src/gallium/auxiliary/gallivm/lp_bld_sample.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  * @file
  30  * Texture sampling -- common code.
  31  *
  32  * @author Jose Fonseca <jfonseca@vmware.com>
  33  */
  34
  35 #include "pipe/p_defines.h"
  36 #include "pipe/p_state.h"
  37 #include "util/u_format.h"
  38 #include "util/u_math.h"
  39 #include "lp_bld_debug.h"
  40 #include "lp_bld_const.h"
  41 #include "lp_bld_arit.h"
  42 #include "lp_bld_type.h"
  43 #include "lp_bld_format.h"
  44 #include "lp_bld_sample.h"
  45
  46
  47 /**
  48  * Initialize lp_sampler_static_state object with the gallium sampler
  49  * and texture state.
  50  * The former is considered to be static and the later dynamic.
  51  */
  52 void
  53 lp_sampler_static_state(struct lp_sampler_static_state *state,
  54                         const struct pipe_texture *texture,
  55                         const struct pipe_sampler_state *sampler)
  56 {
  57    memset(state, 0, sizeof *state);
  58
  59    if(!texture)
  60       return;
  61
  62    if(!sampler)
  63       return;
  64
  65    /*
  66     * We don't copy sampler state over unless it is actually enabled, to avoid
  67     * spurious recompiles, as the sampler static state is part of the shader
  68     * key.
  69     *
  70     * Ideally the state tracker or cso_cache module would make all state
  71     * canonical, but until that happens it's better to be safe than sorry here.
  72     *
  73     * XXX: Actually there's much more than can be done here, especially
  74     * regarding 1D/2D/3D/CUBE textures, wrap modes, etc.
  75     */
  76
  77    state->format            = texture->format;
  78    state->target            = texture->target;
  79    state->pot_width         = util_is_pot(texture->width0);
  80    state->pot_height        = util_is_pot(texture->height0);
  81    state->pot_depth         = util_is_pot(texture->depth0);
  82
  83    state->wrap_s            = sampler->wrap_s;
  84    state->wrap_t            = sampler->wrap_t;
  85    state->wrap_r            = sampler->wrap_r;
  86    state->min_img_filter    = sampler->min_img_filter;
  87    state->mag_img_filter    = sampler->mag_img_filter;
  88    if (texture->last_level) {
  89       state->min_mip_filter = sampler->min_mip_filter;
  90    } else {
  91       state->min_mip_filter = PIPE_TEX_MIPFILTER_NONE;
  92    }
  93
  94    state->compare_mode      = sampler->compare_mode;
  95    if (sampler->compare_mode != PIPE_TEX_COMPARE_NONE) {
  96       state->compare_func   = sampler->compare_func;
  97    }
  98
  99    state->normalized_coords = sampler->normalized_coords;
 100    state->lod_bias          = sampler->lod_bias;
 101    state->min_lod           = sampler->min_lod;
 102    state->max_lod           = sampler->max_lod;
 103    state->border_color[0]   = sampler->border_color[0];
 104    state->border_color[1]   = sampler->border_color[1];
 105    state->border_color[2]   = sampler->border_color[2];
 106    state->border_color[3]   = sampler->border_color[3];
 107 }
 108
 109
 110 /**
 111  * Gather elements from scatter positions in memory into a single vector.
 112  *
 113  * @param src_width src element width
 114  * @param dst_width result element width (source will be expanded to fit)
 115  * @param length length of the offsets,
 116  * @param base_ptr base pointer, should be a i8 pointer type.
 117  * @param offsets vector with offsets
 118  */
 119 LLVMValueRef
 120 lp_build_gather(LLVMBuilderRef builder,
 121                 unsigned length,
 122                 unsigned src_width,
 123                 unsigned dst_width,
 124                 LLVMValueRef base_ptr,
 125                 LLVMValueRef offsets)
 126 {
 127    LLVMTypeRef src_type = LLVMIntType(src_width);
 128    LLVMTypeRef src_ptr_type = LLVMPointerType(src_type, 0);
 129    LLVMTypeRef dst_elem_type = LLVMIntType(dst_width);
 130    LLVMTypeRef dst_vec_type = LLVMVectorType(dst_elem_type, length);
 131    LLVMValueRef res;
 132    unsigned i;
 133
 134    res = LLVMGetUndef(dst_vec_type);
 135    for(i = 0; i < length; ++i) {
 136       LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), i, 0);
 137       LLVMValueRef elem_offset;
 138       LLVMValueRef elem_ptr;
 139       LLVMValueRef elem;
 140
 141       elem_offset = LLVMBuildExtractElement(builder, offsets, index, "");
 142       elem_ptr = LLVMBuildGEP(builder, base_ptr, &elem_offset, 1, "");
 143       elem_ptr = LLVMBuildBitCast(builder, elem_ptr, src_ptr_type, "");
 144       elem = LLVMBuildLoad(builder, elem_ptr, "");
 145
 146       assert(src_width <= dst_width);
 147       if(src_width > dst_width)
 148          elem = LLVMBuildTrunc(builder, elem, dst_elem_type, "");
 149       if(src_width < dst_width)
 150          elem = LLVMBuildZExt(builder, elem, dst_elem_type, "");
 151
 152       res = LLVMBuildInsertElement(builder, res, elem, index, "");
 153    }
 154
 155    return res;
 156 }
 157
 158
 159 /**
 160  * Compute the offset of a pixel.
 161  *
 162  * x, y, z, y_stride, z_stride are vectors
 163  */
 164 LLVMValueRef
 165 lp_build_sample_offset(struct lp_build_context *bld,
 166                        const struct util_format_description *format_desc,
 167                        LLVMValueRef x,
 168                        LLVMValueRef y,
 169                        LLVMValueRef z,
 170                        LLVMValueRef y_stride,
 171                        LLVMValueRef z_stride)
 172 {
 173    LLVMValueRef x_stride;
 174    LLVMValueRef offset;
 175
 176    x_stride = lp_build_const_vec(bld->type, format_desc->block.bits/8);
 177
 178    if(format_desc->colorspace == UTIL_FORMAT_COLORSPACE_ZS) {
 179       LLVMValueRef x_lo, x_hi;
 180       LLVMValueRef y_lo, y_hi;
 181       LLVMValueRef x_stride_lo, x_stride_hi;
 182       LLVMValueRef y_stride_lo, y_stride_hi;
 183       LLVMValueRef x_offset_lo, x_offset_hi;
 184       LLVMValueRef y_offset_lo, y_offset_hi;
 185       LLVMValueRef offset_lo, offset_hi;
 186
 187       /* XXX 1D & 3D addressing not done yet */
 188       assert(!z);
 189       assert(!z_stride);
 190
 191       x_lo = LLVMBuildAnd(bld->builder, x, bld->one, "");
 192       y_lo = LLVMBuildAnd(bld->builder, y, bld->one, "");
 193
 194       x_hi = LLVMBuildLShr(bld->builder, x, bld->one, "");
 195       y_hi = LLVMBuildLShr(bld->builder, y, bld->one, "");
 196
 197       x_stride_lo = x_stride;
 198       y_stride_lo = lp_build_const_vec(bld->type, 2*format_desc->block.bits/8);
 199
 200       x_stride_hi = lp_build_const_vec(bld->type, 4*format_desc->block.bits/8);
 201       y_stride_hi = LLVMBuildShl(bld->builder, y_stride, bld->one, "");
 202
 203       x_offset_lo = lp_build_mul(bld, x_lo, x_stride_lo);
 204       y_offset_lo = lp_build_mul(bld, y_lo, y_stride_lo);
 205       offset_lo = lp_build_add(bld, x_offset_lo, y_offset_lo);
 206
 207       x_offset_hi = lp_build_mul(bld, x_hi, x_stride_hi);
 208       y_offset_hi = lp_build_mul(bld, y_hi, y_stride_hi);
 209       offset_hi = lp_build_add(bld, x_offset_hi, y_offset_hi);
 210
 211       offset = lp_build_add(bld, offset_hi, offset_lo);
 212    }
 213    else {
 214       offset = lp_build_mul(bld, x, x_stride);
 215
 216       if (y && y_stride) {
 217          LLVMValueRef y_offset = lp_build_mul(bld, y, y_stride);
 218          offset = lp_build_add(bld, offset, y_offset);
 219       }
 220
 221       if (z && z_stride) {
 222          LLVMValueRef z_offset = lp_build_mul(bld, z, z_stride);
 223          offset = lp_build_add(bld, offset, z_offset);
 224       }
 225    }
 226
 227    return offset;
 228 }