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 * Unit tests for blend LLVM IR generation
33 * @author Jose Fonseca <jfonseca@vmware.com>
35 * Blend computation code derived from code written by
36 * @author Brian Paul <brian@vmware.com>
40 #include "gallivm/lp_bld_init.h"
41 #include "gallivm/lp_bld_type.h"
42 #include "gallivm/lp_bld_debug.h"
43 #include "lp_bld_blend.h"
54 typedef void (*blend_test_ptr_t
)(const void *src
, const void *dst
, const void *con
, void *res
);
57 static blend_test_ptr_t
58 voidptr_to_blend_test_ptr_t(void *p
)
71 write_tsv_header(FILE *fp
)
75 "cycles_per_channel\t"
86 "alpha_dst_factor\n");
93 write_tsv_row(FILE *fp
,
94 const struct pipe_blend_state
*blend
,
95 enum vector_mode mode
,
100 fprintf(fp
, "%s\t", success
? "pass" : "fail");
103 fprintf(fp
, "%.1f\t", cycles
/ type
.length
);
104 fprintf(fp
, "aos\t");
108 fprintf(fp
, "%.1f\t", cycles
/ (4 * type
.length
));
109 fprintf(fp
, "soa\t");
112 fprintf(fp
, "%s%u%sx%u\t",
113 type
.floating
? "f" : (type
.fixed
? "h" : (type
.sign
? "s" : "u")),
115 type
.norm
? "n" : "",
120 blend
->rt
[0].rgb_func
!= blend
->rt
[0].alpha_func
? "true" : "false",
121 blend
->rt
[0].rgb_src_factor
!= blend
->rt
[0].alpha_src_factor
? "true" : "false",
122 blend
->rt
[0].rgb_dst_factor
!= blend
->rt
[0].alpha_dst_factor
? "true" : "false");
125 "%s\t%s\t%s\t%s\t%s\t%s\n",
126 util_dump_blend_func(blend
->rt
[0].rgb_func
, TRUE
),
127 util_dump_blend_factor(blend
->rt
[0].rgb_src_factor
, TRUE
),
128 util_dump_blend_factor(blend
->rt
[0].rgb_dst_factor
, TRUE
),
129 util_dump_blend_func(blend
->rt
[0].alpha_func
, TRUE
),
130 util_dump_blend_factor(blend
->rt
[0].alpha_src_factor
, TRUE
),
131 util_dump_blend_factor(blend
->rt
[0].alpha_dst_factor
, TRUE
));
138 dump_blend_type(FILE *fp
,
139 const struct pipe_blend_state
*blend
,
140 enum vector_mode mode
,
143 fprintf(fp
, "%s", mode
? "soa" : "aos");
145 fprintf(fp
, " type=%s%u%sx%u",
146 type
.floating
? "f" : (type
.fixed
? "h" : (type
.sign
? "s" : "u")),
148 type
.norm
? "n" : "",
152 " %s=%s %s=%s %s=%s %s=%s %s=%s %s=%s",
153 "rgb_func", util_dump_blend_func(blend
->rt
[0].rgb_func
, TRUE
),
154 "rgb_src_factor", util_dump_blend_factor(blend
->rt
[0].rgb_src_factor
, TRUE
),
155 "rgb_dst_factor", util_dump_blend_factor(blend
->rt
[0].rgb_dst_factor
, TRUE
),
156 "alpha_func", util_dump_blend_func(blend
->rt
[0].alpha_func
, TRUE
),
157 "alpha_src_factor", util_dump_blend_factor(blend
->rt
[0].alpha_src_factor
, TRUE
),
158 "alpha_dst_factor", util_dump_blend_factor(blend
->rt
[0].alpha_dst_factor
, TRUE
));
160 fprintf(fp
, " ...\n");
166 add_blend_test(LLVMModuleRef module
,
167 const struct pipe_blend_state
*blend
,
168 enum vector_mode mode
,
171 LLVMTypeRef vec_type
;
174 LLVMValueRef src_ptr
;
175 LLVMValueRef dst_ptr
;
176 LLVMValueRef const_ptr
;
177 LLVMValueRef res_ptr
;
178 LLVMBasicBlockRef block
;
179 LLVMBuilderRef builder
;
180 const unsigned rt
= 0;
182 vec_type
= lp_build_vec_type(type
);
184 args
[3] = args
[2] = args
[1] = args
[0] = LLVMPointerType(vec_type
, 0);
185 func
= LLVMAddFunction(module
, "test", LLVMFunctionType(LLVMVoidType(), args
, 4, 0));
186 LLVMSetFunctionCallConv(func
, LLVMCCallConv
);
187 src_ptr
= LLVMGetParam(func
, 0);
188 dst_ptr
= LLVMGetParam(func
, 1);
189 const_ptr
= LLVMGetParam(func
, 2);
190 res_ptr
= LLVMGetParam(func
, 3);
192 block
= LLVMAppendBasicBlock(func
, "entry");
193 builder
= LLVMCreateBuilder();
194 LLVMPositionBuilderAtEnd(builder
, block
);
202 src
= LLVMBuildLoad(builder
, src_ptr
, "src");
203 dst
= LLVMBuildLoad(builder
, dst_ptr
, "dst");
204 con
= LLVMBuildLoad(builder
, const_ptr
, "const");
206 res
= lp_build_blend_aos(builder
, blend
, type
, rt
, src
, dst
, con
, 3);
208 lp_build_name(res
, "res");
210 LLVMBuildStore(builder
, res
, res_ptr
);
220 for(i
= 0; i
< 4; ++i
) {
221 LLVMValueRef index
= LLVMConstInt(LLVMInt32Type(), i
, 0);
222 src
[i
] = LLVMBuildLoad(builder
, LLVMBuildGEP(builder
, src_ptr
, &index
, 1, ""), "");
223 dst
[i
] = LLVMBuildLoad(builder
, LLVMBuildGEP(builder
, dst_ptr
, &index
, 1, ""), "");
224 con
[i
] = LLVMBuildLoad(builder
, LLVMBuildGEP(builder
, const_ptr
, &index
, 1, ""), "");
225 lp_build_name(src
[i
], "src.%c", "rgba"[i
]);
226 lp_build_name(con
[i
], "con.%c", "rgba"[i
]);
227 lp_build_name(dst
[i
], "dst.%c", "rgba"[i
]);
230 lp_build_blend_soa(builder
, blend
, type
, rt
, src
, dst
, con
, res
);
232 for(i
= 0; i
< 4; ++i
) {
233 LLVMValueRef index
= LLVMConstInt(LLVMInt32Type(), i
, 0);
234 lp_build_name(res
[i
], "res.%c", "rgba"[i
]);
235 LLVMBuildStore(builder
, res
[i
], LLVMBuildGEP(builder
, res_ptr
, &index
, 1, ""));
239 LLVMBuildRetVoid(builder
);;
241 LLVMDisposeBuilder(builder
);
246 /** Add and limit result to ceiling of 1.0 */
247 #define ADD_SAT(R, A, B) \
249 R = (A) + (B); if (R > 1.0f) R = 1.0f; \
252 /** Subtract and limit result to floor of 0.0 */
253 #define SUB_SAT(R, A, B) \
255 R = (A) - (B); if (R < 0.0f) R = 0.0f; \
260 compute_blend_ref_term(unsigned rgb_factor
,
261 unsigned alpha_factor
,
262 const double *factor
,
270 switch (rgb_factor
) {
271 case PIPE_BLENDFACTOR_ONE
:
272 term
[0] = factor
[0]; /* R */
273 term
[1] = factor
[1]; /* G */
274 term
[2] = factor
[2]; /* B */
276 case PIPE_BLENDFACTOR_SRC_COLOR
:
277 term
[0] = factor
[0] * src
[0]; /* R */
278 term
[1] = factor
[1] * src
[1]; /* G */
279 term
[2] = factor
[2] * src
[2]; /* B */
281 case PIPE_BLENDFACTOR_SRC_ALPHA
:
282 term
[0] = factor
[0] * src
[3]; /* R */
283 term
[1] = factor
[1] * src
[3]; /* G */
284 term
[2] = factor
[2] * src
[3]; /* B */
286 case PIPE_BLENDFACTOR_DST_COLOR
:
287 term
[0] = factor
[0] * dst
[0]; /* R */
288 term
[1] = factor
[1] * dst
[1]; /* G */
289 term
[2] = factor
[2] * dst
[2]; /* B */
291 case PIPE_BLENDFACTOR_DST_ALPHA
:
292 term
[0] = factor
[0] * dst
[3]; /* R */
293 term
[1] = factor
[1] * dst
[3]; /* G */
294 term
[2] = factor
[2] * dst
[3]; /* B */
296 case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE
:
297 temp
= MIN2(src
[3], 1.0f
- dst
[3]);
298 term
[0] = factor
[0] * temp
; /* R */
299 term
[1] = factor
[1] * temp
; /* G */
300 term
[2] = factor
[2] * temp
; /* B */
302 case PIPE_BLENDFACTOR_CONST_COLOR
:
303 term
[0] = factor
[0] * con
[0]; /* R */
304 term
[1] = factor
[1] * con
[1]; /* G */
305 term
[2] = factor
[2] * con
[2]; /* B */
307 case PIPE_BLENDFACTOR_CONST_ALPHA
:
308 term
[0] = factor
[0] * con
[3]; /* R */
309 term
[1] = factor
[1] * con
[3]; /* G */
310 term
[2] = factor
[2] * con
[3]; /* B */
312 case PIPE_BLENDFACTOR_SRC1_COLOR
:
313 assert(0); /* to do */
315 case PIPE_BLENDFACTOR_SRC1_ALPHA
:
316 assert(0); /* to do */
318 case PIPE_BLENDFACTOR_ZERO
:
319 term
[0] = 0.0f
; /* R */
320 term
[1] = 0.0f
; /* G */
321 term
[2] = 0.0f
; /* B */
323 case PIPE_BLENDFACTOR_INV_SRC_COLOR
:
324 term
[0] = factor
[0] * (1.0f
- src
[0]); /* R */
325 term
[1] = factor
[1] * (1.0f
- src
[1]); /* G */
326 term
[2] = factor
[2] * (1.0f
- src
[2]); /* B */
328 case PIPE_BLENDFACTOR_INV_SRC_ALPHA
:
329 term
[0] = factor
[0] * (1.0f
- src
[3]); /* R */
330 term
[1] = factor
[1] * (1.0f
- src
[3]); /* G */
331 term
[2] = factor
[2] * (1.0f
- src
[3]); /* B */
333 case PIPE_BLENDFACTOR_INV_DST_ALPHA
:
334 term
[0] = factor
[0] * (1.0f
- dst
[3]); /* R */
335 term
[1] = factor
[1] * (1.0f
- dst
[3]); /* G */
336 term
[2] = factor
[2] * (1.0f
- dst
[3]); /* B */
338 case PIPE_BLENDFACTOR_INV_DST_COLOR
:
339 term
[0] = factor
[0] * (1.0f
- dst
[0]); /* R */
340 term
[1] = factor
[1] * (1.0f
- dst
[1]); /* G */
341 term
[2] = factor
[2] * (1.0f
- dst
[2]); /* B */
343 case PIPE_BLENDFACTOR_INV_CONST_COLOR
:
344 term
[0] = factor
[0] * (1.0f
- con
[0]); /* R */
345 term
[1] = factor
[1] * (1.0f
- con
[1]); /* G */
346 term
[2] = factor
[2] * (1.0f
- con
[2]); /* B */
348 case PIPE_BLENDFACTOR_INV_CONST_ALPHA
:
349 term
[0] = factor
[0] * (1.0f
- con
[3]); /* R */
350 term
[1] = factor
[1] * (1.0f
- con
[3]); /* G */
351 term
[2] = factor
[2] * (1.0f
- con
[3]); /* B */
353 case PIPE_BLENDFACTOR_INV_SRC1_COLOR
:
354 assert(0); /* to do */
356 case PIPE_BLENDFACTOR_INV_SRC1_ALPHA
:
357 assert(0); /* to do */
364 * Compute src/first term A
366 switch (alpha_factor
) {
367 case PIPE_BLENDFACTOR_ONE
:
368 term
[3] = factor
[3]; /* A */
370 case PIPE_BLENDFACTOR_SRC_COLOR
:
371 case PIPE_BLENDFACTOR_SRC_ALPHA
:
372 term
[3] = factor
[3] * src
[3]; /* A */
374 case PIPE_BLENDFACTOR_DST_COLOR
:
375 case PIPE_BLENDFACTOR_DST_ALPHA
:
376 term
[3] = factor
[3] * dst
[3]; /* A */
378 case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE
:
379 term
[3] = src
[3]; /* A */
381 case PIPE_BLENDFACTOR_CONST_COLOR
:
382 case PIPE_BLENDFACTOR_CONST_ALPHA
:
383 term
[3] = factor
[3] * con
[3]; /* A */
385 case PIPE_BLENDFACTOR_ZERO
:
386 term
[3] = 0.0f
; /* A */
388 case PIPE_BLENDFACTOR_INV_SRC_COLOR
:
389 case PIPE_BLENDFACTOR_INV_SRC_ALPHA
:
390 term
[3] = factor
[3] * (1.0f
- src
[3]); /* A */
392 case PIPE_BLENDFACTOR_INV_DST_COLOR
:
393 case PIPE_BLENDFACTOR_INV_DST_ALPHA
:
394 term
[3] = factor
[3] * (1.0f
- dst
[3]); /* A */
396 case PIPE_BLENDFACTOR_INV_CONST_COLOR
:
397 case PIPE_BLENDFACTOR_INV_CONST_ALPHA
:
398 term
[3] = factor
[3] * (1.0f
- con
[3]);
407 compute_blend_ref(const struct pipe_blend_state
*blend
,
416 compute_blend_ref_term(blend
->rt
[0].rgb_src_factor
, blend
->rt
[0].alpha_src_factor
,
417 src
, src
, dst
, con
, src_term
);
418 compute_blend_ref_term(blend
->rt
[0].rgb_dst_factor
, blend
->rt
[0].alpha_dst_factor
,
419 dst
, src
, dst
, con
, dst_term
);
424 switch (blend
->rt
[0].rgb_func
) {
426 ADD_SAT(res
[0], src_term
[0], dst_term
[0]); /* R */
427 ADD_SAT(res
[1], src_term
[1], dst_term
[1]); /* G */
428 ADD_SAT(res
[2], src_term
[2], dst_term
[2]); /* B */
430 case PIPE_BLEND_SUBTRACT
:
431 SUB_SAT(res
[0], src_term
[0], dst_term
[0]); /* R */
432 SUB_SAT(res
[1], src_term
[1], dst_term
[1]); /* G */
433 SUB_SAT(res
[2], src_term
[2], dst_term
[2]); /* B */
435 case PIPE_BLEND_REVERSE_SUBTRACT
:
436 SUB_SAT(res
[0], dst_term
[0], src_term
[0]); /* R */
437 SUB_SAT(res
[1], dst_term
[1], src_term
[1]); /* G */
438 SUB_SAT(res
[2], dst_term
[2], src_term
[2]); /* B */
441 res
[0] = MIN2(src_term
[0], dst_term
[0]); /* R */
442 res
[1] = MIN2(src_term
[1], dst_term
[1]); /* G */
443 res
[2] = MIN2(src_term
[2], dst_term
[2]); /* B */
446 res
[0] = MAX2(src_term
[0], dst_term
[0]); /* R */
447 res
[1] = MAX2(src_term
[1], dst_term
[1]); /* G */
448 res
[2] = MAX2(src_term
[2], dst_term
[2]); /* B */
457 switch (blend
->rt
[0].alpha_func
) {
459 ADD_SAT(res
[3], src_term
[3], dst_term
[3]); /* A */
461 case PIPE_BLEND_SUBTRACT
:
462 SUB_SAT(res
[3], src_term
[3], dst_term
[3]); /* A */
464 case PIPE_BLEND_REVERSE_SUBTRACT
:
465 SUB_SAT(res
[3], dst_term
[3], src_term
[3]); /* A */
468 res
[3] = MIN2(src_term
[3], dst_term
[3]); /* A */
471 res
[3] = MAX2(src_term
[3], dst_term
[3]); /* A */
481 test_one(unsigned verbose
,
483 const struct pipe_blend_state
*blend
,
484 enum vector_mode mode
,
487 LLVMModuleRef module
= NULL
;
488 LLVMValueRef func
= NULL
;
489 LLVMExecutionEngineRef engine
= lp_build_engine
;
490 LLVMPassManagerRef pass
= NULL
;
492 blend_test_ptr_t blend_test_ptr
;
494 const unsigned n
= LP_TEST_NUM_SAMPLES
;
495 int64_t cycles
[LP_TEST_NUM_SAMPLES
];
496 double cycles_avg
= 0.0;
501 dump_blend_type(stdout
, blend
, mode
, type
);
503 module
= LLVMModuleCreateWithName("test");
505 func
= add_blend_test(module
, blend
, mode
, type
);
507 if(LLVMVerifyModule(module
, LLVMPrintMessageAction
, &error
)) {
508 LLVMDumpModule(module
);
511 LLVMDisposeMessage(error
);
514 pass
= LLVMCreatePassManager();
515 LLVMAddTargetData(LLVMGetExecutionEngineTargetData(engine
), pass
);
516 /* These are the passes currently listed in llvm-c/Transforms/Scalar.h,
517 * but there are more on SVN. */
518 LLVMAddConstantPropagationPass(pass
);
519 LLVMAddInstructionCombiningPass(pass
);
520 LLVMAddPromoteMemoryToRegisterPass(pass
);
521 LLVMAddGVNPass(pass
);
522 LLVMAddCFGSimplificationPass(pass
);
523 LLVMRunPassManager(pass
, module
);
529 LLVMDumpModule(module
);
531 code
= LLVMGetPointerToGlobal(engine
, func
);
532 blend_test_ptr
= voidptr_to_blend_test_ptr_t(code
);
535 lp_disassemble(code
);
538 for(i
= 0; i
< n
&& success
; ++i
) {
540 PIPE_ALIGN_VAR(16) uint8_t src
[LP_NATIVE_VECTOR_WIDTH
/8];
541 PIPE_ALIGN_VAR(16) uint8_t dst
[LP_NATIVE_VECTOR_WIDTH
/8];
542 PIPE_ALIGN_VAR(16) uint8_t con
[LP_NATIVE_VECTOR_WIDTH
/8];
543 PIPE_ALIGN_VAR(16) uint8_t res
[LP_NATIVE_VECTOR_WIDTH
/8];
544 PIPE_ALIGN_VAR(16) uint8_t ref
[LP_NATIVE_VECTOR_WIDTH
/8];
545 int64_t start_counter
= 0;
546 int64_t end_counter
= 0;
548 random_vec(type
, src
);
549 random_vec(type
, dst
);
550 random_vec(type
, con
);
553 double fsrc
[LP_MAX_VECTOR_LENGTH
];
554 double fdst
[LP_MAX_VECTOR_LENGTH
];
555 double fcon
[LP_MAX_VECTOR_LENGTH
];
556 double fref
[LP_MAX_VECTOR_LENGTH
];
558 read_vec(type
, src
, fsrc
);
559 read_vec(type
, dst
, fdst
);
560 read_vec(type
, con
, fcon
);
562 for(j
= 0; j
< type
.length
; j
+= 4)
563 compute_blend_ref(blend
, fsrc
+ j
, fdst
+ j
, fcon
+ j
, fref
+ j
);
565 write_vec(type
, ref
, fref
);
568 start_counter
= rdtsc();
569 blend_test_ptr(src
, dst
, con
, res
);
570 end_counter
= rdtsc();
572 cycles
[i
] = end_counter
- start_counter
;
574 if(!compare_vec(type
, res
, ref
)) {
578 dump_blend_type(stderr
, blend
, mode
, type
);
579 fprintf(stderr
, "MISMATCH\n");
581 fprintf(stderr
, " Src: ");
582 dump_vec(stderr
, type
, src
);
583 fprintf(stderr
, "\n");
585 fprintf(stderr
, " Dst: ");
586 dump_vec(stderr
, type
, dst
);
587 fprintf(stderr
, "\n");
589 fprintf(stderr
, " Con: ");
590 dump_vec(stderr
, type
, con
);
591 fprintf(stderr
, "\n");
593 fprintf(stderr
, " Res: ");
594 dump_vec(stderr
, type
, res
);
595 fprintf(stderr
, "\n");
597 fprintf(stderr
, " Ref: ");
598 dump_vec(stderr
, type
, ref
);
599 fprintf(stderr
, "\n");
604 const unsigned stride
= type
.length
*type
.width
/8;
605 PIPE_ALIGN_VAR(16) uint8_t src
[4*LP_NATIVE_VECTOR_WIDTH
/8];
606 PIPE_ALIGN_VAR(16) uint8_t dst
[4*LP_NATIVE_VECTOR_WIDTH
/8];
607 PIPE_ALIGN_VAR(16) uint8_t con
[4*LP_NATIVE_VECTOR_WIDTH
/8];
608 PIPE_ALIGN_VAR(16) uint8_t res
[4*LP_NATIVE_VECTOR_WIDTH
/8];
609 PIPE_ALIGN_VAR(16) uint8_t ref
[4*LP_NATIVE_VECTOR_WIDTH
/8];
610 int64_t start_counter
= 0;
611 int64_t end_counter
= 0;
614 for(j
= 0; j
< 4; ++j
) {
615 random_vec(type
, src
+ j
*stride
);
616 random_vec(type
, dst
+ j
*stride
);
617 random_vec(type
, con
+ j
*stride
);
627 for(k
= 0; k
< type
.length
; ++k
) {
628 for(j
= 0; j
< 4; ++j
) {
629 fsrc
[j
] = read_elem(type
, src
+ j
*stride
, k
);
630 fdst
[j
] = read_elem(type
, dst
+ j
*stride
, k
);
631 fcon
[j
] = read_elem(type
, con
+ j
*stride
, k
);
634 compute_blend_ref(blend
, fsrc
, fdst
, fcon
, fref
);
636 for(j
= 0; j
< 4; ++j
)
637 write_elem(type
, ref
+ j
*stride
, k
, fref
[j
]);
641 start_counter
= rdtsc();
642 blend_test_ptr(src
, dst
, con
, res
);
643 end_counter
= rdtsc();
645 cycles
[i
] = end_counter
- start_counter
;
648 for (j
= 0; j
< 4; ++j
)
649 if(!compare_vec(type
, res
+ j
*stride
, ref
+ j
*stride
))
656 dump_blend_type(stderr
, blend
, mode
, type
);
657 fprintf(stderr
, "MISMATCH\n");
658 for(j
= 0; j
< 4; ++j
) {
659 char channel
= "RGBA"[j
];
660 fprintf(stderr
, " Src%c: ", channel
);
661 dump_vec(stderr
, type
, src
+ j
*stride
);
662 fprintf(stderr
, "\n");
664 fprintf(stderr
, " Dst%c: ", channel
);
665 dump_vec(stderr
, type
, dst
+ j
*stride
);
666 fprintf(stderr
, "\n");
668 fprintf(stderr
, " Con%c: ", channel
);
669 dump_vec(stderr
, type
, con
+ j
*stride
);
670 fprintf(stderr
, "\n");
672 fprintf(stderr
, " Res%c: ", channel
);
673 dump_vec(stderr
, type
, res
+ j
*stride
);
674 fprintf(stderr
, "\n");
676 fprintf(stderr
, " Ref%c: ", channel
);
677 dump_vec(stderr
, type
, ref
+ j
*stride
);
678 fprintf(stderr
, "\n");
685 * Unfortunately the output of cycle counter is not very reliable as it comes
686 * -- sometimes we get outliers (due IRQs perhaps?) which are
687 * better removed to avoid random or biased data.
690 double sum
= 0.0, sum2
= 0.0;
694 for(i
= 0; i
< n
; ++i
) {
696 sum2
+= cycles
[i
]*cycles
[i
];
700 std
= sqrtf((sum2
- n
*avg
*avg
)/n
);
704 for(i
= 0; i
< n
; ++i
) {
705 if(fabs(cycles
[i
] - avg
) <= 4.0*std
) {
716 write_tsv_row(fp
, blend
, mode
, type
, cycles_avg
, success
);
720 LLVMDumpModule(module
);
721 LLVMWriteBitcodeToFile(module
, "blend.bc");
722 fprintf(stderr
, "blend.bc written\n");
723 fprintf(stderr
, "Invoke as \"llc -o - blend.bc\"\n");
727 LLVMFreeMachineCodeForFunction(engine
, func
);
730 LLVMDisposePassManager(pass
);
738 PIPE_BLENDFACTOR_ZERO
,
739 PIPE_BLENDFACTOR_ONE
,
740 PIPE_BLENDFACTOR_SRC_COLOR
,
741 PIPE_BLENDFACTOR_SRC_ALPHA
,
742 PIPE_BLENDFACTOR_DST_COLOR
,
743 PIPE_BLENDFACTOR_DST_ALPHA
,
744 PIPE_BLENDFACTOR_CONST_COLOR
,
745 PIPE_BLENDFACTOR_CONST_ALPHA
,
747 PIPE_BLENDFACTOR_SRC1_COLOR
,
748 PIPE_BLENDFACTOR_SRC1_ALPHA
,
750 PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE
,
751 PIPE_BLENDFACTOR_INV_SRC_COLOR
,
752 PIPE_BLENDFACTOR_INV_SRC_ALPHA
,
753 PIPE_BLENDFACTOR_INV_DST_COLOR
,
754 PIPE_BLENDFACTOR_INV_DST_ALPHA
,
755 PIPE_BLENDFACTOR_INV_CONST_COLOR
,
756 PIPE_BLENDFACTOR_INV_CONST_ALPHA
,
758 PIPE_BLENDFACTOR_INV_SRC1_COLOR
,
759 PIPE_BLENDFACTOR_INV_SRC1_ALPHA
,
768 PIPE_BLEND_REVERSE_SUBTRACT
,
774 const struct lp_type blend_types
[] = {
775 /* float, fixed, sign, norm, width, len */
776 { TRUE
, FALSE
, FALSE
, TRUE
, 32, 4 }, /* f32 x 4 */
777 { FALSE
, FALSE
, FALSE
, TRUE
, 8, 16 }, /* u8n x 16 */
781 const unsigned num_funcs
= sizeof(blend_funcs
)/sizeof(blend_funcs
[0]);
782 const unsigned num_factors
= sizeof(blend_factors
)/sizeof(blend_factors
[0]);
783 const unsigned num_types
= sizeof(blend_types
)/sizeof(blend_types
[0]);
787 test_all(unsigned verbose
, FILE *fp
)
789 const unsigned *rgb_func
;
790 const unsigned *rgb_src_factor
;
791 const unsigned *rgb_dst_factor
;
792 const unsigned *alpha_func
;
793 const unsigned *alpha_src_factor
;
794 const unsigned *alpha_dst_factor
;
795 struct pipe_blend_state blend
;
796 enum vector_mode mode
;
797 const struct lp_type
*type
;
798 boolean success
= TRUE
;
800 for(rgb_func
= blend_funcs
; rgb_func
< &blend_funcs
[num_funcs
]; ++rgb_func
) {
801 for(alpha_func
= blend_funcs
; alpha_func
< &blend_funcs
[num_funcs
]; ++alpha_func
) {
802 for(rgb_src_factor
= blend_factors
; rgb_src_factor
< &blend_factors
[num_factors
]; ++rgb_src_factor
) {
803 for(rgb_dst_factor
= blend_factors
; rgb_dst_factor
<= rgb_src_factor
; ++rgb_dst_factor
) {
804 for(alpha_src_factor
= blend_factors
; alpha_src_factor
< &blend_factors
[num_factors
]; ++alpha_src_factor
) {
805 for(alpha_dst_factor
= blend_factors
; alpha_dst_factor
<= alpha_src_factor
; ++alpha_dst_factor
) {
806 for(mode
= 0; mode
< 2; ++mode
) {
807 for(type
= blend_types
; type
< &blend_types
[num_types
]; ++type
) {
809 if(*rgb_dst_factor
== PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE
||
810 *alpha_dst_factor
== PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE
)
813 memset(&blend
, 0, sizeof blend
);
814 blend
.rt
[0].blend_enable
= 1;
815 blend
.rt
[0].rgb_func
= *rgb_func
;
816 blend
.rt
[0].rgb_src_factor
= *rgb_src_factor
;
817 blend
.rt
[0].rgb_dst_factor
= *rgb_dst_factor
;
818 blend
.rt
[0].alpha_func
= *alpha_func
;
819 blend
.rt
[0].alpha_src_factor
= *alpha_src_factor
;
820 blend
.rt
[0].alpha_dst_factor
= *alpha_dst_factor
;
821 blend
.rt
[0].colormask
= PIPE_MASK_RGBA
;
823 if(!test_one(verbose
, fp
, &blend
, mode
, *type
))
840 test_some(unsigned verbose
, FILE *fp
, unsigned long n
)
842 const unsigned *rgb_func
;
843 const unsigned *rgb_src_factor
;
844 const unsigned *rgb_dst_factor
;
845 const unsigned *alpha_func
;
846 const unsigned *alpha_src_factor
;
847 const unsigned *alpha_dst_factor
;
848 struct pipe_blend_state blend
;
849 enum vector_mode mode
;
850 const struct lp_type
*type
;
852 boolean success
= TRUE
;
854 for(i
= 0; i
< n
; ++i
) {
855 rgb_func
= &blend_funcs
[rand() % num_funcs
];
856 alpha_func
= &blend_funcs
[rand() % num_funcs
];
857 rgb_src_factor
= &blend_factors
[rand() % num_factors
];
858 alpha_src_factor
= &blend_factors
[rand() % num_factors
];
861 rgb_dst_factor
= &blend_factors
[rand() % num_factors
];
862 } while(*rgb_dst_factor
== PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE
);
865 alpha_dst_factor
= &blend_factors
[rand() % num_factors
];
866 } while(*alpha_dst_factor
== PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE
);
870 type
= &blend_types
[rand() % num_types
];
872 memset(&blend
, 0, sizeof blend
);
873 blend
.rt
[0].blend_enable
= 1;
874 blend
.rt
[0].rgb_func
= *rgb_func
;
875 blend
.rt
[0].rgb_src_factor
= *rgb_src_factor
;
876 blend
.rt
[0].rgb_dst_factor
= *rgb_dst_factor
;
877 blend
.rt
[0].alpha_func
= *alpha_func
;
878 blend
.rt
[0].alpha_src_factor
= *alpha_src_factor
;
879 blend
.rt
[0].alpha_dst_factor
= *alpha_dst_factor
;
880 blend
.rt
[0].colormask
= PIPE_MASK_RGBA
;
882 if(!test_one(verbose
, fp
, &blend
, mode
, *type
))
891 test_single(unsigned verbose
, FILE *fp
)
893 printf("no test_single()");