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"
49 #include "gallivm/lp_bld_type.h"
50 #include "gallivm/lp_bld_const.h"
51 #include "gallivm/lp_bld_arit.h"
52 #include "gallivm/lp_bld_logic.h"
53 #include "gallivm/lp_bld_swizzle.h"
54 #include "gallivm/lp_bld_debug.h"
56 #include "lp_bld_blend.h"
60 * We may the same values several times, so we keep them here to avoid
61 * recomputing them. Also reusing the values allows us to do simplifications
62 * that LLVM optimization passes wouldn't normally be able to do.
64 struct lp_build_blend_aos_context
66 struct lp_build_context base
;
74 LLVMValueRef inv_const
;
75 LLVMValueRef saturate
;
77 LLVMValueRef rgb_src_factor
;
78 LLVMValueRef alpha_src_factor
;
79 LLVMValueRef rgb_dst_factor
;
80 LLVMValueRef alpha_dst_factor
;
85 lp_build_blend_factor_unswizzled(struct lp_build_blend_aos_context
*bld
,
90 case PIPE_BLENDFACTOR_ZERO
:
91 return bld
->base
.zero
;
92 case PIPE_BLENDFACTOR_ONE
:
94 case PIPE_BLENDFACTOR_SRC_COLOR
:
95 case PIPE_BLENDFACTOR_SRC_ALPHA
:
97 case PIPE_BLENDFACTOR_DST_COLOR
:
98 case PIPE_BLENDFACTOR_DST_ALPHA
:
100 case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE
:
102 return bld
->base
.one
;
105 bld
->inv_dst
= lp_build_comp(&bld
->base
, bld
->dst
);
107 bld
->saturate
= lp_build_min(&bld
->base
, bld
->src
, bld
->inv_dst
);
108 return bld
->saturate
;
110 case PIPE_BLENDFACTOR_CONST_COLOR
:
111 case PIPE_BLENDFACTOR_CONST_ALPHA
:
113 case PIPE_BLENDFACTOR_SRC1_COLOR
:
114 case PIPE_BLENDFACTOR_SRC1_ALPHA
:
117 return bld
->base
.zero
;
118 case PIPE_BLENDFACTOR_INV_SRC_COLOR
:
119 case PIPE_BLENDFACTOR_INV_SRC_ALPHA
:
121 bld
->inv_src
= lp_build_comp(&bld
->base
, bld
->src
);
123 case PIPE_BLENDFACTOR_INV_DST_COLOR
:
124 case PIPE_BLENDFACTOR_INV_DST_ALPHA
:
126 bld
->inv_dst
= lp_build_comp(&bld
->base
, bld
->dst
);
128 case PIPE_BLENDFACTOR_INV_CONST_COLOR
:
129 case PIPE_BLENDFACTOR_INV_CONST_ALPHA
:
131 bld
->inv_const
= lp_build_comp(&bld
->base
, bld
->const_
);
132 return bld
->inv_const
;
133 case PIPE_BLENDFACTOR_INV_SRC1_COLOR
:
134 case PIPE_BLENDFACTOR_INV_SRC1_ALPHA
:
137 return bld
->base
.zero
;
140 return bld
->base
.zero
;
145 enum lp_build_blend_swizzle
{
146 LP_BUILD_BLEND_SWIZZLE_RGBA
= 0,
147 LP_BUILD_BLEND_SWIZZLE_AAAA
= 1
152 * How should we shuffle the base factor.
154 static enum lp_build_blend_swizzle
155 lp_build_blend_factor_swizzle(unsigned factor
)
158 case PIPE_BLENDFACTOR_ONE
:
159 case PIPE_BLENDFACTOR_ZERO
:
160 case PIPE_BLENDFACTOR_SRC_COLOR
:
161 case PIPE_BLENDFACTOR_DST_COLOR
:
162 case PIPE_BLENDFACTOR_CONST_COLOR
:
163 case PIPE_BLENDFACTOR_SRC1_COLOR
:
164 case PIPE_BLENDFACTOR_INV_SRC_COLOR
:
165 case PIPE_BLENDFACTOR_INV_DST_COLOR
:
166 case PIPE_BLENDFACTOR_INV_CONST_COLOR
:
167 case PIPE_BLENDFACTOR_INV_SRC1_COLOR
:
168 return LP_BUILD_BLEND_SWIZZLE_RGBA
;
169 case PIPE_BLENDFACTOR_SRC_ALPHA
:
170 case PIPE_BLENDFACTOR_DST_ALPHA
:
171 case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE
:
172 case PIPE_BLENDFACTOR_SRC1_ALPHA
:
173 case PIPE_BLENDFACTOR_CONST_ALPHA
:
174 case PIPE_BLENDFACTOR_INV_SRC_ALPHA
:
175 case PIPE_BLENDFACTOR_INV_DST_ALPHA
:
176 case PIPE_BLENDFACTOR_INV_CONST_ALPHA
:
177 case PIPE_BLENDFACTOR_INV_SRC1_ALPHA
:
178 return LP_BUILD_BLEND_SWIZZLE_AAAA
;
181 return LP_BUILD_BLEND_SWIZZLE_RGBA
;
187 lp_build_blend_swizzle(struct lp_build_blend_aos_context
*bld
,
190 enum lp_build_blend_swizzle rgb_swizzle
,
191 unsigned alpha_swizzle
)
193 LLVMValueRef swizzled_rgb
;
195 switch (rgb_swizzle
) {
196 case LP_BUILD_BLEND_SWIZZLE_RGBA
:
199 case LP_BUILD_BLEND_SWIZZLE_AAAA
:
200 swizzled_rgb
= lp_build_broadcast_aos(&bld
->base
, rgb
, alpha_swizzle
);
204 swizzled_rgb
= bld
->base
.undef
;
208 boolean cond
[4] = {0, 0, 0, 0};
209 cond
[alpha_swizzle
] = 1;
210 swizzled_rgb
= lp_build_select_aos(&bld
->base
, alpha
, swizzled_rgb
, cond
);
218 * @sa http://www.opengl.org/sdk/docs/man/xhtml/glBlendFuncSeparate.xml
221 lp_build_blend_factor(struct lp_build_blend_aos_context
*bld
,
222 LLVMValueRef factor1
,
224 unsigned alpha_factor
,
225 unsigned alpha_swizzle
)
227 LLVMValueRef rgb_factor_
;
228 LLVMValueRef alpha_factor_
;
229 LLVMValueRef factor2
;
230 enum lp_build_blend_swizzle rgb_swizzle
;
232 rgb_factor_
= lp_build_blend_factor_unswizzled(bld
, rgb_factor
, FALSE
);
233 alpha_factor_
= lp_build_blend_factor_unswizzled(bld
, alpha_factor
, TRUE
);
235 rgb_swizzle
= lp_build_blend_factor_swizzle(rgb_factor
);
237 factor2
= lp_build_blend_swizzle(bld
, rgb_factor_
, alpha_factor_
, rgb_swizzle
, alpha_swizzle
);
239 return lp_build_mul(&bld
->base
, factor1
, factor2
);
244 * Is (a OP b) == (b OP a)?
247 lp_build_blend_func_commutative(unsigned func
)
254 case PIPE_BLEND_SUBTRACT
:
255 case PIPE_BLEND_REVERSE_SUBTRACT
:
265 lp_build_blend_func_reverse(unsigned rgb_func
, unsigned alpha_func
)
267 if(rgb_func
== alpha_func
)
269 if(rgb_func
== PIPE_BLEND_SUBTRACT
&& alpha_func
== PIPE_BLEND_REVERSE_SUBTRACT
)
271 if(rgb_func
== PIPE_BLEND_REVERSE_SUBTRACT
&& alpha_func
== PIPE_BLEND_SUBTRACT
)
278 * @sa http://www.opengl.org/sdk/docs/man/xhtml/glBlendEquationSeparate.xml
281 lp_build_blend_func(struct lp_build_context
*bld
,
288 return lp_build_add(bld
, term1
, term2
);
289 case PIPE_BLEND_SUBTRACT
:
290 return lp_build_sub(bld
, term1
, term2
);
291 case PIPE_BLEND_REVERSE_SUBTRACT
:
292 return lp_build_sub(bld
, term2
, term1
);
294 return lp_build_min(bld
, term1
, term2
);
296 return lp_build_max(bld
, term1
, term2
);
305 lp_build_blend_aos(LLVMBuilderRef builder
,
306 const struct pipe_blend_state
*blend
,
312 unsigned alpha_swizzle
)
314 struct lp_build_blend_aos_context bld
;
315 LLVMValueRef src_term
;
316 LLVMValueRef dst_term
;
318 /* FIXME: color masking not implemented yet */
319 assert(blend
->rt
[rt
].colormask
== 0xf);
321 if(!blend
->rt
[rt
].blend_enable
)
324 /* It makes no sense to blend unless values are normalized */
327 /* Setup build context */
328 memset(&bld
, 0, sizeof bld
);
329 lp_build_context_init(&bld
.base
, builder
, type
);
334 /* TODO: There are still a few optimization opportunities here. For certain
335 * combinations it is possible to reorder the operations and therefore saving
336 * some instructions. */
338 src_term
= lp_build_blend_factor(&bld
, src
, blend
->rt
[rt
].rgb_src_factor
,
339 blend
->rt
[rt
].alpha_src_factor
, alpha_swizzle
);
340 dst_term
= lp_build_blend_factor(&bld
, dst
, blend
->rt
[rt
].rgb_dst_factor
,
341 blend
->rt
[rt
].alpha_dst_factor
, alpha_swizzle
);
343 lp_build_name(src_term
, "src_term");
344 lp_build_name(dst_term
, "dst_term");
346 if(blend
->rt
[rt
].rgb_func
== blend
->rt
[rt
].alpha_func
) {
347 return lp_build_blend_func(&bld
.base
, blend
->rt
[rt
].rgb_func
, src_term
, dst_term
);
350 /* Seperate RGB / A functions */
355 rgb
= lp_build_blend_func(&bld
.base
, blend
->rt
[rt
].rgb_func
, src_term
, dst_term
);
356 alpha
= lp_build_blend_func(&bld
.base
, blend
->rt
[rt
].alpha_func
, src_term
, dst_term
);
358 return lp_build_blend_swizzle(&bld
, rgb
, alpha
, LP_BUILD_BLEND_SWIZZLE_RGBA
, alpha_swizzle
);