src/gallium/drivers/llvmpipe/lp_bld_format_soa.c

   1 /**************************************************************************
   2  *
   3  * Copyright 2009 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 above copyright notice and this permission notice (including the
  15  * next paragraph) shall be included in all copies or substantial portions
  16  * of the Software.
  17  *
  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.
  25  *
  26  **************************************************************************/
  27
  28
  29 #include "util/u_format.h"
  30
  31 #include "lp_bld_type.h"
  32 #include "lp_bld_const.h"
  33 #include "lp_bld_conv.h"
  34 #include "lp_bld_format.h"
  35
  36
  37 /**
  38  * Gather elements from scatter positions in memory into a single vector.
  39  *
  40  * @param src_width src element width
  41  * @param dst_width result element width (source will be expanded to fit)
  42  * @param length length of the offsets,
  43  * @param base_ptr base pointer, should be a i8 pointer type.
  44  * @param offsets vector with offsets
  45  */
  46 LLVMValueRef
  47 lp_build_gather(LLVMBuilderRef builder,
  48                 unsigned length,
  49                 unsigned src_width,
  50                 unsigned dst_width,
  51                 LLVMValueRef base_ptr,
  52                 LLVMValueRef offsets)
  53 {
  54    LLVMTypeRef src_type = LLVMIntType(src_width);
  55    LLVMTypeRef src_ptr_type = LLVMPointerType(src_type, 0);
  56    LLVMTypeRef dst_elem_type = LLVMIntType(dst_width);
  57    LLVMTypeRef dst_vec_type = LLVMVectorType(dst_elem_type, length);
  58    LLVMValueRef res;
  59    unsigned i;
  60
  61    res = LLVMGetUndef(dst_vec_type);
  62    for(i = 0; i < length; ++i) {
  63       LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), i, 0);
  64       LLVMValueRef elem_offset;
  65       LLVMValueRef elem_ptr;
  66       LLVMValueRef elem;
  67
  68       elem_offset = LLVMBuildExtractElement(builder, offsets, index, "");
  69       elem_ptr = LLVMBuildGEP(builder, base_ptr, &elem_offset, 1, "");
  70       elem_ptr = LLVMBuildBitCast(builder, elem_ptr, src_ptr_type, "");
  71       elem = LLVMBuildLoad(builder, elem_ptr, "");
  72
  73       assert(src_width <= dst_width);
  74       if(src_width > dst_width)
  75          elem = LLVMBuildTrunc(builder, elem, dst_elem_type, "");
  76       if(src_width < dst_width)
  77          elem = LLVMBuildZExt(builder, elem, dst_elem_type, "");
  78
  79       res = LLVMBuildInsertElement(builder, res, elem, index, "");
  80    }
  81
  82    return res;
  83 }
  84
  85
  86 static LLVMValueRef
  87 lp_build_format_swizzle(struct lp_type type,
  88                         const LLVMValueRef *inputs,
  89                         enum util_format_swizzle swizzle)
  90 {
  91    switch (swizzle) {
  92    case UTIL_FORMAT_SWIZZLE_X:
  93    case UTIL_FORMAT_SWIZZLE_Y:
  94    case UTIL_FORMAT_SWIZZLE_Z:
  95    case UTIL_FORMAT_SWIZZLE_W:
  96       return inputs[swizzle];
  97    case UTIL_FORMAT_SWIZZLE_0:
  98       return lp_build_zero(type);
  99    case UTIL_FORMAT_SWIZZLE_1:
 100       return lp_build_one(type);
 101    case UTIL_FORMAT_SWIZZLE_NONE:
 102       return lp_build_undef(type);
 103    default:
 104       assert(0);
 105       return lp_build_undef(type);
 106    }
 107 }
 108
 109
 110 void
 111 lp_build_unpack_rgba_soa(LLVMBuilderRef builder,
 112                          const struct util_format_description *format_desc,
 113                          struct lp_type type,
 114                          LLVMValueRef packed,
 115                          LLVMValueRef *rgba)
 116 {
 117    LLVMValueRef inputs[4];
 118    unsigned start;
 119    unsigned chan;
 120
 121    /* FIXME: Support more formats */
 122    assert(format_desc->layout == UTIL_FORMAT_LAYOUT_ARITH ||
 123           (format_desc->layout == UTIL_FORMAT_LAYOUT_ARRAY &&
 124            format_desc->block.bits == format_desc->channel[0].size));
 125    assert(format_desc->block.width == 1);
 126    assert(format_desc->block.height == 1);
 127    assert(format_desc->block.bits <= 32);
 128
 129    /* Decode the input vector components */
 130    start = 0;
 131    for (chan = 0; chan < 4; ++chan) {
 132       unsigned width = format_desc->channel[chan].size;
 133       unsigned stop = start + width;
 134       LLVMValueRef input;
 135
 136       input = packed;
 137
 138       switch(format_desc->channel[chan].type) {
 139       case UTIL_FORMAT_TYPE_VOID:
 140          input = NULL;
 141          break;
 142
 143       case UTIL_FORMAT_TYPE_UNSIGNED:
 144          if(type.floating) {
 145             if(start)
 146                input = LLVMBuildLShr(builder, input, lp_build_int_const_scalar(type, start), "");
 147             if(stop < format_desc->block.bits) {
 148                unsigned mask = ((unsigned long long)1 << width) - 1;
 149                input = LLVMBuildAnd(builder, input, lp_build_int_const_scalar(type, mask), "");
 150             }
 151
 152             if(format_desc->channel[chan].normalized)
 153                input = lp_build_unsigned_norm_to_float(builder, width, type, input);
 154             else
 155                input = LLVMBuildFPToSI(builder, input, lp_build_vec_type(type), "");
 156          }
 157          else {
 158             /* FIXME */
 159             assert(0);
 160             input = lp_build_undef(type);
 161          }
 162          break;
 163
 164       default:
 165          /* fall through */
 166          input = lp_build_undef(type);
 167          break;
 168       }
 169
 170       inputs[chan] = input;
 171
 172       start = stop;
 173    }
 174
 175    if(format_desc->colorspace == UTIL_FORMAT_COLORSPACE_ZS) {
 176       enum util_format_swizzle swizzle = format_desc->swizzle[0];
 177       LLVMValueRef depth = lp_build_format_swizzle(type, inputs, swizzle);
 178       rgba[2] = rgba[1] = rgba[0] = depth;
 179       rgba[3] = lp_build_one(type);
 180    }
 181    else {
 182       for (chan = 0; chan < 4; ++chan) {
 183          enum util_format_swizzle swizzle = format_desc->swizzle[chan];
 184          rgba[chan] = lp_build_format_swizzle(type, inputs, swizzle);
 185       }
 186    }
 187 }
 188
 189
 190 void
 191 lp_build_load_rgba_soa(LLVMBuilderRef builder,
 192                        const struct util_format_description *format_desc,
 193                        struct lp_type type,
 194                        LLVMValueRef base_ptr,
 195                        LLVMValueRef offsets,
 196                        LLVMValueRef *rgba)
 197 {
 198    LLVMValueRef packed;
 199
 200    assert(format_desc->block.width == 1);
 201    assert(format_desc->block.height == 1);
 202    assert(format_desc->block.bits <= type.width);
 203
 204    packed = lp_build_gather(builder,
 205                             type.length, format_desc->block.bits, type.width,
 206                             base_ptr, offsets);
 207
 208    lp_build_unpack_rgba_soa(builder, format_desc, type, packed, rgba);
 209 }