src/gallium/drivers/panfrost/midgard/nir_lower_framebuffer.c

   1 /*
   2  * Copyright (C) 2019 Collabora, Ltd.
   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
  20  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
  21  * IN THE SOFTWARE.
  22  *
  23  * Authors (Collabora):
  24  *      Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
  25  */
  26
  27 /**
  28  * @file
  29  *
  30  * Implements framebuffer format conversions in software, specifically for
  31  * blend shaders on Midgard/Bifrost. load_output/store_output (derefs more
  32  * correctly -- pre I/O lowering) normally for the fragment stage within the
  33  * blend shader will operate with purely vec4 float ("nir") encodings. This
  34  * lowering stage, to be run before I/O is lowered, converts the native
  35  * framebuffer format to a NIR encoding after loads and vice versa before
  36  * stores. This pass is designed for a single render target; Midgard duplicates
  37  * blend shaders for MRT to simplify everything.
  38  */
  39
  40 #include "compiler/nir/nir.h"
  41 #include "compiler/nir/nir_builder.h"
  42 #include "nir_lower_blend.h"
  43
  44 static nir_ssa_def *
  45 nir_float_to_native(nir_builder *b, nir_ssa_def *c_float)
  46 {
  47    /* First, we scale from [0, 1] to [0, 255.0] */
  48    nir_ssa_def *scaled = nir_fmul_imm(b, nir_fsat(b, c_float), 255.0);
  49
  50    /* Next, we type convert */
  51    nir_ssa_def *converted = nir_u2u8(b, nir_f2u32(b,
  52             nir_fround_even(b, scaled)));
  53
  54    return converted;
  55 }
  56
  57 static nir_ssa_def *
  58 nir_native_to_float(nir_builder *b, nir_ssa_def *c_native)
  59 {
  60    return c_native;
  61 }
  62
  63 void
  64 nir_lower_framebuffer(nir_shader *shader)
  65 {
  66    /* Blend shaders are represented as special fragment shaders */
  67    assert(shader->info.stage == MESA_SHADER_FRAGMENT);
  68
  69    nir_foreach_function(func, shader) {
  70       nir_foreach_block(block, func->impl) {
  71          nir_foreach_instr_safe(instr, block) {
  72             if (instr->type != nir_instr_type_intrinsic)
  73                continue;
  74
  75             nir_intrinsic_instr *intr = nir_instr_as_intrinsic(instr);
  76
  77             bool is_load = intr->intrinsic == nir_intrinsic_load_deref;
  78             bool is_store = intr->intrinsic == nir_intrinsic_store_deref;
  79
  80             if (!(is_load || is_store))
  81                continue;
  82
  83             /* Don't worry about MRT */
  84             nir_variable *var = nir_intrinsic_get_var(intr, 0);
  85
  86             if (var->data.location != FRAG_RESULT_COLOR)
  87                continue;
  88
  89             nir_builder b;
  90             nir_builder_init(&b, func->impl);
  91
  92             if (is_store) {
  93                /* For stores, add conversion before */
  94                b.cursor = nir_before_instr(instr);
  95
  96                /* Grab the input color */
  97                nir_ssa_def *c_nir = nir_ssa_for_src(&b, intr->src[1], 4);
  98
  99                /* Format convert */
 100                nir_ssa_def *converted = nir_float_to_native(&b, c_nir);
 101
 102                /* Rewrite to use a native store by creating a new intrinsic */
 103                nir_intrinsic_instr *new =
 104                   nir_intrinsic_instr_create(shader, nir_intrinsic_store_raw_output_pan);
 105                new->src[0] = nir_src_for_ssa(converted);
 106
 107                /* TODO: What about non-RGBA? Is that different? */
 108                new->num_components = 4;
 109
 110                nir_builder_instr_insert(&b, &new->instr);
 111
 112                /* (And finally removing the old) */
 113                nir_instr_remove(instr);
 114             } else {
 115                /* For loads, add conversion after */
 116                b.cursor = nir_after_instr(instr);
 117
 118                /* Rewrite to use a native load by creating a new intrinsic */
 119
 120                nir_intrinsic_instr *new =
 121                   nir_intrinsic_instr_create(shader, nir_intrinsic_load_raw_output_pan);
 122
 123                new->num_components = 4;
 124
 125                unsigned bitsize = 32;
 126                nir_ssa_dest_init(&new->instr, &new->dest, 4, bitsize, NULL);
 127                nir_builder_instr_insert(&b, &new->instr);
 128
 129                /* Convert the raw value */
 130                nir_ssa_def *raw = &new->dest.ssa;
 131                nir_ssa_def *converted = nir_native_to_float(&b, raw);
 132
 133                /* Rewrite to use the converted value */
 134                nir_src rewritten = nir_src_for_ssa(converted);
 135                nir_ssa_def_rewrite_uses_after(&intr->dest.ssa, rewritten, instr);
 136
 137                /* Finally, remove the old load */
 138                nir_instr_remove(instr);
 139             }
 140          }
 141       }
 142
 143       nir_metadata_preserve(func->impl, nir_metadata_block_index |
 144                             nir_metadata_dominance);
 145    }
 146 }