summaryrefslogtreecommitdiffstats
path: root/src
diff options
context:
space:
mode:
authorBas Nieuwenhuizen <[email protected]>2019-03-21 01:29:52 +0100
committerBas Nieuwenhuizen <[email protected]>2019-04-25 19:56:20 +0000
commit91702374d5d70296c967c324fff1844cc2933dd4 (patch)
tree2f3c219537c11d6ec27cb87ae99e6c5c2a7caae3 /src
parent5564c38212ac3d3aa2fdfca4da03f10db79a07a4 (diff)
radv: Add ycbcr lowering pass.
Reviewed-by: Samuel Pitoiset <[email protected]>
Diffstat (limited to 'src')
-rw-r--r--src/amd/vulkan/Makefile.sources1
-rw-r--r--src/amd/vulkan/meson.build1
-rw-r--r--src/amd/vulkan/radv_nir_lower_ycbcr_textures.c371
-rw-r--r--src/amd/vulkan/radv_shader.h3
4 files changed, 376 insertions, 0 deletions
diff --git a/src/amd/vulkan/Makefile.sources b/src/amd/vulkan/Makefile.sources
index f6327563ee9..df90c1150af 100644
--- a/src/amd/vulkan/Makefile.sources
+++ b/src/amd/vulkan/Makefile.sources
@@ -54,6 +54,7 @@ VULKAN_FILES := \
radv_meta_resolve.c \
radv_meta_resolve_cs.c \
radv_meta_resolve_fs.c \
+ radv_nir_lower_ycbcr_textures.c \
radv_nir_to_llvm.c \
radv_llvm_helper.cpp \
radv_pass.c \
diff --git a/src/amd/vulkan/meson.build b/src/amd/vulkan/meson.build
index 3d7ad9f55a1..532267343d1 100644
--- a/src/amd/vulkan/meson.build
+++ b/src/amd/vulkan/meson.build
@@ -82,6 +82,7 @@ libradv_files = files(
'radv_meta_resolve.c',
'radv_meta_resolve_cs.c',
'radv_meta_resolve_fs.c',
+ 'radv_nir_lower_ycbcr_textures.c',
'radv_nir_to_llvm.c',
'radv_pass.c',
'radv_pipeline.c',
diff --git a/src/amd/vulkan/radv_nir_lower_ycbcr_textures.c b/src/amd/vulkan/radv_nir_lower_ycbcr_textures.c
new file mode 100644
index 00000000000..e37cc10a8a0
--- /dev/null
+++ b/src/amd/vulkan/radv_nir_lower_ycbcr_textures.c
@@ -0,0 +1,371 @@
+/*
+ * Copyright © 2017 Intel Corporation
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+ * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
+ * IN THE SOFTWARE.
+ */
+
+#include "radv_private.h"
+#include "radv_shader.h"
+#include "vk_format.h"
+#include "nir/nir.h"
+#include "nir/nir_builder.h"
+
+struct ycbcr_state {
+ nir_builder *builder;
+ nir_ssa_def *image_size;
+ nir_tex_instr *origin_tex;
+ nir_deref_instr *tex_deref;
+ const struct radv_sampler_ycbcr_conversion *conversion;
+};
+
+static nir_ssa_def *
+y_range(nir_builder *b,
+ nir_ssa_def *y_channel,
+ int bpc,
+ VkSamplerYcbcrRange range)
+{
+ switch (range) {
+ case VK_SAMPLER_YCBCR_RANGE_ITU_FULL:
+ return y_channel;
+ case VK_SAMPLER_YCBCR_RANGE_ITU_NARROW:
+ return nir_fmul(b,
+ nir_fadd(b,
+ nir_fmul(b, y_channel,
+ nir_imm_float(b, pow(2, bpc) - 1)),
+ nir_imm_float(b, -16.0f * pow(2, bpc - 8))),
+ nir_frcp(b, nir_imm_float(b, 219.0f * pow(2, bpc - 8))));
+ default:
+ unreachable("missing Ycbcr range");
+ return NULL;
+ }
+}
+
+static nir_ssa_def *
+chroma_range(nir_builder *b,
+ nir_ssa_def *chroma_channel,
+ int bpc,
+ VkSamplerYcbcrRange range)
+{
+ switch (range) {
+ case VK_SAMPLER_YCBCR_RANGE_ITU_FULL:
+ return nir_fadd(b, chroma_channel,
+ nir_imm_float(b, -pow(2, bpc - 1) / (pow(2, bpc) - 1.0f)));
+ case VK_SAMPLER_YCBCR_RANGE_ITU_NARROW:
+ return nir_fmul(b,
+ nir_fadd(b,
+ nir_fmul(b, chroma_channel,
+ nir_imm_float(b, pow(2, bpc) - 1)),
+ nir_imm_float(b, -128.0f * pow(2, bpc - 8))),
+ nir_frcp(b, nir_imm_float(b, 224.0f * pow(2, bpc - 8))));
+ default:
+ unreachable("missing Ycbcr range");
+ return NULL;
+ }
+}
+
+typedef struct nir_const_value_3_4 {
+ nir_const_value v[3][4];
+} nir_const_value_3_4;
+
+static const nir_const_value_3_4 *
+ycbcr_model_to_rgb_matrix(VkSamplerYcbcrModelConversion model)
+{
+ switch (model) {
+ case VK_SAMPLER_YCBCR_MODEL_CONVERSION_YCBCR_601: {
+ static const nir_const_value_3_4 bt601 = { {
+ { { .f32 = 1.402f }, { .f32 = 1.0f }, { .f32 = 0.0f }, { .f32 = 0.0f } },
+ { { .f32 = -0.714136286201022f }, { .f32 = 1.0f }, { .f32 = -0.344136286201022f }, { .f32 = 0.0f } },
+ { { .f32 = 0.0f }, { .f32 = 1.0f }, { .f32 = 1.772f }, { .f32 = 0.0f } },
+ } };
+
+ return &bt601;
+ }
+ case VK_SAMPLER_YCBCR_MODEL_CONVERSION_YCBCR_709: {
+ static const nir_const_value_3_4 bt709 = { {
+ { { .f32 = 1.5748031496063f }, { .f32 = 1.0f }, { .f32 = 0.0f }, { .f32 = 0.0f } },
+ { { .f32 = -0.468125209181067f }, { .f32 = 1.0f }, { .f32 = -0.187327487470334f }, { .f32 = 0.0f } },
+ { { .f32 = 0.0f }, { .f32 = 1.0f }, { .f32 = 1.85563184264242f }, { .f32 = 0.0f } },
+ } };
+
+ return &bt709;
+ }
+ case VK_SAMPLER_YCBCR_MODEL_CONVERSION_YCBCR_2020: {
+ static const nir_const_value_3_4 bt2020 = { {
+ { { .f32 = 1.4746f }, { .f32 = 1.0f }, { .f32 = 0.0f }, { .f32 = 0.0f } },
+ { { .f32 = -0.571353126843658f }, { .f32 = 1.0f }, { .f32 = -0.164553126843658f }, { .f32 = 0.0f } },
+ { { .f32 = 0.0f }, { .f32 = 1.0f }, { .f32 = 1.8814f }, { .f32 = 0.0f } },
+ } };
+
+ return &bt2020;
+ }
+ default:
+ unreachable("missing Ycbcr model");
+ return NULL;
+ }
+}
+
+static nir_ssa_def *
+convert_ycbcr(struct ycbcr_state *state,
+ nir_ssa_def *raw_channels,
+ uint8_t bits)
+{
+ nir_builder *b = state->builder;
+ const struct radv_sampler_ycbcr_conversion *conversion = state->conversion;
+
+ nir_ssa_def *expanded_channels =
+ nir_vec4(b,
+ chroma_range(b, nir_channel(b, raw_channels, 0),
+ bits, conversion->ycbcr_range),
+ y_range(b, nir_channel(b, raw_channels, 1),
+ bits, conversion->ycbcr_range),
+ chroma_range(b, nir_channel(b, raw_channels, 2),
+ bits, conversion->ycbcr_range),
+ nir_imm_float(b, 1.0f));
+
+ if (conversion->ycbcr_model == VK_SAMPLER_YCBCR_MODEL_CONVERSION_YCBCR_IDENTITY)
+ return expanded_channels;
+
+ const nir_const_value_3_4 *conversion_matrix =
+ ycbcr_model_to_rgb_matrix(conversion->ycbcr_model);
+
+ nir_ssa_def *converted_channels[] = {
+ nir_fdot4(b, expanded_channels, nir_build_imm(b, 4, 32, conversion_matrix->v[0])),
+ nir_fdot4(b, expanded_channels, nir_build_imm(b, 4, 32, conversion_matrix->v[1])),
+ nir_fdot4(b, expanded_channels, nir_build_imm(b, 4, 32, conversion_matrix->v[2]))
+ };
+
+ return nir_vec4(b,
+ converted_channels[0], converted_channels[1],
+ converted_channels[2], nir_imm_float(b, 1.0f));
+}
+
+static nir_ssa_def *
+create_plane_tex_instr_implicit(struct ycbcr_state *state,
+ uint32_t plane)
+{
+ nir_builder *b = state->builder;
+ nir_tex_instr *old_tex = state->origin_tex;
+ nir_tex_instr *tex = nir_tex_instr_create(b->shader, old_tex->num_srcs+ 1);
+
+ for (uint32_t i = 0; i < old_tex->num_srcs; i++) {
+ tex->src[i].src_type = old_tex->src[i].src_type;
+ nir_src_copy(&tex->src[i].src, &old_tex->src[i].src, tex);
+ }
+
+ tex->src[tex->num_srcs - 1].src = nir_src_for_ssa(nir_imm_int(b, plane));
+ tex->src[tex->num_srcs - 1].src_type = nir_tex_src_plane;
+
+ tex->sampler_dim = old_tex->sampler_dim;
+ tex->dest_type = old_tex->dest_type;
+
+ tex->op = old_tex->op;
+ tex->coord_components = old_tex->coord_components;
+ tex->is_new_style_shadow = old_tex->is_new_style_shadow;
+ tex->component = old_tex->component;
+
+ tex->texture_index = old_tex->texture_index;
+ tex->texture_array_size = old_tex->texture_array_size;
+ tex->sampler_index = old_tex->sampler_index;
+
+ nir_ssa_dest_init(&tex->instr, &tex->dest,
+ old_tex->dest.ssa.num_components,
+ nir_dest_bit_size(old_tex->dest), NULL);
+ nir_builder_instr_insert(b, &tex->instr);
+
+ return &tex->dest.ssa;
+}
+
+struct swizzle_info {
+ unsigned plane[4];
+ unsigned swizzle[4];
+};
+
+static struct swizzle_info
+get_plane_swizzles(VkFormat format)
+{
+ int planes = vk_format_get_plane_count(format);
+ switch (planes) {
+ case 3:
+ return (struct swizzle_info) {
+ {2, 0, 1, 0},
+ {0, 0, 0, 3}
+ };
+ case 2:
+ return (struct swizzle_info) {
+ {1, 0, 1, 0},
+ {1, 0, 0, 3}
+ };
+ case 1:
+ return (struct swizzle_info) {
+ {0, 0, 0, 0},
+ {0, 1, 2, 3}
+ };
+ default:
+ unreachable("unhandled plane count for ycbcr swizzling");
+ }
+}
+
+
+static nir_ssa_def *
+build_swizzled_components(nir_builder *builder,
+ VkFormat format,
+ VkComponentMapping mapping,
+ nir_ssa_def **plane_values)
+{
+ struct swizzle_info plane_swizzle = get_plane_swizzles(format);
+ enum vk_swizzle swizzles[4];
+ nir_ssa_def *values[4];
+
+ vk_format_compose_swizzles(&mapping, (const unsigned char[4]){0,1,2,3}, swizzles);
+
+ nir_ssa_def *zero = nir_imm_float(builder, 0.0f);
+ nir_ssa_def *one = nir_imm_float(builder, 1.0f);
+
+ for (unsigned i = 0; i < 4; ++i) {
+ switch(swizzles[i]) {
+ case VK_SWIZZLE_X:
+ case VK_SWIZZLE_Y:
+ case VK_SWIZZLE_Z:
+ case VK_SWIZZLE_W: {
+ unsigned channel = swizzles[i] - VK_SWIZZLE_X;
+ values[i] = nir_channel(builder,
+ plane_values[plane_swizzle.plane[channel]],
+ plane_swizzle.swizzle[channel]);
+ break;
+ }
+ case VK_SWIZZLE_0:
+ values[i] = zero;
+ break;
+ case VK_SWIZZLE_1:
+ values[i] = one;
+ break;
+ default:
+ unreachable("unhandled swizzle");
+ }
+ }
+ return nir_vec(builder, values, 4);
+}
+
+static bool
+try_lower_tex_ycbcr(const struct radv_pipeline_layout *layout,
+ nir_builder *builder,
+ nir_tex_instr *tex)
+{
+ int deref_src_idx = nir_tex_instr_src_index(tex, nir_tex_src_texture_deref);
+ assert(deref_src_idx >= 0);
+ nir_deref_instr *deref = nir_src_as_deref(tex->src[deref_src_idx].src);
+
+ nir_variable *var = nir_deref_instr_get_variable(deref);
+ const struct radv_descriptor_set_layout *set_layout =
+ layout->set[var->data.descriptor_set].layout;
+ const struct radv_descriptor_set_binding_layout *binding =
+ &set_layout->binding[var->data.binding];
+ const struct radv_sampler_ycbcr_conversion *ycbcr_samplers =
+ radv_immutable_ycbcr_samplers(set_layout, var->data.binding);
+
+ if (!ycbcr_samplers)
+ return false;
+
+ /* For the following instructions, we don't apply any change and let the
+ * instruction apply to the first plane.
+ */
+ if (tex->op == nir_texop_txs ||
+ tex->op == nir_texop_query_levels ||
+ tex->op == nir_texop_lod)
+ return false;
+
+ assert(tex->texture_index == 0);
+ unsigned array_index = 0;
+ if (deref->deref_type != nir_deref_type_var) {
+ assert(deref->deref_type == nir_deref_type_array);
+ if (!nir_src_is_const(deref->arr.index))
+ return false;
+ array_index = nir_src_as_uint(deref->arr.index);
+ array_index = MIN2(array_index, binding->array_size - 1);
+ }
+ const struct radv_sampler_ycbcr_conversion *ycbcr_sampler = ycbcr_samplers + array_index;
+
+ if (ycbcr_sampler->format == VK_FORMAT_UNDEFINED)
+ return false;
+
+ struct ycbcr_state state = {
+ .builder = builder,
+ .origin_tex = tex,
+ .tex_deref = deref,
+ .conversion = ycbcr_sampler,
+ };
+
+ builder->cursor = nir_before_instr(&tex->instr);
+
+ VkFormat format = state.conversion->format;
+ const int plane_count = vk_format_get_plane_count(format);
+ nir_ssa_def *plane_values[3];
+
+ for (int p = 0; p < plane_count; ++p) {
+ plane_values[p] = create_plane_tex_instr_implicit(&state, p);
+ }
+
+ nir_ssa_def *result = build_swizzled_components(builder, format, ycbcr_sampler->components, plane_values);
+ if (state.conversion->ycbcr_model != VK_SAMPLER_YCBCR_MODEL_CONVERSION_RGB_IDENTITY) {
+ VkFormat first_format = vk_format_get_plane_format(format, 0);
+ result = convert_ycbcr(&state, result, vk_format_get_component_bits(first_format, VK_FORMAT_COLORSPACE_RGB, VK_SWIZZLE_X));
+ }
+
+ nir_ssa_def_rewrite_uses(&tex->dest.ssa, nir_src_for_ssa(result));
+ nir_instr_remove(&tex->instr);
+
+ return true;
+}
+
+bool
+radv_nir_lower_ycbcr_textures(nir_shader *shader,
+ const struct radv_pipeline_layout *layout)
+{
+ bool progress = false;
+
+ nir_foreach_function(function, shader) {
+ if (!function->impl)
+ continue;
+
+ bool function_progress = false;
+ nir_builder builder;
+ nir_builder_init(&builder, function->impl);
+
+ nir_foreach_block(block, function->impl) {
+ nir_foreach_instr_safe(instr, block) {
+ if (instr->type != nir_instr_type_tex)
+ continue;
+
+ nir_tex_instr *tex = nir_instr_as_tex(instr);
+ function_progress |= try_lower_tex_ycbcr(layout, &builder, tex);
+ }
+ }
+
+ if (function_progress) {
+ nir_metadata_preserve(function->impl,
+ nir_metadata_block_index |
+ nir_metadata_dominance);
+ }
+
+ progress |= function_progress;
+ }
+
+ return progress;
+}
diff --git a/src/amd/vulkan/radv_shader.h b/src/amd/vulkan/radv_shader.h
index fe2f2868630..ac367bfc82e 100644
--- a/src/amd/vulkan/radv_shader.h
+++ b/src/amd/vulkan/radv_shader.h
@@ -337,6 +337,9 @@ struct radv_shader_slab {
void
radv_optimize_nir(struct nir_shader *shader, bool optimize_conservatively,
bool allow_copies);
+bool
+radv_nir_lower_ycbcr_textures(nir_shader *shader,
+ const struct radv_pipeline_layout *layout);
nir_shader *
radv_shader_compile_to_nir(struct radv_device *device,