gallivm: use f16c hw support for float->half and half->float conversion
[mesa.git] / src / gallium / auxiliary / gallivm / lp_bld_gather.c
1 /**************************************************************************
2 *
3 * Copyright 2010 VMware, Inc.
4 * All Rights Reserved.
5 *
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:
13 *
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.
21 *
22 * The above copyright notice and this permission notice (including the
23 * next paragraph) shall be included in all copies or substantial portions
24 * of the Software.
25 *
26 **************************************************************************/
27
28
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"
35 #include "lp_bld_intr.h"
36
37
38 /**
39 * Get the pointer to one element from scatter positions in memory.
40 *
41 * @sa lp_build_gather()
42 */
43 LLVMValueRef
44 lp_build_gather_elem_ptr(struct gallivm_state *gallivm,
45 unsigned length,
46 LLVMValueRef base_ptr,
47 LLVMValueRef offsets,
48 unsigned i)
49 {
50 LLVMValueRef offset;
51 LLVMValueRef ptr;
52
53 assert(LLVMTypeOf(base_ptr) == LLVMPointerType(LLVMInt8TypeInContext(gallivm->context), 0));
54
55 if (length == 1) {
56 assert(i == 0);
57 offset = offsets;
58 } else {
59 LLVMValueRef index = lp_build_const_int32(gallivm, i);
60 offset = LLVMBuildExtractElement(gallivm->builder, offsets, index, "");
61 }
62
63 ptr = LLVMBuildGEP(gallivm->builder, base_ptr, &offset, 1, "");
64
65 return ptr;
66 }
67
68
69 /**
70 * Gather one element from scatter positions in memory.
71 *
72 * @sa lp_build_gather()
73 */
74 LLVMValueRef
75 lp_build_gather_elem(struct gallivm_state *gallivm,
76 unsigned length,
77 unsigned src_width,
78 unsigned dst_width,
79 LLVMValueRef base_ptr,
80 LLVMValueRef offsets,
81 unsigned i)
82 {
83 LLVMTypeRef src_type = LLVMIntTypeInContext(gallivm->context, src_width);
84 LLVMTypeRef src_ptr_type = LLVMPointerType(src_type, 0);
85 LLVMTypeRef dst_elem_type = LLVMIntTypeInContext(gallivm->context, dst_width);
86 LLVMValueRef ptr;
87 LLVMValueRef res;
88
89 assert(LLVMTypeOf(base_ptr) == LLVMPointerType(LLVMInt8TypeInContext(gallivm->context), 0));
90
91 ptr = lp_build_gather_elem_ptr(gallivm, length, base_ptr, offsets, i);
92 ptr = LLVMBuildBitCast(gallivm->builder, ptr, src_ptr_type, "");
93 res = LLVMBuildLoad(gallivm->builder, ptr, "");
94
95 assert(src_width <= dst_width);
96 if (src_width > dst_width) {
97 res = LLVMBuildTrunc(gallivm->builder, res, dst_elem_type, "");
98 } else if (src_width < dst_width) {
99 res = LLVMBuildZExt(gallivm->builder, res, dst_elem_type, "");
100 #ifdef PIPE_ARCH_BIG_ENDIAN
101 res = LLVMBuildShl(gallivm->builder, res,
102 LLVMConstInt(dst_elem_type, dst_width - src_width, 0), "");
103 #endif
104 }
105
106 return res;
107 }
108
109
110 /**
111 * Gather elements from scatter positions in memory into a single vector.
112 * Use for fetching texels from a texture.
113 * For SSE, typical values are length=4, src_width=32, dst_width=32.
114 *
115 * @param length length of the offsets
116 * @param src_width src element width in bits
117 * @param dst_width result element width in bits (src will be expanded to fit)
118 * @param base_ptr base pointer, should be a i8 pointer type.
119 * @param offsets vector with offsets
120 */
121 LLVMValueRef
122 lp_build_gather(struct gallivm_state *gallivm,
123 unsigned length,
124 unsigned src_width,
125 unsigned dst_width,
126 LLVMValueRef base_ptr,
127 LLVMValueRef offsets)
128 {
129 LLVMValueRef res;
130
131 if (length == 1) {
132 /* Scalar */
133 return lp_build_gather_elem(gallivm, length,
134 src_width, dst_width,
135 base_ptr, offsets, 0);
136 } else {
137 /* Vector */
138
139 LLVMTypeRef dst_elem_type = LLVMIntTypeInContext(gallivm->context, dst_width);
140 LLVMTypeRef dst_vec_type = LLVMVectorType(dst_elem_type, length);
141 unsigned i;
142
143 res = LLVMGetUndef(dst_vec_type);
144 for (i = 0; i < length; ++i) {
145 LLVMValueRef index = lp_build_const_int32(gallivm, i);
146 LLVMValueRef elem;
147 elem = lp_build_gather_elem(gallivm, length,
148 src_width, dst_width,
149 base_ptr, offsets, i);
150 res = LLVMBuildInsertElement(gallivm->builder, res, elem, index, "");
151 }
152 }
153
154 return res;
155 }
156
157 LLVMValueRef
158 lp_build_gather_values(struct gallivm_state * gallivm,
159 LLVMValueRef * values,
160 unsigned value_count)
161 {
162 LLVMTypeRef vec_type = LLVMVectorType(LLVMTypeOf(values[0]), value_count);
163 LLVMBuilderRef builder = gallivm->builder;
164 LLVMValueRef vec = LLVMGetUndef(vec_type);
165 unsigned i;
166
167 for (i = 0; i < value_count; i++) {
168 LLVMValueRef index = lp_build_const_int32(gallivm, i);
169 vec = LLVMBuildInsertElement(builder, vec, values[i], index, "");
170 }
171 return vec;
172 }