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
  35
  36 /**
  37  * Get the pointer to one element from scatter positions in memory.
  38  *
  39  * @sa lp_build_gather()
  40  */
  41 LLVMValueRef
  42 lp_build_gather_elem_ptr(LLVMBuilderRef builder,
  43                          unsigned length,
  44                          LLVMValueRef base_ptr,
  45                          LLVMValueRef offsets,
  46                          unsigned i)
  47 {
  48    LLVMValueRef offset;
  49    LLVMValueRef ptr;
  50
  51    assert(LLVMTypeOf(base_ptr) == LLVMPointerType(LLVMInt8Type(), 0));
  52
  53    if (length == 1) {
  54       assert(i == 0);
  55       offset = offsets;
  56    } else {
  57       LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), i, 0);
  58       offset = LLVMBuildExtractElement(builder, offsets, index, "");
  59    }
  60
  61    ptr = LLVMBuildGEP(builder, base_ptr, &offset, 1, "");
  62
  63    return ptr;
  64 }
  65
  66
  67 /**
  68  * Gather one element from scatter positions in memory.
  69  *
  70  * @sa lp_build_gather()
  71  */
  72 LLVMValueRef
  73 lp_build_gather_elem(LLVMBuilderRef builder,
  74                      unsigned length,
  75                      unsigned src_width,
  76                      unsigned dst_width,
  77                      LLVMValueRef base_ptr,
  78                      LLVMValueRef offsets,
  79                      unsigned i)
  80 {
  81    LLVMTypeRef src_type = LLVMIntType(src_width);
  82    LLVMTypeRef src_ptr_type = LLVMPointerType(src_type, 0);
  83    LLVMTypeRef dst_elem_type = LLVMIntType(dst_width);
  84    LLVMValueRef ptr;
  85    LLVMValueRef res;
  86
  87    assert(LLVMTypeOf(base_ptr) == LLVMPointerType(LLVMInt8Type(), 0));
  88
  89    ptr = lp_build_gather_elem_ptr(builder, length, base_ptr, offsets, i);
  90    ptr = LLVMBuildBitCast(builder, ptr, src_ptr_type, "");
  91    res = LLVMBuildLoad(builder, ptr, "");
  92
  93    assert(src_width <= dst_width);
  94    if (src_width > dst_width)
  95       res = LLVMBuildTrunc(builder, res, dst_elem_type, "");
  96    if (src_width < dst_width)
  97       res = LLVMBuildZExt(builder, res, dst_elem_type, "");
  98
  99    return res;
 100 }
 101
 102
 103 /**
 104  * Gather elements from scatter positions in memory into a single vector.
 105  * Use for fetching texels from a texture.
 106  * For SSE, typical values are length=4, src_width=32, dst_width=32.
 107  *
 108  * @param length length of the offsets
 109  * @param src_width src element width in bits
 110  * @param dst_width result element width in bits (src will be expanded to fit)
 111  * @param base_ptr base pointer, should be a i8 pointer type.
 112  * @param offsets vector with offsets
 113  */
 114 LLVMValueRef
 115 lp_build_gather(LLVMBuilderRef builder,
 116                 unsigned length,
 117                 unsigned src_width,
 118                 unsigned dst_width,
 119                 LLVMValueRef base_ptr,
 120                 LLVMValueRef offsets)
 121 {
 122    LLVMValueRef res;
 123
 124    if (length == 1) {
 125       /* Scalar */
 126       return lp_build_gather_elem(builder, length,
 127                                   src_width, dst_width,
 128                                   base_ptr, offsets, 0);
 129    } else {
 130       /* Vector */
 131
 132       LLVMTypeRef dst_elem_type = LLVMIntType(dst_width);
 133       LLVMTypeRef dst_vec_type = LLVMVectorType(dst_elem_type, length);
 134       unsigned i;
 135
 136       res = LLVMGetUndef(dst_vec_type);
 137       for (i = 0; i < length; ++i) {
 138          LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), i, 0);
 139          LLVMValueRef elem;
 140          elem = lp_build_gather_elem(builder, length,
 141                                      src_width, dst_width,
 142                                      base_ptr, offsets, i);
 143          res = LLVMBuildInsertElement(builder, res, elem, index, "");
 144       }
 145    }
 146
 147    return res;
 148 }