1 /**************************************************************************
3 * Copyright 2010 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 SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
15 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16 * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
17 * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM,
18 * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
19 * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
20 * USE OR OTHER DEALINGS IN THE SOFTWARE.
22 * The above copyright notice and this permission notice (including the
23 * next paragraph) shall be included in all copies or substantial portions
26 **************************************************************************/
29 #include "util/u_debug.h"
30 #include "lp_bld_debug.h"
31 #include "lp_bld_const.h"
32 #include "lp_bld_format.h"
33 #include "lp_bld_gather.h"
34 #include "lp_bld_init.h"
38 * Get the pointer to one element from scatter positions in memory.
40 * @sa lp_build_gather()
43 lp_build_gather_elem_ptr(struct gallivm_state
*gallivm
,
45 LLVMValueRef base_ptr
,
52 assert(LLVMTypeOf(base_ptr
) == LLVMPointerType(LLVMInt8TypeInContext(gallivm
->context
), 0));
58 LLVMValueRef index
= lp_build_const_int32(gallivm
, i
);
59 offset
= LLVMBuildExtractElement(gallivm
->builder
, offsets
, index
, "");
62 ptr
= LLVMBuildGEP(gallivm
->builder
, base_ptr
, &offset
, 1, "");
69 * Gather one element from scatter positions in memory.
71 * @sa lp_build_gather()
74 lp_build_gather_elem(struct gallivm_state
*gallivm
,
78 LLVMValueRef base_ptr
,
82 LLVMTypeRef src_type
= LLVMIntTypeInContext(gallivm
->context
, src_width
);
83 LLVMTypeRef src_ptr_type
= LLVMPointerType(src_type
, 0);
84 LLVMTypeRef dst_elem_type
= LLVMIntTypeInContext(gallivm
->context
, dst_width
);
88 assert(LLVMTypeOf(base_ptr
) == LLVMPointerType(LLVMInt8TypeInContext(gallivm
->context
), 0));
90 ptr
= lp_build_gather_elem_ptr(gallivm
, length
, base_ptr
, offsets
, i
);
91 ptr
= LLVMBuildBitCast(gallivm
->builder
, ptr
, src_ptr_type
, "");
92 res
= LLVMBuildLoad(gallivm
->builder
, ptr
, "");
94 assert(src_width
<= dst_width
);
95 if (src_width
> dst_width
)
96 res
= LLVMBuildTrunc(gallivm
->builder
, res
, dst_elem_type
, "");
97 if (src_width
< dst_width
)
98 res
= LLVMBuildZExt(gallivm
->builder
, res
, dst_elem_type
, "");
105 * Gather elements from scatter positions in memory into a single vector.
106 * Use for fetching texels from a texture.
107 * For SSE, typical values are length=4, src_width=32, dst_width=32.
109 * @param length length of the offsets
110 * @param src_width src element width in bits
111 * @param dst_width result element width in bits (src will be expanded to fit)
112 * @param base_ptr base pointer, should be a i8 pointer type.
113 * @param offsets vector with offsets
116 lp_build_gather(struct gallivm_state
*gallivm
,
120 LLVMValueRef base_ptr
,
121 LLVMValueRef offsets
)
127 return lp_build_gather_elem(gallivm
, length
,
128 src_width
, dst_width
,
129 base_ptr
, offsets
, 0);
133 LLVMTypeRef dst_elem_type
= LLVMIntTypeInContext(gallivm
->context
, dst_width
);
134 LLVMTypeRef dst_vec_type
= LLVMVectorType(dst_elem_type
, length
);
137 res
= LLVMGetUndef(dst_vec_type
);
138 for (i
= 0; i
< length
; ++i
) {
139 LLVMValueRef index
= lp_build_const_int32(gallivm
, i
);
141 elem
= lp_build_gather_elem(gallivm
, length
,
142 src_width
, dst_width
,
143 base_ptr
, offsets
, i
);
144 res
= LLVMBuildInsertElement(gallivm
->builder
, res
, elem
, index
, "");