X-Git-Url: https://git.libre-soc.org/?a=blobdiff_plain;f=src%2Famd%2Fvulkan%2Fradv_meta_resolve_cs.c;h=b6c9e1c2b1faa9076115be2939a8c626c0570706;hb=cd247cf45648159cb2676e88108ec8d8dc97095b;hp=ad02594614ceb29e50959af8d63b03a701178bfe;hpb=0a8127bbfbe8a8af45dacf6c0998a8f0f1152f4b;p=mesa.git diff --git a/src/amd/vulkan/radv_meta_resolve_cs.c b/src/amd/vulkan/radv_meta_resolve_cs.c index ad02594614c..b6c9e1c2b1f 100644 --- a/src/amd/vulkan/radv_meta_resolve_cs.c +++ b/src/amd/vulkan/radv_meta_resolve_cs.c @@ -34,33 +34,27 @@ static nir_ssa_def *radv_meta_build_resolve_srgb_conversion(nir_builder *b, nir_ssa_def *input) { - nir_const_value v; unsigned i; - v.u32[0] = 0x3b4d2e1c; // 0.00313080009 nir_ssa_def *cmp[3]; for (i = 0; i < 3; i++) cmp[i] = nir_flt(b, nir_channel(b, input, i), - nir_build_imm(b, 1, 32, v)); + nir_imm_int(b, 0x3b4d2e1c)); nir_ssa_def *ltvals[3]; - v.f32[0] = 12.92; for (i = 0; i < 3; i++) ltvals[i] = nir_fmul(b, nir_channel(b, input, i), - nir_build_imm(b, 1, 32, v)); + nir_imm_float(b, 12.92)); nir_ssa_def *gtvals[3]; for (i = 0; i < 3; i++) { - v.f32[0] = 1.0/2.4; gtvals[i] = nir_fpow(b, nir_channel(b, input, i), - nir_build_imm(b, 1, 32, v)); - v.f32[0] = 1.055; + nir_imm_float(b, 1.0/2.4)); gtvals[i] = nir_fmul(b, gtvals[i], - nir_build_imm(b, 1, 32, v)); - v.f32[0] = 0.055; + nir_imm_float(b, 1.055)); gtvals[i] = nir_fsub(b, gtvals[i], - nir_build_imm(b, 1, 32, v)); + nir_imm_float(b, 0.055)); } nir_ssa_def *comp[4]; @@ -99,8 +93,8 @@ build_resolve_compute_shader(struct radv_device *dev, bool is_integer, bool is_s img_type, "out_img"); output_img->data.descriptor_set = 0; output_img->data.binding = 1; - nir_ssa_def *invoc_id = nir_load_system_value(&b, nir_intrinsic_load_local_invocation_id, 0); - nir_ssa_def *wg_id = nir_load_system_value(&b, nir_intrinsic_load_work_group_id, 0); + nir_ssa_def *invoc_id = nir_load_local_invocation_id(&b); + nir_ssa_def *wg_id = nir_load_work_group_id(&b); nir_ssa_def *block_size = nir_imm_ivec4(&b, b.shader->info.cs.local_size[0], b.shader->info.cs.local_size[1], @@ -136,6 +130,7 @@ build_resolve_compute_shader(struct radv_device *dev, bool is_integer, bool is_s nir_ssa_def *coord = nir_iadd(&b, global_id, &dst_offset->dest.ssa); nir_intrinsic_instr *store = nir_intrinsic_instr_create(b.shader, nir_intrinsic_image_deref_store); + store->num_components = 4; store->src[0] = nir_src_for_ssa(&nir_build_deref_var(&b, output_img)->dest.ssa); store->src[1] = nir_src_for_ssa(coord); store->src[2] = nir_src_for_ssa(nir_ssa_undef(&b, 1, 32)); @@ -212,6 +207,12 @@ create_resolve_pipeline(struct radv_device *device, VkResult result; struct radv_shader_module cs = { .nir = NULL }; + mtx_lock(&device->meta_state.mtx); + if (*pipeline) { + mtx_unlock(&device->meta_state.mtx); + return VK_SUCCESS; + } + cs.nir = build_resolve_compute_shader(device, is_integer, is_srgb, samples); /* compute shader */ @@ -239,14 +240,16 @@ create_resolve_pipeline(struct radv_device *device, goto fail; ralloc_free(cs.nir); + mtx_unlock(&device->meta_state.mtx); return VK_SUCCESS; fail: ralloc_free(cs.nir); + mtx_unlock(&device->meta_state.mtx); return result; } VkResult -radv_device_init_meta_resolve_compute_state(struct radv_device *device) +radv_device_init_meta_resolve_compute_state(struct radv_device *device, bool on_demand) { struct radv_meta_state *state = &device->meta_state; VkResult res; @@ -255,6 +258,9 @@ radv_device_init_meta_resolve_compute_state(struct radv_device *device) if (res != VK_SUCCESS) goto fail; + if (on_demand) + return VK_SUCCESS; + for (uint32_t i = 0; i < MAX_SAMPLES_LOG2; ++i) { uint32_t samples = 1 << i; @@ -307,6 +313,39 @@ radv_device_finish_meta_resolve_compute_state(struct radv_device *device) &state->alloc); } +static VkPipeline * +radv_get_resolve_pipeline(struct radv_cmd_buffer *cmd_buffer, + struct radv_image_view *src_iview) +{ + struct radv_device *device = cmd_buffer->device; + struct radv_meta_state *state = &device->meta_state; + uint32_t samples = src_iview->image->info.samples; + uint32_t samples_log2 = ffs(samples) - 1; + VkPipeline *pipeline; + + if (vk_format_is_int(src_iview->vk_format)) + pipeline = &state->resolve_compute.rc[samples_log2].i_pipeline; + else if (vk_format_is_srgb(src_iview->vk_format)) + pipeline = &state->resolve_compute.rc[samples_log2].srgb_pipeline; + else + pipeline = &state->resolve_compute.rc[samples_log2].pipeline; + + if (!*pipeline) { + VkResult ret; + + ret = create_resolve_pipeline(device, samples, + vk_format_is_int(src_iview->vk_format), + vk_format_is_srgb(src_iview->vk_format), + pipeline); + if (ret != VK_SUCCESS) { + cmd_buffer->record_result = ret; + return NULL; + } + } + + return pipeline; +} + static void emit_resolve(struct radv_cmd_buffer *cmd_buffer, struct radv_image_view *src_iview, @@ -316,8 +355,8 @@ emit_resolve(struct radv_cmd_buffer *cmd_buffer, const VkExtent2D *resolve_extent) { struct radv_device *device = cmd_buffer->device; - const uint32_t samples = src_iview->image->info.samples; - const uint32_t samples_log2 = ffs(samples) - 1; + VkPipeline *pipeline; + radv_meta_push_descriptor_set(cmd_buffer, VK_PIPELINE_BIND_POINT_COMPUTE, device->meta_state.resolve_compute.p_layout, @@ -353,16 +392,10 @@ emit_resolve(struct radv_cmd_buffer *cmd_buffer, } }); - VkPipeline pipeline; - if (vk_format_is_int(src_iview->image->vk_format)) - pipeline = device->meta_state.resolve_compute.rc[samples_log2].i_pipeline; - else if (vk_format_is_srgb(src_iview->image->vk_format)) - pipeline = device->meta_state.resolve_compute.rc[samples_log2].srgb_pipeline; - else - pipeline = device->meta_state.resolve_compute.rc[samples_log2].pipeline; + pipeline = radv_get_resolve_pipeline(cmd_buffer, src_iview); radv_CmdBindPipeline(radv_cmd_buffer_to_handle(cmd_buffer), - VK_PIPELINE_BIND_POINT_COMPUTE, pipeline); + VK_PIPELINE_BIND_POINT_COMPUTE, *pipeline); unsigned push_constants[4] = { src_offset->x, @@ -380,8 +413,10 @@ emit_resolve(struct radv_cmd_buffer *cmd_buffer, void radv_meta_resolve_compute_image(struct radv_cmd_buffer *cmd_buffer, struct radv_image *src_image, + VkFormat src_format, VkImageLayout src_image_layout, struct radv_image *dest_image, + VkFormat dest_format, VkImageLayout dest_image_layout, uint32_t region_count, const VkImageResolve *regions) @@ -427,7 +462,7 @@ void radv_meta_resolve_compute_image(struct radv_cmd_buffer *cmd_buffer, .sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO, .image = radv_image_to_handle(src_image), .viewType = radv_meta_get_view_type(src_image), - .format = src_image->vk_format, + .format = src_format, .subresourceRange = { .aspectMask = VK_IMAGE_ASPECT_COLOR_BIT, .baseMipLevel = region->srcSubresource.mipLevel, @@ -443,7 +478,7 @@ void radv_meta_resolve_compute_image(struct radv_cmd_buffer *cmd_buffer, .sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO, .image = radv_image_to_handle(dest_image), .viewType = radv_meta_get_view_type(dest_image), - .format = vk_to_non_srgb_format(dest_image->vk_format), + .format = vk_to_non_srgb_format(dest_format), .subresourceRange = { .aspectMask = VK_IMAGE_ASPECT_COLOR_BIT, .baseMipLevel = region->dstSubresource.mipLevel, @@ -472,21 +507,8 @@ radv_cmd_buffer_resolve_subpass_cs(struct radv_cmd_buffer *cmd_buffer) { struct radv_framebuffer *fb = cmd_buffer->state.framebuffer; const struct radv_subpass *subpass = cmd_buffer->state.subpass; - struct radv_meta_saved_state saved_state; struct radv_subpass_barrier barrier; - /* FINISHME(perf): Skip clears for resolve attachments. - * - * From the Vulkan 1.0 spec: - * - * If the first use of an attachment in a render pass is as a resolve - * attachment, then the loadOp is effectively ignored as the resolve is - * guaranteed to overwrite all pixels in the render area. - */ - - if (!subpass->has_resolve) - return; - /* Resolves happen before the end-of-subpass barriers get executed, so * we have to make the attachment shader-readable. */ @@ -495,67 +517,43 @@ radv_cmd_buffer_resolve_subpass_cs(struct radv_cmd_buffer *cmd_buffer) barrier.dst_access_mask = VK_ACCESS_INPUT_ATTACHMENT_READ_BIT; radv_subpass_barrier(cmd_buffer, &barrier); - radv_decompress_resolve_subpass_src(cmd_buffer); - - radv_meta_save(&saved_state, cmd_buffer, - RADV_META_SAVE_COMPUTE_PIPELINE | - RADV_META_SAVE_CONSTANTS | - RADV_META_SAVE_DESCRIPTORS); - for (uint32_t i = 0; i < subpass->color_count; ++i) { struct radv_subpass_attachment src_att = subpass->color_attachments[i]; - struct radv_subpass_attachment dest_att = subpass->resolve_attachments[i]; - struct radv_image_view *src_iview = cmd_buffer->state.framebuffer->attachments[src_att.attachment].attachment; - struct radv_image_view *dst_iview = cmd_buffer->state.framebuffer->attachments[dest_att.attachment].attachment; - if (dest_att.attachment == VK_ATTACHMENT_UNUSED) - continue; - - struct radv_image *src_image = src_iview->image; - struct radv_image *dst_image = dst_iview->image; - for (uint32_t layer = 0; layer < src_image->info.array_size; layer++) { + struct radv_subpass_attachment dst_att = subpass->resolve_attachments[i]; + struct radv_image_view *src_iview = fb->attachments[src_att.attachment].attachment; + struct radv_image_view *dst_iview = fb->attachments[dst_att.attachment].attachment; - struct radv_image_view tsrc_iview; - radv_image_view_init(&tsrc_iview, cmd_buffer->device, - &(VkImageViewCreateInfo) { - .sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO, - .image = radv_image_to_handle(src_image), - .viewType = radv_meta_get_view_type(src_image), - .format = src_image->vk_format, - .subresourceRange = { - .aspectMask = VK_IMAGE_ASPECT_COLOR_BIT, - .baseMipLevel = src_iview->base_mip, - .levelCount = 1, - .baseArrayLayer = layer, - .layerCount = 1, - }, - }); + if (dst_att.attachment == VK_ATTACHMENT_UNUSED) + continue; - struct radv_image_view tdst_iview; - radv_image_view_init(&tdst_iview, cmd_buffer->device, - &(VkImageViewCreateInfo) { - .sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO, - .image = radv_image_to_handle(dst_image), - .viewType = radv_meta_get_view_type(dst_image), - .format = vk_to_non_srgb_format(dst_image->vk_format), - .subresourceRange = { - .aspectMask = VK_IMAGE_ASPECT_COLOR_BIT, - .baseMipLevel = dst_iview->base_mip, - .levelCount = 1, - .baseArrayLayer = layer, - .layerCount = 1, - }, - }); - emit_resolve(cmd_buffer, - &tsrc_iview, - &tdst_iview, - &(VkOffset2D) { 0, 0 }, - &(VkOffset2D) { 0, 0 }, - &(VkExtent2D) { fb->width, fb->height }); - } + VkImageResolve region = { + .extent = (VkExtent3D){ fb->width, fb->height, 0 }, + .srcSubresource = (VkImageSubresourceLayers) { + .aspectMask = VK_IMAGE_ASPECT_COLOR_BIT, + .mipLevel = src_iview->base_mip, + .baseArrayLayer = 0, + .layerCount = src_iview->image->info.array_size + }, + .dstSubresource = (VkImageSubresourceLayers) { + .aspectMask = VK_IMAGE_ASPECT_COLOR_BIT, + .mipLevel = dst_iview->base_mip, + .baseArrayLayer = 0, + .layerCount = dst_iview->image->info.array_size + }, + .srcOffset = (VkOffset3D){ 0, 0, 0 }, + .dstOffset = (VkOffset3D){ 0, 0, 0 }, + }; + + radv_meta_resolve_compute_image(cmd_buffer, + src_iview->image, + src_iview->vk_format, + src_att.layout, + dst_iview->image, + dst_iview->vk_format, + dst_att.layout, + 1, ®ion); } cmd_buffer->state.flush_bits |= RADV_CMD_FLAG_CS_PARTIAL_FLUSH | RADV_CMD_FLAG_INV_VMEM_L1; - - radv_meta_restore(&saved_state, cmd_buffer); }