Emulation/ppsspp
1
0
Fork 0
mirror of https://github.com/hrydgard/ppsspp.git synced 2025-04-02 11:01:50 -04:00
Code Issues Projects Releases Packages Wiki Activity

Merge pull request #13265 from hrydgard/depth-texturing-prep

Browse source 
Depth texturing preparations
This commit is contained in:
Henrik Rydgård 2020-08-09 20:30:49 +02:00 • committed by GitHub
commit 4de34de57f
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
10 changed files with 137 additions and 29 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

12
GPU/Common/GPUStateUtils.cpp
View file

@ -531,14 +531,20 @@ float ToScaledDepthFromIntegerScale(float z) {
}
}
float FromScaledDepth(float z) {
// See struct DepthScaleFactors for how to apply.
DepthScaleFactors GetDepthScaleFactors() {
DepthScaleFactors factors;
if (!gstate_c.Supports(GPU_SUPPORTS_ACCURATE_DEPTH)) {
return z * 65535.0f;
factors.offset = 0;
factors.scale = 65535.0f;
return factors;
}
const float depthSliceFactor = DepthSliceFactor();
const float offset = 0.5f * (depthSliceFactor - 1.0f) * (1.0f / depthSliceFactor);
return (z - offset) * depthSliceFactor * 65535.0f;
factors.scale = depthSliceFactor * 65535.0f;
factors.offset = offset;
return factors;
}
void ConvertViewportAndScissor(bool useBufferedRendering, float renderWidth, float renderHeight, int bufferWidth, int bufferHeight, ViewportAndScissor &out) {

13
GPU/Common/GPUStateUtils.h
View file

@ -72,7 +72,18 @@ struct ViewportAndScissor {
};
void ConvertViewportAndScissor(bool useBufferedRendering, float renderWidth, float renderHeight, int bufferWidth, int bufferHeight, ViewportAndScissor &out);
float ToScaledDepthFromIntegerScale(float z);
float FromScaledDepth(float z);
// Use like this: (z - offset) * scale
struct DepthScaleFactors {
float offset;
float scale;
float Apply(float z) const {
return (z - offset) * scale;
}
};
DepthScaleFactors GetDepthScaleFactors();
float DepthSliceFactor();
// These are common to all modern APIs and can be easily converted with a lookup table.

2
GPU/Common/TextureCacheCommon.h
View file

@ -119,6 +119,8 @@ struct TexCacheEntry {
STATUS_FREE_CHANGE = 0x200, // Allow one change before marking "frequent".
STATUS_BAD_MIPS = 0x400, // Has bad or unusable mipmap levels.
STATUS_DEPTH = 0x800,
};
// Status, but int so we can zero initialize.

3
GPU/Directx9/FramebufferManagerDX9.cpp
View file

@ -538,12 +538,13 @@ static const D3DVERTEXELEMENT9 g_FramebufferVertexElements[] = {
const u32 *packed = (const u32 *)locked.pBits;
u16 *depth = (u16 *)Memory::GetPointer(z_address);
DepthScaleFactors depthScale = GetDepthScaleFactors();
// TODO: Optimize.
for (int yp = 0; yp < h; ++yp) {
for (int xp = 0; xp < w; ++xp) {
const int offset = (yp + y) * vfb->z_stride + x + xp;
float scaled = FromScaledDepth((packed[offset] & 0x00FFFFFF) * (1.0f / 16777215.0f));
float scaled = depthScale.Apply((packed[offset] & 0x00FFFFFF) * (1.0f / 16777215.0f));
if (scaled <= 0.0f) {
depth[offset] = 0;
} else if (scaled >= 65535.0f) {

5
GPU/GLES/DepthBufferGLES.cpp
View file

@ -169,6 +169,7 @@ void FramebufferManagerGLES::PackDepthbuffer(VirtualFramebuffer *vfb, int x, int
int totalPixels = h == 1 ? packWidth : vfb->z_stride * h;
if (format16Bit) {
// TODO: We have to apply GetDepthScaleFactors here too, right?
for (int yp = 0; yp < h; ++yp) {
int row_offset = vfb->z_stride * yp;
for (int xp = 0; xp < packWidth; ++xp) {
@ -177,11 +178,13 @@ void FramebufferManagerGLES::PackDepthbuffer(VirtualFramebuffer *vfb, int x, int
}
}
} else {
// TODO: Apply this in the shader.
DepthScaleFactors depthScale = GetDepthScaleFactors();
for (int yp = 0; yp < h; ++yp) {
int row_offset = vfb->z_stride * yp;
for (int xp = 0; xp < packWidth; ++xp) {
const int i = row_offset + xp;
float scaled = FromScaledDepth(packedf[i]);
float scaled = depthScale.Apply(packedf[i]);
if (scaled <= 0.0f) {
depth[i] = 0;
} else if (scaled >= 65535.0f) {

18
Windows/GEDebugger/GEDebugger.cpp
View file

@ -546,9 +546,12 @@ void CGEDebugger::DescribePixel(u32 pix, GPUDebugBufferFormat fmt, int x, int y,
break;
case GPU_DBG_FORMAT_24BIT_8X:
{
DepthScaleFactors depthScale = GetDepthScaleFactors();
// These are only ever going to be depth values, so let's also show scaled to 16 bit.
_snwprintf(desc, 256, L"%d,%d: %d / %f / %f", x, y, pix & 0x00FFFFFF, (pix & 0x00FFFFFF) * (1.0f / 16777215.0f), FromScaledDepth((pix & 0x00FFFFFF) * (1.0f / 16777215.0f)));
_snwprintf(desc, 256, L"%d,%d: %d / %f / %f", x, y, pix & 0x00FFFFFF, (pix & 0x00FFFFFF) * (1.0f / 16777215.0f), depthScale.Apply((pix & 0x00FFFFFF) * (1.0f / 16777215.0f)));
break;
}
case GPU_DBG_FORMAT_24BIT_8X_DIV_256:
{
@ -563,15 +566,24 @@ void CGEDebugger::DescribePixel(u32 pix, GPUDebugBufferFormat fmt, int x, int y,
_snwprintf(desc, 256, L"%d,%d: %d / %f", x, y, (pix >> 24) & 0xFF, ((pix >> 24) & 0xFF) * (1.0f / 255.0f));
break;
case GPU_DBG_FORMAT_FLOAT:
_snwprintf(desc, 256, L"%d,%d: %f / %f", x, y, *(float *)&pix, FromScaledDepth(*(float *)&pix));
case GPU_DBG_FORMAT_FLOAT: {
float pixf = *(float *)&pix;
DepthScaleFactors depthScale = GetDepthScaleFactors();
_snwprintf(desc, 256, L"%d,%d: %f / %f", x, y, pixf, depthScale.Apply(pixf));
break;
}
case GPU_DBG_FORMAT_FLOAT_DIV_256:
{
double z = *(float *)&pix;
int z24 = (int)(z * 16777215.0);
DepthScaleFactors factors = GetDepthScaleFactors();
// TODO: Use GetDepthScaleFactors here too, verify it's the same.
int z16 = z24 - 0x800000 + 0x8000;
int z16_2 = factors.Apply(z);
_snwprintf(desc, 256, L"%d,%d: %d / %f", x, y, z16, (z - 0.5 + (1.0 / 512.0)) * 256.0);
}
break;

47
ext/native/thin3d/VulkanQueueRunner.cpp
View file

@ -949,9 +949,8 @@ void VulkanQueueRunner::LogReadbackImage(const VKRStep &step) {
void VulkanQueueRunner::PerformRenderPass(const VKRStep &step, VkCommandBuffer cmd) {
// TODO: If there are multiple, we can transition them together.
for (const auto &iter : step.preTransitions) {
if (iter.fb->color.layout != iter.targetLayout) {
VkImageMemoryBarrier barrier{};
barrier.sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER;
if (iter.aspect == VK_IMAGE_ASPECT_COLOR_BIT && iter.fb->color.layout != iter.targetLayout) {
VkImageMemoryBarrier barrier{ VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER };
barrier.oldLayout = iter.fb->color.layout;
barrier.subresourceRange.layerCount = 1;
barrier.subresourceRange.levelCount = 1;
@ -992,9 +991,49 @@ void VulkanQueueRunner::PerformRenderPass(const VKRStep &step, VkCommandBuffer c
barrier.subresourceRange.aspectMask = VK_IMAGE_ASPECT_COLOR_BIT;
barrier.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED;
barrier.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED;
vkCmdPipelineBarrier(cmd, srcStage, dstStage, 0, 0, nullptr, 0, nullptr, 1, &barrier);
iter.fb->color.layout = barrier.newLayout;
} else if (iter.aspect == VK_IMAGE_ASPECT_DEPTH_BIT && iter.fb->depth.layout != iter.targetLayout) {
VkImageMemoryBarrier barrier{ VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER };
barrier.oldLayout = iter.fb->depth.layout;
barrier.subresourceRange.layerCount = 1;
barrier.subresourceRange.levelCount = 1;
barrier.image = iter.fb->depth.image;
barrier.srcAccessMask = 0;
VkPipelineStageFlags srcStage;
VkPipelineStageFlags dstStage;
switch (barrier.oldLayout) {
case VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL:
barrier.srcAccessMask = VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT | VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_READ_BIT;
srcStage = VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT | VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT;
break;
case VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL:
barrier.srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT;
srcStage = VK_PIPELINE_STAGE_TRANSFER_BIT;
break;
case VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL:
barrier.srcAccessMask = VK_ACCESS_TRANSFER_READ_BIT;
srcStage = VK_PIPELINE_STAGE_TRANSFER_BIT;
break;
default:
Crash();
break;
}
barrier.newLayout = iter.targetLayout;
switch (barrier.newLayout) {
case VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL:
barrier.dstAccessMask = VK_ACCESS_SHADER_READ_BIT;
dstStage = VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT;
break;
default:
Crash();
break;
}
barrier.subresourceRange.aspectMask = VK_IMAGE_ASPECT_DEPTH_BIT | VK_IMAGE_ASPECT_STENCIL_BIT;
barrier.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED;
barrier.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED;
vkCmdPipelineBarrier(cmd, srcStage, dstStage, 0, 0, nullptr, 0, nullptr, 1, &barrier);
iter.fb->depth.layout = barrier.newLayout;
}
}

2
ext/native/thin3d/VulkanQueueRunner.h
View file

@ -107,6 +107,7 @@ enum class VKRRenderPassAction : uint8_t {
};
struct TransitionRequest {
VkImageAspectFlags aspect; // COLOR or DEPTH
VKRFramebuffer *fb;
VkImageLayout targetLayout;
};
@ -141,6 +142,7 @@ struct VKRStep {
// Downloads and textures from this pass.
int numReads;
VkImageLayout finalColorLayout;
VkImageLayout finalDepthStencilLayout;
} render;
struct {
VKRFramebuffer *src;

56
ext/native/thin3d/VulkanRenderManager.cpp
View file

@ -580,7 +580,7 @@ void VulkanRenderManager::BindFramebufferAsRenderTarget(VKRFramebuffer *fb, VKRR
}
}
bool VulkanRenderManager::CopyFramebufferToMemorySync(VKRFramebuffer *src, int aspectBits, int x, int y, int w, int h, Draw::DataFormat destFormat, uint8_t *pixels, int pixelStride, const char *tag) {
bool VulkanRenderManager::CopyFramebufferToMemorySync(VKRFramebuffer *src, VkImageAspectFlags aspectBits, int x, int y, int w, int h, Draw::DataFormat destFormat, uint8_t *pixels, int pixelStride, const char *tag) {
_dbg_assert_(insideFrame_);
for (int i = (int)steps_.size() - 1; i >= 0; i--) {
if (steps_[i]->stepType == VKRStepType::RENDER && steps_[i]->render.framebuffer == src) {
@ -869,7 +869,7 @@ void VulkanRenderManager::Clear(uint32_t clearColor, float clearZ, int clearSten
}
}
void VulkanRenderManager::CopyFramebuffer(VKRFramebuffer *src, VkRect2D srcRect, VKRFramebuffer *dst, VkOffset2D dstPos, int aspectMask, const char *tag) {
void VulkanRenderManager::CopyFramebuffer(VKRFramebuffer *src, VkRect2D srcRect, VKRFramebuffer *dst, VkOffset2D dstPos, VkImageAspectFlags aspectMask, const char *tag) {
_dbg_assert_msg_(srcRect.offset.x >= 0, "srcrect offset x (%d) < 0", srcRect.offset.x);
_dbg_assert_msg_(srcRect.offset.y >= 0, "srcrect offset y (%d) < 0", srcRect.offset.y);
_dbg_assert_msg_(srcRect.offset.x + srcRect.extent.width <= (uint32_t)src->width, "srcrect offset x (%d) + extent (%d) > width (%d)", srcRect.offset.x, srcRect.extent.width, (uint32_t)src->width);
@ -885,8 +885,15 @@ void VulkanRenderManager::CopyFramebuffer(VKRFramebuffer *src, VkRect2D srcRect,
for (int i = (int)steps_.size() - 1; i >= 0; i--) {
if (steps_[i]->stepType == VKRStepType::RENDER && steps_[i]->render.framebuffer == src) {
if (steps_[i]->render.finalColorLayout == VK_IMAGE_LAYOUT_UNDEFINED) {
steps_[i]->render.finalColorLayout = VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL;
if (aspectMask & VK_IMAGE_ASPECT_COLOR_BIT) {
if (steps_[i]->render.finalColorLayout == VK_IMAGE_LAYOUT_UNDEFINED) {
steps_[i]->render.finalColorLayout = VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL;
}
}
if (aspectMask & (VK_IMAGE_ASPECT_DEPTH_BIT | VK_IMAGE_ASPECT_STENCIL_BIT)) {
if (steps_[i]->render.finalDepthStencilLayout == VK_IMAGE_LAYOUT_UNDEFINED) {
steps_[i]->render.finalDepthStencilLayout = VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL;
}
}
steps_[i]->render.numReads++;
break;
@ -894,8 +901,15 @@ void VulkanRenderManager::CopyFramebuffer(VKRFramebuffer *src, VkRect2D srcRect,
}
for (int i = (int)steps_.size() - 1; i >= 0; i--) {
if (steps_[i]->stepType == VKRStepType::RENDER && steps_[i]->render.framebuffer == dst) {
if (steps_[i]->render.finalColorLayout == VK_IMAGE_LAYOUT_UNDEFINED) {
steps_[i]->render.finalColorLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL;
if (aspectMask & VK_IMAGE_ASPECT_COLOR_BIT) {
if (steps_[i]->render.finalColorLayout == VK_IMAGE_LAYOUT_UNDEFINED) {
steps_[i]->render.finalColorLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL;
}
}
if (aspectMask & (VK_IMAGE_ASPECT_DEPTH_BIT | VK_IMAGE_ASPECT_STENCIL_BIT)) {
if (steps_[i]->render.finalDepthStencilLayout == VK_IMAGE_LAYOUT_UNDEFINED) {
steps_[i]->render.finalDepthStencilLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL;
}
}
break;
}
@ -919,7 +933,7 @@ void VulkanRenderManager::CopyFramebuffer(VKRFramebuffer *src, VkRect2D srcRect,
curRenderStep_ = nullptr;
}
void VulkanRenderManager::BlitFramebuffer(VKRFramebuffer *src, VkRect2D srcRect, VKRFramebuffer *dst, VkRect2D dstRect, int aspectMask, VkFilter filter, const char *tag) {
void VulkanRenderManager::BlitFramebuffer(VKRFramebuffer *src, VkRect2D srcRect, VKRFramebuffer *dst, VkRect2D dstRect, VkImageAspectFlags aspectMask, VkFilter filter, const char *tag) {
_dbg_assert_msg_(srcRect.offset.x >= 0, "srcrect offset x (%d) < 0", srcRect.offset.x);
_dbg_assert_msg_(srcRect.offset.y >= 0, "srcrect offset y (%d) < 0", srcRect.offset.y);
_dbg_assert_msg_(srcRect.offset.x + srcRect.extent.width <= (uint32_t)src->width, "srcrect offset x (%d) + extent (%d) > width (%d)", srcRect.offset.x, srcRect.extent.width, (uint32_t)src->width);
@ -936,6 +950,8 @@ void VulkanRenderManager::BlitFramebuffer(VKRFramebuffer *src, VkRect2D srcRect,
_dbg_assert_msg_(dstRect.extent.width > 0, "blit dstwidth == 0");
_dbg_assert_msg_(dstRect.extent.height > 0, "blit dstheight == 0");
// TODO: Seem to be missing final layouts here like in Copy...
for (int i = (int)steps_.size() - 1; i >= 0; i--) {
if (steps_[i]->stepType == VKRStepType::RENDER && steps_[i]->render.framebuffer == src) {
steps_[i]->render.numReads++;
@ -962,15 +978,31 @@ void VulkanRenderManager::BlitFramebuffer(VKRFramebuffer *src, VkRect2D srcRect,
curRenderStep_ = nullptr;
}
VkImageView VulkanRenderManager::BindFramebufferAsTexture(VKRFramebuffer *fb, int binding, int aspectBit, int attachment) {
VkImageView VulkanRenderManager::BindFramebufferAsTexture(VKRFramebuffer *fb, int binding, VkImageAspectFlags aspectBit, int attachment) {
_dbg_assert_(curRenderStep_ != nullptr);
// Mark the dependency, check for required transitions, and return the image.
for (int i = (int)steps_.size() - 1; i >= 0; i--) {
if (steps_[i]->stepType == VKRStepType::RENDER && steps_[i]->render.framebuffer == fb) {
// If this framebuffer was rendered to earlier in this frame, make sure to pre-transition it to the correct layout.
if (steps_[i]->render.finalColorLayout == VK_IMAGE_LAYOUT_UNDEFINED) {
steps_[i]->render.finalColorLayout = VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL;
if (aspectBit & VK_IMAGE_ASPECT_COLOR_BIT) {
// If this framebuffer was rendered to earlier in this frame, make sure to pre-transition it to the correct layout.
if (steps_[i]->render.finalColorLayout == VK_IMAGE_LAYOUT_UNDEFINED) {
steps_[i]->render.finalColorLayout = VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL;
break;
} else if (steps_[i]->render.finalColorLayout != VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL) {
_assert_msg_(false, "Unexpected color layout %d", (int)steps_[i]->render.finalColorLayout);
// May need to shadow the framebuffer if we re-order passes later.
}
}
if (aspectBit & (VK_IMAGE_ASPECT_DEPTH_BIT | VK_IMAGE_ASPECT_STENCIL_BIT)) {
// If this framebuffer was rendered to earlier in this frame, make sure to pre-transition it to the correct layout.
if (steps_[i]->render.finalDepthStencilLayout == VK_IMAGE_LAYOUT_UNDEFINED) {
steps_[i]->render.finalDepthStencilLayout = VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL;
break;
} else if (steps_[i]->render.finalDepthStencilLayout != VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL) {
_assert_msg_(false, "Unexpected depth layout %d", (int)steps_[i]->render.finalDepthStencilLayout);
// May need to shadow the framebuffer if we re-order passes later.
}
}
steps_[i]->render.numReads++;
break;
@ -986,7 +1018,7 @@ VkImageView VulkanRenderManager::BindFramebufferAsTexture(VKRFramebuffer *fb, in
// We're done.
return fb->color.imageView;
} else {
curRenderStep_->preTransitions.push_back({ fb, VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL });
curRenderStep_->preTransitions.push_back({ aspectBit, fb, VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL });
return fb->color.imageView;
}
}

8
ext/native/thin3d/VulkanRenderManager.h
View file

@ -130,13 +130,13 @@ public:
// Returns an ImageView corresponding to a framebuffer. Is called BindFramebufferAsTexture to maintain a similar interface
// as the other backends, even though there's no actual binding happening here.
VkImageView BindFramebufferAsTexture(VKRFramebuffer *fb, int binding, int aspectBit, int attachment);
VkImageView BindFramebufferAsTexture(VKRFramebuffer *fb, int binding, VkImageAspectFlags aspectBits, int attachment);
bool CopyFramebufferToMemorySync(VKRFramebuffer *src, int aspectBits, int x, int y, int w, int h, Draw::DataFormat destFormat, uint8_t *pixels, int pixelStride, const char *tag);
bool CopyFramebufferToMemorySync(VKRFramebuffer *src, VkImageAspectFlags aspectBits, int x, int y, int w, int h, Draw::DataFormat destFormat, uint8_t *pixels, int pixelStride, const char *tag);
void CopyImageToMemorySync(VkImage image, int mipLevel, int x, int y, int w, int h, Draw::DataFormat destFormat, uint8_t *pixels, int pixelStride, const char *tag);
void CopyFramebuffer(VKRFramebuffer *src, VkRect2D srcRect, VKRFramebuffer *dst, VkOffset2D dstPos, int aspectMask, const char *tag);
void BlitFramebuffer(VKRFramebuffer *src, VkRect2D srcRect, VKRFramebuffer *dst, VkRect2D dstRect, int aspectMask, VkFilter filter, const char *tag);
void CopyFramebuffer(VKRFramebuffer *src, VkRect2D srcRect, VKRFramebuffer *dst, VkOffset2D dstPos, VkImageAspectFlags aspectMask, const char *tag);
void BlitFramebuffer(VKRFramebuffer *src, VkRect2D srcRect, VKRFramebuffer *dst, VkRect2D dstRect, VkImageAspectFlags aspectMask, VkFilter filter, const char *tag);
void BindPipeline(VkPipeline pipeline) {
_dbg_assert_(curRenderStep_ && curRenderStep_->stepType == VKRStepType::RENDER);