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"
48 #include "lp_bld_type.h"
49 #include "lp_bld_const.h"
50 #include "lp_bld_arit.h"
51 #include "lp_bld_logic.h"
52 #include "lp_bld_swizzle.h"
53 #include "lp_bld_blend.h"
54 #include "lp_bld_debug.h"
58 * We may the same values several times, so we keep them here to avoid
59 * recomputing them. Also reusing the values allows us to do simplifications
60 * that LLVM optimization passes wouldn't normally be able to do.
62 struct lp_build_blend_aos_context
64 struct lp_build_context base
;
72 LLVMValueRef inv_const
;
73 LLVMValueRef saturate
;
75 LLVMValueRef rgb_src_factor
;
76 LLVMValueRef alpha_src_factor
;
77 LLVMValueRef rgb_dst_factor
;
78 LLVMValueRef alpha_dst_factor
;
83 lp_build_blend_factor_unswizzled(struct lp_build_blend_aos_context
*bld
,
88 case PIPE_BLENDFACTOR_ZERO
:
89 return bld
->base
.zero
;
90 case PIPE_BLENDFACTOR_ONE
:
92 case PIPE_BLENDFACTOR_SRC_COLOR
:
93 case PIPE_BLENDFACTOR_SRC_ALPHA
:
95 case PIPE_BLENDFACTOR_DST_COLOR
:
96 case PIPE_BLENDFACTOR_DST_ALPHA
:
98 case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE
:
100 return bld
->base
.one
;
103 bld
->inv_dst
= lp_build_comp(&bld
->base
, bld
->dst
);
105 bld
->saturate
= lp_build_min(&bld
->base
, bld
->src
, bld
->inv_dst
);
106 return bld
->saturate
;
108 case PIPE_BLENDFACTOR_CONST_COLOR
:
109 case PIPE_BLENDFACTOR_CONST_ALPHA
:
111 case PIPE_BLENDFACTOR_SRC1_COLOR
:
112 case PIPE_BLENDFACTOR_SRC1_ALPHA
:
115 return bld
->base
.zero
;
116 case PIPE_BLENDFACTOR_INV_SRC_COLOR
:
117 case PIPE_BLENDFACTOR_INV_SRC_ALPHA
:
119 bld
->inv_src
= lp_build_comp(&bld
->base
, bld
->src
);
121 case PIPE_BLENDFACTOR_INV_DST_COLOR
:
122 case PIPE_BLENDFACTOR_INV_DST_ALPHA
:
124 bld
->inv_dst
= lp_build_comp(&bld
->base
, bld
->dst
);
126 case PIPE_BLENDFACTOR_INV_CONST_COLOR
:
127 case PIPE_BLENDFACTOR_INV_CONST_ALPHA
:
129 bld
->inv_const
= lp_build_comp(&bld
->base
, bld
->const_
);
130 return bld
->inv_const
;
131 case PIPE_BLENDFACTOR_INV_SRC1_COLOR
:
132 case PIPE_BLENDFACTOR_INV_SRC1_ALPHA
:
135 return bld
->base
.zero
;
138 return bld
->base
.zero
;
143 enum lp_build_blend_swizzle
{
144 LP_BUILD_BLEND_SWIZZLE_RGBA
= 0,
145 LP_BUILD_BLEND_SWIZZLE_AAAA
= 1
150 * How should we shuffle the base factor.
152 static enum lp_build_blend_swizzle
153 lp_build_blend_factor_swizzle(unsigned factor
)
156 case PIPE_BLENDFACTOR_ONE
:
157 case PIPE_BLENDFACTOR_ZERO
:
158 case PIPE_BLENDFACTOR_SRC_COLOR
:
159 case PIPE_BLENDFACTOR_DST_COLOR
:
160 case PIPE_BLENDFACTOR_CONST_COLOR
:
161 case PIPE_BLENDFACTOR_SRC1_COLOR
:
162 case PIPE_BLENDFACTOR_INV_SRC_COLOR
:
163 case PIPE_BLENDFACTOR_INV_DST_COLOR
:
164 case PIPE_BLENDFACTOR_INV_CONST_COLOR
:
165 case PIPE_BLENDFACTOR_INV_SRC1_COLOR
:
166 return LP_BUILD_BLEND_SWIZZLE_RGBA
;
167 case PIPE_BLENDFACTOR_SRC_ALPHA
:
168 case PIPE_BLENDFACTOR_DST_ALPHA
:
169 case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE
:
170 case PIPE_BLENDFACTOR_SRC1_ALPHA
:
171 case PIPE_BLENDFACTOR_CONST_ALPHA
:
172 case PIPE_BLENDFACTOR_INV_SRC_ALPHA
:
173 case PIPE_BLENDFACTOR_INV_DST_ALPHA
:
174 case PIPE_BLENDFACTOR_INV_CONST_ALPHA
:
175 case PIPE_BLENDFACTOR_INV_SRC1_ALPHA
:
176 return LP_BUILD_BLEND_SWIZZLE_AAAA
;
179 return LP_BUILD_BLEND_SWIZZLE_RGBA
;
185 lp_build_blend_swizzle(struct lp_build_blend_aos_context
*bld
,
188 enum lp_build_blend_swizzle rgb_swizzle
,
189 unsigned alpha_swizzle
)
192 if(rgb_swizzle
== LP_BUILD_BLEND_SWIZZLE_RGBA
)
194 if(rgb_swizzle
== LP_BUILD_BLEND_SWIZZLE_AAAA
)
195 return lp_build_broadcast_aos(&bld
->base
, rgb
, alpha_swizzle
);
198 if(rgb_swizzle
== LP_BUILD_BLEND_SWIZZLE_RGBA
) {
199 boolean cond
[4] = {0, 0, 0, 0};
200 cond
[alpha_swizzle
] = 1;
201 return lp_build_select_aos(&bld
->base
, alpha
, rgb
, cond
);
203 if(rgb_swizzle
== LP_BUILD_BLEND_SWIZZLE_AAAA
) {
204 unsigned char swizzle
[4];
205 swizzle
[0] = alpha_swizzle
;
206 swizzle
[1] = alpha_swizzle
;
207 swizzle
[2] = alpha_swizzle
;
208 swizzle
[3] = alpha_swizzle
;
209 swizzle
[alpha_swizzle
] += 4;
210 return lp_build_swizzle2_aos(&bld
->base
, rgb
, alpha
, swizzle
);
214 return bld
->base
.undef
;
219 * @sa http://www.opengl.org/sdk/docs/man/xhtml/glBlendFuncSeparate.xml
222 lp_build_blend_factor(struct lp_build_blend_aos_context
*bld
,
223 LLVMValueRef factor1
,
225 unsigned alpha_factor
,
226 unsigned alpha_swizzle
)
228 LLVMValueRef rgb_factor_
;
229 LLVMValueRef alpha_factor_
;
230 LLVMValueRef factor2
;
231 enum lp_build_blend_swizzle rgb_swizzle
;
233 rgb_factor_
= lp_build_blend_factor_unswizzled(bld
, rgb_factor
, FALSE
);
234 alpha_factor_
= lp_build_blend_factor_unswizzled(bld
, alpha_factor
, TRUE
);
236 rgb_swizzle
= lp_build_blend_factor_swizzle(rgb_factor
);
238 factor2
= lp_build_blend_swizzle(bld
, rgb_factor_
, alpha_factor_
, rgb_swizzle
, alpha_swizzle
);
240 return lp_build_mul(&bld
->base
, factor1
, factor2
);
245 lp_build_blend_func_commutative(unsigned func
)
252 case PIPE_BLEND_SUBTRACT
:
253 case PIPE_BLEND_REVERSE_SUBTRACT
:
263 lp_build_blend_func_reverse(unsigned rgb_func
, unsigned alpha_func
)
265 if(rgb_func
== alpha_func
)
267 if(rgb_func
== PIPE_BLEND_SUBTRACT
&& alpha_func
== PIPE_BLEND_REVERSE_SUBTRACT
)
269 if(rgb_func
== PIPE_BLEND_REVERSE_SUBTRACT
&& alpha_func
== PIPE_BLEND_SUBTRACT
)
276 * @sa http://www.opengl.org/sdk/docs/man/xhtml/glBlendEquationSeparate.xml
279 lp_build_blend_func(struct lp_build_context
*bld
,
286 return lp_build_add(bld
, term1
, term2
);
288 case PIPE_BLEND_SUBTRACT
:
289 return lp_build_sub(bld
, term1
, term2
);
290 case PIPE_BLEND_REVERSE_SUBTRACT
:
291 return lp_build_sub(bld
, term2
, term1
);
293 return lp_build_min(bld
, term1
, term2
);
295 return lp_build_max(bld
, term1
, term2
);
304 lp_build_blend_aos(LLVMBuilderRef builder
,
305 const struct pipe_blend_state
*blend
,
310 unsigned alpha_swizzle
)
312 struct lp_build_blend_aos_context bld
;
313 LLVMValueRef src_term
;
314 LLVMValueRef dst_term
;
317 assert(blend
->colormask
== 0xf);
319 if(!blend
->blend_enable
)
322 /* It makes no sense to blend unless values are normalized */
325 /* Setup build context */
326 memset(&bld
, 0, sizeof bld
);
327 lp_build_context_init(&bld
.base
, builder
, type
);
332 /* TODO: There are still a few optimization opportunities here. For certain
333 * combinations it is possible to reorder the operations and therefore saving
334 * some instructions. */
336 src_term
= lp_build_blend_factor(&bld
, src
, blend
->rgb_src_factor
, blend
->alpha_src_factor
, alpha_swizzle
);
337 dst_term
= lp_build_blend_factor(&bld
, dst
, blend
->rgb_dst_factor
, blend
->alpha_dst_factor
, alpha_swizzle
);
339 lp_build_name(src_term
, "src_term");
340 lp_build_name(dst_term
, "dst_term");
342 if(blend
->rgb_func
== blend
->alpha_func
) {
343 return lp_build_blend_func(&bld
.base
, blend
->rgb_func
, src_term
, dst_term
);
346 /* Seperate RGB / A functions */
351 rgb
= lp_build_blend_func(&bld
.base
, blend
->rgb_func
, src_term
, dst_term
);
352 alpha
= lp_build_blend_func(&bld
.base
, blend
->alpha_func
, src_term
, dst_term
);
354 return lp_build_blend_swizzle(&bld
, rgb
, alpha
, LP_BUILD_BLEND_SWIZZLE_RGBA
, alpha_swizzle
);