src/gallium/drivers/panfrost/pan_assemble.c

   1 /*
   2  * © Copyright 2018 Alyssa Rosenzweig
   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  */
  24
  25 #include <stdio.h>
  26 #include <stdlib.h>
  27 #include <string.h>
  28 #include "pan_context.h"
  29
  30 #include "compiler/nir/nir.h"
  31 #include "nir/tgsi_to_nir.h"
  32 #include "midgard/midgard_compile.h"
  33 #include "util/u_dynarray.h"
  34
  35 #include "tgsi/tgsi_dump.h"
  36
  37 void
  38 panfrost_shader_compile(struct panfrost_context *ctx, struct mali_shader_meta *meta, const char *src, int type, struct panfrost_shader_state *state)
  39 {
  40         uint8_t *dst;
  41
  42         nir_shader *s;
  43
  44         struct pipe_shader_state *cso = state->base;
  45
  46         if (cso->type == PIPE_SHADER_IR_NIR) {
  47                 s = nir_shader_clone(NULL, cso->ir.nir);
  48         } else {
  49                 assert (cso->type == PIPE_SHADER_IR_TGSI);
  50                 //tgsi_dump(cso->tokens, 0);
  51                 s = tgsi_to_nir(cso->tokens, &midgard_nir_options);
  52         }
  53
  54         s->info.stage = type == JOB_TYPE_VERTEX ? MESA_SHADER_VERTEX : MESA_SHADER_FRAGMENT;
  55
  56         if (s->info.stage == MESA_SHADER_FRAGMENT) {
  57                 /* Inject the alpha test now if we need to */
  58
  59                 if (state->alpha_state.enabled) {
  60                         NIR_PASS_V(s, nir_lower_alpha_test, state->alpha_state.func, false);
  61                 }
  62         }
  63
  64         /* Call out to Midgard compiler given the above NIR */
  65
  66         midgard_program program = {
  67                 .alpha_ref = state->alpha_state.ref_value
  68         };
  69
  70         midgard_compile_shader_nir(s, &program, false);
  71
  72         /* Prepare the compiled binary for upload */
  73         int size = program.compiled.size;
  74         dst = program.compiled.data;
  75
  76         /* Upload the shader. The lookahead tag is ORed on as a tagged pointer.
  77          * I bet someone just thought that would be a cute pun. At least,
  78          * that's how I'd do it. */
  79
  80         meta->shader = panfrost_upload(&ctx->shaders, dst, size, true) | program.first_tag;
  81
  82         util_dynarray_fini(&program.compiled);
  83
  84         meta->midgard1.uniform_count = MIN2(program.uniform_count, program.uniform_cutoff);
  85         meta->attribute_count = program.attribute_count;
  86         meta->varying_count = program.varying_count;
  87         meta->midgard1.work_count = program.work_register_count;
  88
  89         state->can_discard = program.can_discard;
  90         state->writes_point_size = program.writes_point_size;
  91
  92         /* Separate as primary uniform count is truncated */
  93         state->uniform_count = program.uniform_count;
  94
  95         /* gl_Position eats up an extra spot */
  96         if (type == JOB_TYPE_VERTEX)
  97                 meta->varying_count += 1;
  98
  99         /* gl_FragCoord does -not- eat an extra spot; it will be included in our count if we need it */
 100
 101
 102         meta->midgard1.unknown2 = 8; /* XXX */
 103
 104         /* Varyings are known only through the shader. We choose to upload this
 105          * information with the vertex shader, though the choice is perhaps
 106          * arbitrary */
 107
 108         if (type == JOB_TYPE_VERTEX) {
 109                 struct panfrost_varyings *varyings = &state->varyings;
 110
 111                 /* Measured in vec4 words. Don't include gl_Position */
 112                 int varying_count = program.varying_count;
 113
 114                 /* Setup two buffers, one for position, the other for normal
 115                  * varyings, as seen in traces. TODO: Are there other
 116                  * configurations we might use? */
 117
 118                 varyings->varying_buffer_count = 2;
 119
 120                 /* mediump vec4s sequentially */
 121                 varyings->varyings_stride[0] = (2 * sizeof(float)) * varying_count;
 122
 123                 /* highp gl_Position */
 124                 varyings->varyings_stride[1] = 4 * sizeof(float);
 125
 126                 /* mediump gl_PointSize */
 127                 if (program.writes_point_size) {
 128                         ++varyings->varying_buffer_count;
 129                         varyings->varyings_stride[2] = 2; /* sizeof(fp16) */
 130                 }
 131
 132                 /* Setup gl_Position, its weirdo analogue, and gl_PointSize (optionally) */
 133                 unsigned default_vec1_swizzle = panfrost_get_default_swizzle(1);
 134                 unsigned default_vec4_swizzle = panfrost_get_default_swizzle(4);
 135
 136                 struct mali_attr_meta vertex_special_varyings[] = {
 137                         {
 138                                 .index = 1,
 139                                 .format = MALI_VARYING_POS,
 140
 141                                 .swizzle = default_vec4_swizzle,
 142                                 .unknown1 = 0x2,
 143                         },
 144                         {
 145                                 .index = 1,
 146                                 .format = MALI_RGBA16F,
 147
 148                                 /* TODO: Wat? yyyy swizzle? */
 149                                 .swizzle = 0x249,
 150                                 .unknown1 = 0x0,
 151                         },
 152                         {
 153                                 .index = 2,
 154                                 .format = MALI_R16F,
 155                                 .swizzle =  default_vec1_swizzle,
 156                                 .unknown1 = 0x2
 157                         }
 158                 };
 159
 160                 /* How many special vertex varyings are actually required? */
 161                 int vertex_special_count = 2 + (program.writes_point_size ? 1 : 0);
 162
 163                 /* Setup actual varyings. XXX: Don't assume vec4 */
 164
 165                 struct mali_attr_meta mali_varyings[PIPE_MAX_ATTRIBS];
 166
 167                 for (int i = 0; i < varying_count; ++i) {
 168                         struct mali_attr_meta vec4_varying_meta = {
 169                                 .index = 0,
 170                                 .format = MALI_RGBA16F,
 171                                 .swizzle = default_vec4_swizzle,
 172                                 .unknown1 = 0x2,
 173
 174                                 /* Set offset to keep everything back-to-back in
 175                                  * the same buffer */
 176                                 .src_offset = 8 * i,
 177                         };
 178
 179                         mali_varyings[i] = vec4_varying_meta;
 180                 }
 181
 182                 /* We don't count the weirdo gl_Position in our varying count */
 183                 varyings->varying_count = varying_count - 1;
 184
 185                 /* In this context, position_meta represents the implicit
 186                  * gl_FragCoord varying. So, upload all the varyings */
 187
 188                 unsigned varyings_size = sizeof(struct mali_attr_meta) * varyings->varying_count;
 189                 unsigned vertex_special_size = sizeof(struct mali_attr_meta) * vertex_special_count;
 190                 unsigned vertex_size = vertex_special_size + varyings_size;
 191                 unsigned fragment_size = varyings_size + sizeof(struct mali_attr_meta);
 192
 193                 struct panfrost_transfer transfer = panfrost_allocate_chunk(ctx, vertex_size + fragment_size, HEAP_DESCRIPTOR);
 194
 195                 /* Copy varyings in the follow order:
 196                  *  - Position 1, 2
 197                  *  - Varyings 1, 2, ..., n
 198                  *  - Varyings 1, 2, ..., n (duplicate)
 199                  *  - Position 1
 200                  */
 201
 202                 memcpy(transfer.cpu, vertex_special_varyings, vertex_special_size);
 203                 memcpy(transfer.cpu + vertex_special_size, mali_varyings, varyings_size);
 204                 memcpy(transfer.cpu + vertex_size, mali_varyings, varyings_size);
 205                 memcpy(transfer.cpu + vertex_size + varyings_size, &vertex_special_varyings[0], sizeof(struct mali_attr_meta));
 206
 207                 /* Point to the descriptor */
 208                 varyings->varyings_buffer_cpu = transfer.cpu;
 209                 varyings->varyings_descriptor = transfer.gpu;
 210                 varyings->varyings_descriptor_fragment = transfer.gpu + vertex_size;
 211         }
 212 }