#include "sid.h"
#include "vk_format.h"
+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_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_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;
+ gtvals[i] = nir_fmul(b, gtvals[i],
+ nir_build_imm(b, 1, 32, v));
+ v.f32[0] = 0.055;
+ gtvals[i] = nir_fsub(b, gtvals[i],
+ nir_build_imm(b, 1, 32, v));
+ }
+
+ nir_ssa_def *comp[4];
+ for (i = 0; i < 3; i++)
+ comp[i] = nir_bcsel(b, cmp[i], ltvals[i], gtvals[i]);
+ comp[3] = nir_channels(b, input, 1 << 3);
+ return nir_vec(b, comp, 4);
+}
+
static nir_shader *
build_resolve_compute_shader(struct radv_device *dev, bool is_integer, bool is_srgb, int samples)
{
GLSL_TYPE_FLOAT);
snprintf(name, 64, "meta_resolve_cs-%d-%s", samples, is_integer ? "int" : (is_srgb ? "srgb" : "float"));
nir_builder_init_simple_shader(&b, NULL, MESA_SHADER_COMPUTE, NULL);
- b.shader->info->name = ralloc_strdup(b.shader, name);
- b.shader->info->cs.local_size[0] = 16;
- b.shader->info->cs.local_size[1] = 16;
- b.shader->info->cs.local_size[2] = 1;
+ b.shader->info.name = ralloc_strdup(b.shader, name);
+ b.shader->info.cs.local_size[0] = 16;
+ b.shader->info.cs.local_size[1] = 16;
+ b.shader->info.cs.local_size[2] = 1;
nir_variable *input_img = nir_variable_create(b.shader, nir_var_uniform,
sampler_type, "s_tex");
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 *block_size = nir_imm_ivec4(&b,
- b.shader->info->cs.local_size[0],
- b.shader->info->cs.local_size[1],
- b.shader->info->cs.local_size[2], 0);
+ b.shader->info.cs.local_size[0],
+ b.shader->info.cs.local_size[1],
+ b.shader->info.cs.local_size[2], 0);
nir_ssa_def *global_id = nir_iadd(&b, nir_imul(&b, wg_id, block_size), invoc_id);
nir_intrinsic_instr *src_offset = nir_intrinsic_instr_create(b.shader, nir_intrinsic_load_push_constant);
+ nir_intrinsic_set_base(src_offset, 0);
+ nir_intrinsic_set_range(src_offset, 16);
src_offset->src[0] = nir_src_for_ssa(nir_imm_int(&b, 0));
src_offset->num_components = 2;
nir_ssa_dest_init(&src_offset->instr, &src_offset->dest, 2, 32, "src_offset");
nir_builder_instr_insert(&b, &src_offset->instr);
nir_intrinsic_instr *dst_offset = nir_intrinsic_instr_create(b.shader, nir_intrinsic_load_push_constant);
+ nir_intrinsic_set_base(dst_offset, 0);
+ nir_intrinsic_set_range(dst_offset, 16);
dst_offset->src[0] = nir_src_for_ssa(nir_imm_int(&b, 8));
dst_offset->num_components = 2;
nir_ssa_dest_init(&dst_offset->instr, &dst_offset->dest, 2, 32, "dst_offset");
nir_ssa_def *img_coord = nir_channels(&b, nir_iadd(&b, global_id, &src_offset->dest.ssa), 0x3);
nir_variable *color = nir_local_variable_create(b.impl, glsl_vec4_type(), "color");
- radv_meta_build_resolve_shader_core(&b, is_integer, is_srgb, samples,
- input_img, color, img_coord);
+ radv_meta_build_resolve_shader_core(&b, is_integer, samples, input_img,
+ color, img_coord);
nir_ssa_def *outval = nir_load_var(&b, color);
+ if (is_srgb)
+ outval = radv_meta_build_resolve_srgb_conversion(&b, outval);
+
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_store);
store->src[0] = nir_src_for_ssa(coord);
{
struct radv_meta_state *state = &device->meta_state;
VkResult res;
- memset(&device->meta_state.resolve_compute, 0, sizeof(device->meta_state.resolve_compute));
res = create_layout(device);
if (res != VK_SUCCESS)
- return res;
+ goto fail;
for (uint32_t i = 0; i < MAX_SAMPLES_LOG2; ++i) {
uint32_t samples = 1 << i;
res = create_resolve_pipeline(device, samples, false, false,
&state->resolve_compute.rc[i].pipeline);
+ if (res != VK_SUCCESS)
+ goto fail;
res = create_resolve_pipeline(device, samples, true, false,
&state->resolve_compute.rc[i].i_pipeline);
+ if (res != VK_SUCCESS)
+ goto fail;
res = create_resolve_pipeline(device, samples, false, true,
&state->resolve_compute.rc[i].srgb_pipeline);
+ if (res != VK_SUCCESS)
+ goto fail;
}
+ return VK_SUCCESS;
+fail:
+ radv_device_finish_meta_resolve_compute_state(device);
return res;
}
pipeline = device->meta_state.resolve_compute.rc[samples_log2].srgb_pipeline;
else
pipeline = device->meta_state.resolve_compute.rc[samples_log2].pipeline;
- if (cmd_buffer->state.compute_pipeline != radv_pipeline_from_handle(pipeline)) {
- radv_CmdBindPipeline(radv_cmd_buffer_to_handle(cmd_buffer),
- VK_PIPELINE_BIND_POINT_COMPUTE, pipeline);
- }
+
+ radv_CmdBindPipeline(radv_cmd_buffer_to_handle(cmd_buffer),
+ VK_PIPELINE_BIND_POINT_COMPUTE, pipeline);
unsigned push_constants[4] = {
src_offset->x,
uint32_t region_count,
const VkImageResolve *regions)
{
- struct radv_meta_saved_compute_state saved_state;
+ struct radv_meta_saved_state saved_state;
for (uint32_t r = 0; r < region_count; ++r) {
const VkImageResolve *region = ®ions[r];
radv_fast_clear_flush_image_inplace(cmd_buffer, src_image, &range);
}
- radv_meta_save_compute(&saved_state, cmd_buffer, 16);
+ radv_meta_save(&saved_state, cmd_buffer,
+ RADV_META_SAVE_COMPUTE_PIPELINE |
+ RADV_META_SAVE_CONSTANTS |
+ RADV_META_SAVE_DESCRIPTORS);
for (uint32_t r = 0; r < region_count; ++r) {
const VkImageResolve *region = ®ions[r];
.baseArrayLayer = src_base_layer + layer,
.layerCount = 1,
},
- },
- cmd_buffer, VK_IMAGE_USAGE_SAMPLED_BIT);
+ });
struct radv_image_view dest_iview;
radv_image_view_init(&dest_iview, cmd_buffer->device,
.sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO,
.image = radv_image_to_handle(dest_image),
.viewType = radv_meta_get_view_type(dest_image),
- .format = dest_image->vk_format,
+ .format = vk_to_non_srgb_format(dest_image->vk_format),
.subresourceRange = {
.aspectMask = VK_IMAGE_ASPECT_COLOR_BIT,
.baseMipLevel = region->dstSubresource.mipLevel,
.baseArrayLayer = dest_base_layer + layer,
.layerCount = 1,
},
- },
- cmd_buffer, VK_IMAGE_USAGE_STORAGE_BIT);
+ });
emit_resolve(cmd_buffer,
&src_iview,
&(VkExtent2D) {extent.width, extent.height });
}
}
- radv_meta_restore_compute(&saved_state, cmd_buffer, 16);
+ radv_meta_restore(&saved_state, cmd_buffer);
+}
+
+/**
+ * Emit any needed resolves for the current subpass.
+ */
+void
+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;
+ /* 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 */
+ cmd_buffer->state.flush_bits |= RADV_CMD_FLAG_PS_PARTIAL_FLUSH |
+ RADV_CMD_FLAG_FLUSH_AND_INV_CB |
+ RADV_CMD_FLAG_FLUSH_AND_INV_CB_META |
+ RADV_CMD_FLAG_INV_GLOBAL_L2 |
+ RADV_CMD_FLAG_INV_VMEM_L1;
+
+ for (uint32_t i = 0; i < subpass->color_count; ++i) {
+ VkAttachmentReference src_att = subpass->color_attachments[i];
+ VkAttachmentReference dest_att = subpass->resolve_attachments[i];
+
+ if (src_att.attachment == VK_ATTACHMENT_UNUSED ||
+ dest_att.attachment == VK_ATTACHMENT_UNUSED)
+ continue;
+
+ struct radv_image_view *src_iview = cmd_buffer->state.framebuffer->attachments[src_att.attachment].attachment;
+
+ VkImageSubresourceRange range;
+ range.aspectMask = VK_IMAGE_ASPECT_COLOR_BIT;
+ range.baseMipLevel = 0;
+ range.levelCount = 1;
+ range.baseArrayLayer = 0;
+ range.layerCount = 1;
+ radv_fast_clear_flush_image_inplace(cmd_buffer, src_iview->image, &range);
+ }
+
+ 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) {
+ VkAttachmentReference src_att = subpass->color_attachments[i];
+ VkAttachmentReference 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;
+
+ emit_resolve(cmd_buffer,
+ src_iview,
+ dst_iview,
+ &(VkOffset2D) { 0, 0 },
+ &(VkOffset2D) { 0, 0 },
+ &(VkExtent2D) { fb->width, fb->height });
+ }
+
+ radv_meta_restore(&saved_state, cmd_buffer);
+
+ for (uint32_t i = 0; i < subpass->color_count; ++i) {
+ VkAttachmentReference dest_att = subpass->resolve_attachments[i];
+ struct radv_image *dst_img = cmd_buffer->state.framebuffer->attachments[dest_att.attachment].attachment->image;
+ if (dest_att.attachment == VK_ATTACHMENT_UNUSED)
+ continue;
+ VkImageSubresourceRange range;
+ range.aspectMask = VK_IMAGE_ASPECT_COLOR_BIT;
+ range.baseMipLevel = 0;
+ range.levelCount = 1;
+ range.baseArrayLayer = 0;
+ range.layerCount = 1;
+ radv_fast_clear_flush_image_inplace(cmd_buffer, dst_img, &range);
+ }
}
projects / mesa.git / blobdiff