) {
321 if (blend
->logicop_enable
) {
323 result
= lp_build_logicop(gallivm
->builder
, blend
->logicop_func
, src
, dst
);
328 } else if (!state
->blend_enable
) {
331 boolean rgb_alpha_same
= (state
->rgb_src_factor
== state
->rgb_dst_factor
&& state
->alpha_src_factor
== state
->alpha_dst_factor
) || nr_channels
== 1;
333 src_factor
= lp_build_blend_factor(&bld
, state
->rgb_src_factor
,
334 state
->alpha_src_factor
,
338 dst_factor
= lp_build_blend_factor(&bld
, state
->rgb_dst_factor
,
339 state
->alpha_dst_factor
,
343 result
= lp_build_blend(&bld
.base
,
345 state
->rgb_src_factor
,
346 state
->rgb_dst_factor
,
354 if(state
->rgb_func
!= state
->alpha_func
&& nr_channels
> 1 && alpha_swizzle
!= UTIL_FORMAT_SWIZZLE_NONE
) {
357 alpha
= lp_build_blend(&bld
.base
,
359 state
->alpha_src_factor
,
360 state
->alpha_dst_factor
,
368 result
= lp_build_blend_swizzle(&bld
,
371 LP_BUILD_BLEND_SWIZZLE_RGBA
,
377 /* Check if color mask is necessary */
378 if (!util_format_colormask_full(desc
, state
->colormask
)) {
379 LLVMValueRef color_mask
;
381 color_mask
= lp_build_const_mask_aos_swizzled(gallivm
, bld
.base
.type
, state
->colormask
, nr_channels
, swizzle
);
382 lp_build_name(color_mask
, "color_mask");
384 /* Combine with input mask if necessary */
386 /* We can be blending floating values but masks are always integer... */
387 unsigned floating
= bld
.base
.type
.floating
;
388 bld
.base
.type
.floating
= 0;
390 mask
= lp_build_and(&bld
.base
, color_mask
, mask
);
392 bld
.base
.type
.floating
= floating
;
398 /* Apply mask, if one exists */
400 result
= lp_build_select(&bld
.base
, mask
, result
, dst
);