diff options
| -rw-r--r-- | lib/mesa/src/intel/vulkan/anv_nir_lower_ycbcr_textures.c | 244 |
1 files changed, 185 insertions, 59 deletions
diff --git a/lib/mesa/src/intel/vulkan/anv_nir_lower_ycbcr_textures.c b/lib/mesa/src/intel/vulkan/anv_nir_lower_ycbcr_textures.c index e82cd032e..028f24e2f 100644 --- a/lib/mesa/src/intel/vulkan/anv_nir_lower_ycbcr_textures.c +++ b/lib/mesa/src/intel/vulkan/anv_nir_lower_ycbcr_textures.c @@ -25,35 +25,148 @@ #include "anv_private.h" #include "nir/nir.h" #include "nir/nir_builder.h" -#include "nir/nir_vulkan.h" struct ycbcr_state { nir_builder *builder; nir_ssa_def *image_size; nir_tex_instr *origin_tex; - nir_deref_instr *tex_deref; struct anv_ycbcr_conversion *conversion; }; +static nir_ssa_def * +y_range(nir_builder *b, + nir_ssa_def *y_channel, + int bpc, + VkSamplerYcbcrRangeKHR range) +{ + switch (range) { + case VK_SAMPLER_YCBCR_RANGE_ITU_FULL_KHR: + return y_channel; + case VK_SAMPLER_YCBCR_RANGE_ITU_NARROW_KHR: + 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, + VkSamplerYcbcrRangeKHR range) +{ + switch (range) { + case VK_SAMPLER_YCBCR_RANGE_ITU_FULL_KHR: + 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_KHR: + 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; + } +} + +static const nir_const_value * +ycbcr_model_to_rgb_matrix(VkSamplerYcbcrModelConversionKHR model) +{ + switch (model) { + case VK_SAMPLER_YCBCR_MODEL_CONVERSION_YCBCR_601_KHR: { + static const nir_const_value bt601[3] = { + { .f32 = { 1.402f, 1.0f, 0.0f, 0.0f } }, + { .f32 = { -0.714136286201022f, 1.0f, -0.344136286201022f, 0.0f } }, + { .f32 = { 0.0f, 1.0f, 1.772f, 0.0f } } + }; + + return bt601; + } + case VK_SAMPLER_YCBCR_MODEL_CONVERSION_YCBCR_709_KHR: { + static const nir_const_value bt709[3] = { + { .f32 = { 1.5748031496063f, 1.0f, 0.0, 0.0f } }, + { .f32 = { -0.468125209181067f, 1.0f, -0.187327487470334f, 0.0f } }, + { .f32 = { 0.0f, 1.0f, 1.85563184264242f, 0.0f } } + }; + + return bt709; + } + case VK_SAMPLER_YCBCR_MODEL_CONVERSION_YCBCR_2020_KHR: { + static const nir_const_value bt2020[3] = { + { .f32 = { 1.4746f, 1.0f, 0.0f, 0.0f } }, + { .f32 = { -0.571353126843658f, 1.0f, -0.164553126843658f, 0.0f } }, + { .f32 = { 0.0f, 1.0f, 1.8814f, 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, + uint32_t *bpcs) +{ + nir_builder *b = state->builder; + struct anv_ycbcr_conversion *conversion = state->conversion; + + nir_ssa_def *expanded_channels = + nir_vec4(b, + chroma_range(b, nir_channel(b, raw_channels, 0), + bpcs[0], conversion->ycbcr_range), + y_range(b, nir_channel(b, raw_channels, 1), + bpcs[1], conversion->ycbcr_range), + chroma_range(b, nir_channel(b, raw_channels, 2), + bpcs[2], conversion->ycbcr_range), + nir_imm_float(b, 1.0f)); + + if (conversion->ycbcr_model == VK_SAMPLER_YCBCR_MODEL_CONVERSION_YCBCR_IDENTITY_KHR) + return expanded_channels; + + const nir_const_value *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[0])), + nir_fdot4(b, expanded_channels, nir_build_imm(b, 4, 32, conversion_matrix[1])), + nir_fdot4(b, expanded_channels, nir_build_imm(b, 4, 32, conversion_matrix[2])) + }; + + return nir_vec4(b, + converted_channels[0], converted_channels[1], + converted_channels[2], nir_imm_float(b, 1.0f)); +} + /* TODO: we should probably replace this with a push constant/uniform. */ static nir_ssa_def * -get_texture_size(struct ycbcr_state *state, nir_deref_instr *texture) +get_texture_size(struct ycbcr_state *state, nir_deref_var *texture) { if (state->image_size) return state->image_size; nir_builder *b = state->builder; - const struct glsl_type *type = texture->type; - nir_tex_instr *tex = nir_tex_instr_create(b->shader, 1); + const struct glsl_type *type = nir_deref_tail(&texture->deref)->type; + nir_tex_instr *tex = nir_tex_instr_create(b->shader, 0); tex->op = nir_texop_txs; tex->sampler_dim = glsl_get_sampler_dim(type); tex->is_array = glsl_sampler_type_is_array(type); tex->is_shadow = glsl_sampler_type_is_shadow(type); - tex->dest_type = nir_type_int32; - - tex->src[0].src_type = nir_tex_src_texture_deref; - tex->src[0].src = nir_src_for_ssa(&texture->dest.ssa); + tex->texture = nir_deref_var_clone(texture, tex); + tex->dest_type = nir_type_int; nir_ssa_dest_init(&tex->instr, &tex->dest, nir_tex_instr_dest_size(tex), 32, NULL); @@ -86,13 +199,14 @@ implicit_downsampled_coords(struct ycbcr_state *state, { nir_builder *b = state->builder; struct anv_ycbcr_conversion *conversion = state->conversion; - nir_ssa_def *image_size = get_texture_size(state, state->tex_deref); + nir_ssa_def *image_size = get_texture_size(state, + state->origin_tex->texture); nir_ssa_def *comp[4] = { NULL, }; int c; for (c = 0; c < ARRAY_SIZE(conversion->chroma_offsets); c++) { if (plane_format->denominator_scales[c] > 1 && - conversion->chroma_offsets[c] == VK_CHROMA_LOCATION_COSITED_EVEN) { + conversion->chroma_offsets[c] == VK_CHROMA_LOCATION_COSITED_EVEN_KHR) { comp[c] = implicit_downsampled_coord(b, nir_channel(b, old_coords, c), nir_channel(b, image_size, c), @@ -133,9 +247,9 @@ create_plane_tex_instr_implicit(struct ycbcr_state *state, plane_format)); break; } - FALLTHROUGH; + /* fall through */ default: - nir_src_copy(&tex->src[i].src, &old_tex->src[i].src, &tex->instr); + nir_src_copy(&tex->src[i].src, &old_tex->src[i].src, tex); break; } } @@ -151,8 +265,11 @@ create_plane_tex_instr_implicit(struct ycbcr_state *state, tex->component = old_tex->component; tex->texture_index = old_tex->texture_index; + tex->texture_array_size = old_tex->texture_array_size; + tex->texture = nir_deref_var_clone(old_tex->texture, tex); + tex->sampler_index = old_tex->sampler_index; - tex->is_array = old_tex->is_array; + tex->sampler = nir_deref_var_clone(old_tex->sampler, tex); nir_ssa_dest_init(&tex->instr, &tex->dest, old_tex->dest.ssa.num_components, @@ -199,24 +316,13 @@ swizzle_channel(struct isl_swizzle swizzle, unsigned channel) } static bool -anv_nir_lower_ycbcr_textures_instr(nir_builder *builder, - nir_instr *instr, - void *cb_data) +try_lower_tex_ycbcr(struct anv_pipeline *pipeline, + nir_builder *builder, + nir_tex_instr *tex) { - const struct anv_pipeline_layout *layout = cb_data; - - if (instr->type != nir_instr_type_tex) - return false; - - nir_tex_instr *tex = nir_instr_as_tex(instr); - - 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); + nir_variable *var = tex->texture->var; const struct anv_descriptor_set_layout *set_layout = - layout->set[var->data.descriptor_set].layout; + pipeline->layout->set[var->data.descriptor_set].layout; const struct anv_descriptor_set_binding_layout *binding = &set_layout->binding[var->data.binding]; @@ -231,16 +337,18 @@ anv_nir_lower_ycbcr_textures_instr(nir_builder *builder, if (binding->immutable_samplers == NULL) 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)) + unsigned texture_index = tex->texture_index; + if (tex->texture->deref.child) { + assert(tex->texture->deref.child->deref_type == nir_deref_type_array); + nir_deref_array *deref_array = nir_deref_as_array(tex->texture->deref.child); + if (deref_array->deref_array_type != nir_deref_array_type_direct) return false; - array_index = nir_src_as_uint(deref->arr.index); - array_index = MIN2(array_index, binding->array_size - 1); + size_t hw_binding_size = + anv_descriptor_set_binding_layout_get_hw_size(binding); + texture_index += MIN2(deref_array->base_offset, hw_binding_size - 1); } - const struct anv_sampler *sampler = binding->immutable_samplers[array_index]; + const struct anv_sampler *sampler = + binding->immutable_samplers[texture_index]; if (sampler->conversion == NULL) return false; @@ -248,7 +356,6 @@ anv_nir_lower_ycbcr_textures_instr(nir_builder *builder, struct ycbcr_state state = { .builder = builder, .origin_tex = tex, - .tex_deref = deref, .conversion = sampler->conversion, }; @@ -264,11 +371,11 @@ anv_nir_lower_ycbcr_textures_instr(nir_builder *builder, uint8_t y_bpc = y_isl_layout->channels_array[0].bits; /* |ycbcr_comp| holds components in the order : Cr-Y-Cb */ - nir_ssa_def *zero = nir_imm_float(builder, 0.0f); - nir_ssa_def *one = nir_imm_float(builder, 1.0f); - /* Use extra 2 channels for following swizzle */ - nir_ssa_def *ycbcr_comp[5] = { zero, zero, zero, one, zero }; - + nir_ssa_def *ycbcr_comp[5] = { NULL, NULL, NULL, + /* Use extra 2 channels for following swizzle */ + nir_imm_float(builder, 1.0f), + nir_imm_float(builder, 0.0f), + }; uint8_t ycbcr_bpcs[5]; memset(ycbcr_bpcs, y_bpc, sizeof(ycbcr_bpcs)); @@ -323,27 +430,46 @@ anv_nir_lower_ycbcr_textures_instr(nir_builder *builder, } nir_ssa_def *result = nir_vec(builder, swizzled_comp, 4); - if (state.conversion->ycbcr_model != VK_SAMPLER_YCBCR_MODEL_CONVERSION_RGB_IDENTITY) { - result = nir_convert_ycbcr_to_rgb(builder, - state.conversion->ycbcr_model, - state.conversion->ycbcr_range, - result, - swizzled_bpcs); - } + if (state.conversion->ycbcr_model != VK_SAMPLER_YCBCR_MODEL_CONVERSION_RGB_IDENTITY_KHR) + result = convert_ycbcr(&state, result, swizzled_bpcs); - nir_ssa_def_rewrite_uses(&tex->dest.ssa, result); + nir_ssa_def_rewrite_uses(&tex->dest.ssa, nir_src_for_ssa(result)); nir_instr_remove(&tex->instr); return true; } bool -anv_nir_lower_ycbcr_textures(nir_shader *shader, - const struct anv_pipeline_layout *layout) +anv_nir_lower_ycbcr_textures(nir_shader *shader, struct anv_pipeline *pipeline) { - return nir_shader_instructions_pass(shader, - anv_nir_lower_ycbcr_textures_instr, - nir_metadata_block_index | - nir_metadata_dominance, - (void *)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(pipeline, &builder, tex); + } + } + + if (function_progress) { + nir_metadata_preserve(function->impl, + nir_metadata_block_index | + nir_metadata_dominance); + } + + progress |= function_progress; + } + + return progress; } |