src/freedreno/ir3/ir3_image.c

   1 /*
   2  * Copyright (C) 2017-2018 Rob Clark <robclark@freedesktop.org>
   3  *
   4  * Permission is hereby granted, free of charge, to any person obtaining a
   5  * copy of this software and associated documentation files (the "Software"),
   6  * to deal in the Software without restriction, including without limitation
   7  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
   8  * and/or sell copies of the Software, and to permit persons to whom the
   9  * Software is furnished to do so, subject to the following conditions:
  10  *
  11  * The above copyright notice and this permission notice (including the next
  12  * paragraph) shall be included in all copies or substantial portions of the
  13  * Software.
  14  *
  15  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  16  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  17  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  18  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  19  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
  20  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  21  * SOFTWARE.
  22  *
  23  * Authors:
  24  *    Rob Clark <robclark@freedesktop.org>
  25  */
  26
  27 #include "ir3_image.h"
  28
  29
  30 /*
  31  * SSBO/Image to/from IBO/tex hw mapping table:
  32  */
  33
  34 void
  35 ir3_ibo_mapping_init(struct ir3_ibo_mapping *mapping, unsigned num_textures)
  36 {
  37         memset(mapping, IBO_INVALID, sizeof(*mapping));
  38         mapping->num_ibo = 0;
  39         mapping->num_tex = 0;
  40         mapping->tex_base = num_textures;
  41 }
  42
  43 unsigned
  44 ir3_ssbo_to_ibo(struct ir3_ibo_mapping *mapping, unsigned ssbo)
  45 {
  46         if (mapping->ssbo_to_ibo[ssbo] == IBO_INVALID) {
  47                 unsigned ibo = mapping->num_ibo++;
  48                 mapping->ssbo_to_ibo[ssbo] = ibo;
  49                 mapping->ibo_to_image[ibo] = IBO_SSBO | ssbo;
  50         }
  51         return mapping->ssbo_to_ibo[ssbo];
  52 }
  53
  54 unsigned
  55 ir3_ssbo_to_tex(struct ir3_ibo_mapping *mapping, unsigned ssbo)
  56 {
  57         if (mapping->ssbo_to_tex[ssbo] == IBO_INVALID) {
  58                 unsigned tex = mapping->num_tex++;
  59                 mapping->ssbo_to_tex[ssbo] = tex;
  60                 mapping->tex_to_image[tex] = IBO_SSBO | ssbo;
  61         }
  62         return mapping->ssbo_to_tex[ssbo] + mapping->tex_base;
  63 }
  64
  65 unsigned
  66 ir3_image_to_ibo(struct ir3_ibo_mapping *mapping, unsigned image)
  67 {
  68         if (mapping->image_to_ibo[image] == IBO_INVALID) {
  69                 unsigned ibo = mapping->num_ibo++;
  70                 mapping->image_to_ibo[image] = ibo;
  71                 mapping->ibo_to_image[ibo] = image;
  72         }
  73         return mapping->image_to_ibo[image];
  74 }
  75
  76 unsigned
  77 ir3_image_to_tex(struct ir3_ibo_mapping *mapping, unsigned image)
  78 {
  79         if (mapping->image_to_tex[image] == IBO_INVALID) {
  80                 unsigned tex = mapping->num_tex++;
  81                 mapping->image_to_tex[image] = tex;
  82                 mapping->tex_to_image[tex] = image;
  83         }
  84         return mapping->image_to_tex[image] + mapping->tex_base;
  85 }
  86
  87 /* Helper to parse the deref for an image to get image slot.  This should be
  88  * mapped to tex or ibo idx using ir3_image_to_tex() or ir3_image_to_ibo().
  89  */
  90 unsigned
  91 ir3_get_image_slot(nir_deref_instr *deref)
  92 {
  93         unsigned int loc = 0;
  94         unsigned inner_size = 1;
  95
  96         while (deref->deref_type != nir_deref_type_var) {
  97                 assert(deref->deref_type == nir_deref_type_array);
  98                 unsigned const_index = nir_src_as_uint(deref->arr.index);
  99
 100                 /* Go to the next instruction */
 101                 deref = nir_deref_instr_parent(deref);
 102
 103                 assert(glsl_type_is_array(deref->type));
 104                 const unsigned array_len = glsl_get_length(deref->type);
 105                 loc += MIN2(const_index, array_len - 1) * inner_size;
 106
 107                 /* Update the inner size */
 108                 inner_size *= array_len;
 109         }
 110
 111         loc += deref->var->data.driver_location;
 112
 113         return loc;
 114 }
 115
 116 /* see tex_info() for equiv logic for texture instructions.. it would be
 117  * nice if this could be better unified..
 118  */
 119 unsigned
 120 ir3_get_image_coords(const nir_variable *var, unsigned *flagsp)
 121 {
 122         const struct glsl_type *type = glsl_without_array(var->type);
 123         unsigned coords = glsl_get_sampler_coordinate_components(type);
 124         unsigned flags = 0;
 125
 126         if (coords == 3)
 127                 flags |= IR3_INSTR_3D;
 128
 129         if (glsl_sampler_type_is_array(type))
 130                 flags |= IR3_INSTR_A;
 131
 132         if (flagsp)
 133                 *flagsp = flags;
 134
 135         return coords;
 136 }
 137
 138 type_t
 139 ir3_get_image_type(const nir_variable *var)
 140 {
 141         switch (glsl_get_sampler_result_type(glsl_without_array(var->type))) {
 142         case GLSL_TYPE_UINT:
 143                 return TYPE_U32;
 144         case GLSL_TYPE_INT:
 145                 return TYPE_S32;
 146         case GLSL_TYPE_FLOAT:
 147                 return TYPE_F32;
 148         default:
 149                 unreachable("bad sampler type.");
 150                 return 0;
 151         }
 152 }
 153
 154 /* Returns the number of components for the different image formats
 155  * supported by the GLES 3.1 spec, plus those added by the
 156  * GL_NV_image_formats extension.
 157  */
 158 unsigned
 159 ir3_get_num_components_for_glformat(GLuint format)
 160 {
 161         switch (format) {
 162         case GL_R32F:
 163         case GL_R32I:
 164         case GL_R32UI:
 165         case GL_R16F:
 166         case GL_R16I:
 167         case GL_R16UI:
 168         case GL_R16:
 169         case GL_R16_SNORM:
 170         case GL_R8I:
 171         case GL_R8UI:
 172         case GL_R8:
 173         case GL_R8_SNORM:
 174                 return 1;
 175
 176         case GL_RG32F:
 177         case GL_RG32I:
 178         case GL_RG32UI:
 179         case GL_RG16F:
 180         case GL_RG16I:
 181         case GL_RG16UI:
 182         case GL_RG16:
 183         case GL_RG16_SNORM:
 184         case GL_RG8I:
 185         case GL_RG8UI:
 186         case GL_RG8:
 187         case GL_RG8_SNORM:
 188                 return 2;
 189
 190         case GL_R11F_G11F_B10F:
 191                 return 3;
 192
 193         case GL_RGBA32F:
 194         case GL_RGBA32I:
 195         case GL_RGBA32UI:
 196         case GL_RGBA16F:
 197         case GL_RGBA16I:
 198         case GL_RGBA16UI:
 199         case GL_RGBA16:
 200         case GL_RGBA16_SNORM:
 201         case GL_RGBA8I:
 202         case GL_RGBA8UI:
 203         case GL_RGBA8:
 204         case GL_RGBA8_SNORM:
 205         case GL_RGB10_A2UI:
 206         case GL_RGB10_A2:
 207                 return 4;
 208
 209         case GL_NONE:
 210                 /* Omitting the image format qualifier is allowed on desktop GL
 211                  * profiles. Assuming 4 components is always safe.
 212                  */
 213                 return 4;
 214
 215         default:
 216                 /* Return 4 components also for all other formats we don't know
 217                  * about. The format should have been validated already by
 218                  * the higher level API, but drop a debug message just in case.
 219                  */
 220                 debug_printf("Unhandled GL format %u while emitting imageStore()\n",
 221                                          format);
 222                 return 4;
 223         }
 224 }