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_sampler_view *view,
  55                         const struct pipe_sampler_state *sampler)
  56 {
  57    const struct pipe_resource *texture = view->texture;
  58
  59    memset(state, 0, sizeof *state);
  60
  61    if(!texture)
  62       return;
  63
  64    if(!sampler)
  65       return;
  66
  67    /*
  68     * We don't copy sampler state over unless it is actually enabled, to avoid
  69     * spurious recompiles, as the sampler static state is part of the shader
  70     * key.
  71     *
  72     * Ideally the state tracker or cso_cache module would make all state
  73     * canonical, but until that happens it's better to be safe than sorry here.
  74     *
  75     * XXX: Actually there's much more than can be done here, especially
  76     * regarding 1D/2D/3D/CUBE textures, wrap modes, etc.
  77     */
  78
  79    state->format            = view->format;
  80    state->swizzle_r         = view->swizzle_r;
  81    state->swizzle_g         = view->swizzle_g;
  82    state->swizzle_b         = view->swizzle_b;
  83    state->swizzle_a         = view->swizzle_a;
  84
  85    state->target            = texture->target;
  86    state->pot_width         = util_is_pot(texture->width0);
  87    state->pot_height        = util_is_pot(texture->height0);
  88    state->pot_depth         = util_is_pot(texture->depth0);
  89
  90    state->wrap_s            = sampler->wrap_s;
  91    state->wrap_t            = sampler->wrap_t;
  92    state->wrap_r            = sampler->wrap_r;
  93    state->min_img_filter    = sampler->min_img_filter;
  94    state->mag_img_filter    = sampler->mag_img_filter;
  95    if (view->last_level) {
  96       state->min_mip_filter = sampler->min_mip_filter;
  97    } else {
  98       state->min_mip_filter = PIPE_TEX_MIPFILTER_NONE;
  99    }
 100
 101    state->compare_mode      = sampler->compare_mode;
 102    if (sampler->compare_mode != PIPE_TEX_COMPARE_NONE) {
 103       state->compare_func   = sampler->compare_func;
 104    }
 105
 106    state->normalized_coords = sampler->normalized_coords;
 107    state->lod_bias          = sampler->lod_bias;
 108    if (!view->last_level &&
 109        sampler->min_img_filter == sampler->mag_img_filter) {
 110       state->min_lod        = 0.0f;
 111       state->max_lod        = 0.0f;
 112    } else {
 113       state->min_lod        = MAX2(sampler->min_lod, 0.0f);
 114       state->max_lod        = sampler->max_lod;
 115    }
 116    state->border_color[0]   = sampler->border_color[0];
 117    state->border_color[1]   = sampler->border_color[1];
 118    state->border_color[2]   = sampler->border_color[2];
 119    state->border_color[3]   = sampler->border_color[3];
 120
 121    /*
 122     * FIXME: Handle the remainder of pipe_sampler_view.
 123     */
 124 }
 125
 126
 127 /**
 128  * Gather elements from scatter positions in memory into a single vector.
 129  * Use for fetching texels from a texture.
 130  * For SSE, typical values are length=4, src_width=32, dst_width=32.
 131  *
 132  * @param length length of the offsets
 133  * @param src_width src element width in bits
 134  * @param dst_width result element width in bits (src will be expanded to fit)
 135  * @param base_ptr base pointer, should be a i8 pointer type.
 136  * @param offsets vector with offsets
 137  */
 138 LLVMValueRef
 139 lp_build_gather(LLVMBuilderRef builder,
 140                 unsigned length,
 141                 unsigned src_width,
 142                 unsigned dst_width,
 143                 LLVMValueRef base_ptr,
 144                 LLVMValueRef offsets)
 145 {
 146    LLVMTypeRef src_type = LLVMIntType(src_width);
 147    LLVMTypeRef src_ptr_type = LLVMPointerType(src_type, 0);
 148    LLVMTypeRef dst_elem_type = LLVMIntType(dst_width);
 149    LLVMTypeRef dst_vec_type = LLVMVectorType(dst_elem_type, length);
 150    LLVMValueRef res;
 151    unsigned i;
 152
 153    res = LLVMGetUndef(dst_vec_type);
 154    for(i = 0; i < length; ++i) {
 155       LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), i, 0);
 156       LLVMValueRef elem_offset;
 157       LLVMValueRef elem_ptr;
 158       LLVMValueRef elem;
 159
 160       elem_offset = LLVMBuildExtractElement(builder, offsets, index, "");
 161       elem_ptr = LLVMBuildGEP(builder, base_ptr, &elem_offset, 1, "");
 162       elem_ptr = LLVMBuildBitCast(builder, elem_ptr, src_ptr_type, "");
 163       elem = LLVMBuildLoad(builder, elem_ptr, "");
 164
 165       assert(src_width <= dst_width);
 166       if(src_width > dst_width)
 167          elem = LLVMBuildTrunc(builder, elem, dst_elem_type, "");
 168       if(src_width < dst_width)
 169          elem = LLVMBuildZExt(builder, elem, dst_elem_type, "");
 170
 171       res = LLVMBuildInsertElement(builder, res, elem, index, "");
 172    }
 173
 174    return res;
 175 }
 176
 177
 178 /**
 179  * Compute the offset of a pixel block.
 180  *
 181  * x, y, z, y_stride, z_stride are vectors, and they refer to pixel blocks, as
 182  * per format description, and not individual pixels.
 183  */
 184 LLVMValueRef
 185 lp_build_sample_offset(struct lp_build_context *bld,
 186                        const struct util_format_description *format_desc,
 187                        LLVMValueRef x,
 188                        LLVMValueRef y,
 189                        LLVMValueRef z,
 190                        LLVMValueRef y_stride,
 191                        LLVMValueRef z_stride)
 192 {
 193    LLVMValueRef x_stride;
 194    LLVMValueRef offset;
 195
 196    x_stride = lp_build_const_vec(bld->type, format_desc->block.bits/8);
 197    offset = lp_build_mul(bld, x, x_stride);
 198
 199    if (y && y_stride) {
 200       LLVMValueRef y_offset = lp_build_mul(bld, y, y_stride);
 201       offset = lp_build_add(bld, offset, y_offset);
 202    }
 203
 204    if (z && z_stride) {
 205       LLVMValueRef z_offset = lp_build_mul(bld, z, z_stride);
 206       offset = lp_build_add(bld, offset, z_offset);
 207    }
 208
 209    return offset;
 210 }