dahjah/Cemu
1
0
Fork 0
mirror of https://github.com/cemu-project/Cemu.git synced 2026-04-25 20:26:19 -06:00
Code Issues Packages Projects Releases Wiki Activity

condense attribute pointer cache functions and move handleSpecialState8_clearAsDepth to common renderer

Browse Source
This commit is contained in:
emiyl 2026-04-21 00:49:41 +01:00
parent 828ddea0a8
commit 4be30f6ef9
5 changed files with 109 additions and 133 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

137
src/Cafe/HW/Latte/Renderer/Common/CommonRendererCore.cpp
View File

@ -1,4 +1,12 @@
#include "CommonRendererCore.h"
#include "Cafe/HW/Latte/Renderer/Renderer.h"
#include "Cafe/HW/Latte/ISA/RegDefines.h"
#include "config/CemuConfig.h"
#ifdef ENABLE_OPENGL
#include "Common/GLInclude/glFunctions.h"
#include "Cafe/HW/Latte/Renderer/OpenGL/LatteTextureViewGL.h"
#endif
indexState_t indexState = { 0 };
AttributePointerCacheEntry_t activeAttributePointer[LATTE_VS_ATTRIBUTE_LIMIT] = { 0 };
@ -8,54 +16,35 @@ indexDataCacheEntry2_t* indexDataCacheFirst = nullptr; // points to least recent
indexDataCacheEntry2_t* indexDataCacheLast = nullptr; // points to most recently used item
sint32 indexDataCacheEntryCount = 0;
uint8* CommonRenderer_getAttributePointerCacheVboOutput(uint32 attributeShaderLoc)
void CommonRenderer_resetAttributePointerCache()
{
return activeAttributePointer[attributeShaderLoc].vboOutput;
for (sint32 i = 0; i < LATTE_VS_ATTRIBUTE_LIMIT; i++)
{
activeAttributePointer[i].vboOutput = (uint8*)-1;
activeAttributePointer[i].vboStride = (uint32)-1;
}
}
void CommonRenderer_setAttributePointerCacheVboOutput(uint32 attributeShaderLoc, uint8* vboOutput)
bool CommonRenderer_checkIfAttributePointerCacheChanged(uint32 attributeShaderLoc, uint8* vboOutput, uint32 vboStride, uint8 dataFormat, uint8 nfa, bool isSigned)
{
// don't call glVertexAttribPointer if parameters have not changed
if (
activeAttributePointer[attributeShaderLoc].vboOutput == vboOutput &&
activeAttributePointer[attributeShaderLoc].vboStride == vboStride &&
activeAttributePointer[attributeShaderLoc].dataFormat == dataFormat &&
activeAttributePointer[attributeShaderLoc].nfa == nfa &&
activeAttributePointer[attributeShaderLoc].isSigned == isSigned
)
{
return false;
}
activeAttributePointer[attributeShaderLoc].vboOutput = vboOutput;
}
uint32 CommonRenderer_getAttributePointerCacheVboStride(uint32 attributeShaderLoc)
{
return activeAttributePointer[attributeShaderLoc].vboStride;
}
void CommonRenderer_setAttributePointerCacheVboStride(uint32 attributeShaderLoc, uint32 vboStride)
{
activeAttributePointer[attributeShaderLoc].vboStride = vboStride;
}
uint8 CommonRenderer_getAttributePointerCacheDataFormat(uint32 attributeShaderLoc)
{
return activeAttributePointer[attributeShaderLoc].dataFormat;
}
void CommonRenderer_setAttributePointerCacheDataFormat(uint32 attributeShaderLoc, uint8 dataFormat)
{
activeAttributePointer[attributeShaderLoc].dataFormat = dataFormat;
}
uint8 CommonRenderer_getAttributePointerCacheNfa(uint32 attributeShaderLoc)
{
return activeAttributePointer[attributeShaderLoc].nfa;
}
void CommonRenderer_setAttributePointerCacheNfa(uint32 attributeShaderLoc, uint8 nfa)
{
activeAttributePointer[attributeShaderLoc].nfa = nfa;
}
bool CommonRenderer_getAttributePointerCacheIsSigned(uint32 attributeShaderLoc)
{
return activeAttributePointer[attributeShaderLoc].isSigned;
}
void CommonRenderer_setAttributePointerCacheIsSigned(uint32 attributeShaderLoc, bool isSigned)
{
activeAttributePointer[attributeShaderLoc].isSigned = isSigned;
return true;
}
indexState_t* CommonRenderer_getIndexState()
@ -152,4 +141,74 @@ void LatteDraw_cleanupAfterFrame()
CommonRenderer_removeFromBucket(entry);
free(entry);
}
}
void LatteDraw_handleSpecialState8_clearAsDepth()
{
if (LatteGPUState.contextNew.GetSpecialStateValues()[0] == 0)
cemuLog_logDebug(LogType::Force, "Special state 8 requires special state 0 but it is not set?");
// get depth buffer information
uint32 regDepthBuffer = LatteGPUState.contextRegister[mmDB_HTILE_DATA_BASE];
uint32 regDepthSize = LatteGPUState.contextRegister[mmDB_DEPTH_SIZE];
uint32 regDepthBufferInfo = LatteGPUState.contextRegister[mmDB_DEPTH_INFO];
// get format and tileMode from info reg
uint32 depthBufferTileMode = (regDepthBufferInfo >> 15) & 0xF;
MPTR depthBufferPhysMem = regDepthBuffer << 8;
uint32 depthBufferPitch = (((regDepthSize >> 0) & 0x3FF) + 1);
uint32 depthBufferHeight = ((((regDepthSize >> 10) & 0xFFFFF) + 1) / depthBufferPitch);
depthBufferPitch <<= 3;
depthBufferHeight <<= 3;
uint32 depthBufferWidth = depthBufferPitch;
sint32 sliceIndex = 0; // todo
sint32 mipIndex = 0;
// clear all color buffers that match the format of the depth buffer
sint32 searchIndex = 0;
bool targetFound = false;
while (true)
{
LatteTextureView* view = LatteTC_LookupTextureByData(depthBufferPhysMem, depthBufferWidth, depthBufferHeight, depthBufferPitch, 0, 1, sliceIndex, 1, &searchIndex);
if (!view)
{
// should we clear in RAM instead?
break;
}
sint32 effectiveClearWidth = view->baseTexture->width;
sint32 effectiveClearHeight = view->baseTexture->height;
LatteTexture_scaleToEffectiveSize(view->baseTexture, &effectiveClearWidth, &effectiveClearHeight, 0);
// hacky way to get clear color
float* regClearColor = (float*)(LatteGPUState.contextRegister + 0xC000 + 0); // REG_BASE_ALU_CONST
uint8 clearColor[4] = { 0 };
clearColor[0] = (uint8)(regClearColor[0] * 255.0f);
clearColor[1] = (uint8)(regClearColor[1] * 255.0f);
clearColor[2] = (uint8)(regClearColor[2] * 255.0f);
clearColor[3] = (uint8)(regClearColor[3] * 255.0f);
// todo - use fragment shader software emulation (evoke for one pixel) to determine clear color
// todo - dont clear entire slice, use effectiveClearWidth, effectiveClearHeight
switch (g_renderer->GetType())
{
#ifdef ENABLE_OPENGL
case RendererAPI::OpenGL:
{
//cemu_assert_debug(false); // implement g_renderer->texture_clearColorSlice properly for OpenGL renderer
if (glClearTexSubImage)
glClearTexSubImage(((LatteTextureViewGL*)view)->glTexId, mipIndex, 0, 0, 0, effectiveClearWidth, effectiveClearHeight, 1, GL_RGBA, GL_UNSIGNED_BYTE, clearColor);
}
#endif
default:
{
if (view->baseTexture->isDepth)
g_renderer->texture_clearDepthSlice(view->baseTexture, sliceIndex + view->firstSlice, mipIndex + view->firstMip, true, view->baseTexture->hasStencil, 0.0f, 0);
else
g_renderer->texture_clearColorSlice(view->baseTexture, sliceIndex + view->firstSlice, mipIndex + view->firstMip, clearColor[0], clearColor[1], clearColor[2], clearColor[3]);
}
}
}
}

16
src/Cafe/HW/Latte/Renderer/Common/CommonRendererCore.h
View File

@ -77,16 +77,9 @@ typedef struct _indexDataCacheEntry_t
_indexDataCacheEntry_t* prevInMostRecentUsage; // points to element which was used less recently
}indexDataCacheEntry2_t;
uint8* CommonRenderer_getAttributePointerCacheVboOutput(uint32 attributeShaderLoc);
void CommonRenderer_setAttributePointerCacheVboOutput(uint32 attributeShaderLoc, uint8* vboOutput);
uint32 CommonRenderer_getAttributePointerCacheVboStride(uint32 attributeShaderLoc);
void CommonRenderer_setAttributePointerCacheVboStride(uint32 attributeShaderLoc, uint32 vboStride);
uint8 CommonRenderer_getAttributePointerCacheDataFormat(uint32 attributeShaderLoc);
void CommonRenderer_setAttributePointerCacheDataFormat(uint32 attributeShaderLoc, uint8 dataFormat);
uint8 CommonRenderer_getAttributePointerCacheNfa(uint32 attributeShaderLoc);
void CommonRenderer_setAttributePointerCacheNfa(uint32 attributeShaderLoc, uint8 nfa);
bool CommonRenderer_getAttributePointerCacheIsSigned(uint32 attributeShaderLoc);
void CommonRenderer_setAttributePointerCacheIsSigned(uint32 attributeShaderLoc, bool isSigned);
void CommonRenderer_resetAttributePointerCache();
bool CommonRenderer_checkIfAttributePointerCacheChanged(uint32 attributeShaderLoc, uint8* vboOutput, uint32 vboStride, uint8 dataFormat, uint8 nfa, bool isSigned);
indexState_t* CommonRenderer_getIndexState();
indexDataCacheEntry2_t** CommonRenderer_getIndexDataCacheFirst();
indexDataCacheEntry2_t** CommonRenderer_getIndexDataCacheBucket(uint32 bucketIdx);
@ -95,4 +88,5 @@ sint32* CommonRenderer_getIndexDataCacheEntryCount();
void CommonRenderer_appendToUsageLinkedList(indexDataCacheEntry2_t* entry);
void CommonRenderer_removeFromUsageLinkedList(indexDataCacheEntry2_t* entry);
void CommonRenderer_removeFromBucket(indexDataCacheEntry2_t* entry);
void LatteDraw_cleanupAfterFrame();
void LatteDraw_cleanupAfterFrame();
void LatteDraw_handleSpecialState8_clearAsDepth();

2
src/Cafe/HW/Latte/Renderer/Metal/MetalRenderer.cpp
View File

@ -33,7 +33,7 @@ extern bool hasValidFramebufferAttached;
float supportBufferData[512 * 4];
// Defined in the OpenGL renderer
// Defined in the Common renderer
void LatteDraw_handleSpecialState8_clearAsDepth();
std::vector<MetalRenderer::DeviceInfo> MetalRenderer::GetDevices()

86
src/Cafe/HW/Latte/Renderer/OpenGL/OpenGLRendererCore.cpp
View File

@ -27,11 +27,7 @@ extern bool hasValidFramebufferAttached;
void LatteDraw_resetAttributePointerCache()
{
for (sint32 i = 0; i < LATTE_VS_ATTRIBUTE_LIMIT; i++)
{
CommonRenderer_setAttributePointerCacheVboOutput(i, (uint8*)-1);
CommonRenderer_setAttributePointerCacheVboStride(i, (uint32)-1);
}
CommonRenderer_resetAttributePointerCache();
}
void _setAttributeBufferPointerRaw(uint32 attributeShaderLoc, uint8* buffer, uint32 bufferSize, uint32 stride, LatteParsedFetchShaderAttribute_t* attrib, uint8* vboOutput, uint32 vboStride)
@ -40,21 +36,11 @@ void _setAttributeBufferPointerRaw(uint32 attributeShaderLoc, uint8* buffer, uin
bool isSigned = attrib->isSigned != 0;
uint8 nfa = attrib->nfa;
// don't call glVertexAttribIPointer if parameters have not changed
if (
CommonRenderer_getAttributePointerCacheVboOutput(attributeShaderLoc) == vboOutput &&
CommonRenderer_getAttributePointerCacheVboStride(attributeShaderLoc) == vboStride &&
CommonRenderer_getAttributePointerCacheDataFormat(attributeShaderLoc) == dataFormat &&
CommonRenderer_getAttributePointerCacheNfa(attributeShaderLoc) == nfa &&
CommonRenderer_getAttributePointerCacheIsSigned(attributeShaderLoc) == isSigned
)
{
bool attributePointerChanged = CommonRenderer_checkIfAttributePointerCacheChanged(attributeShaderLoc, vboOutput, vboStride, dataFormat, nfa, isSigned);
if (!attributePointerChanged) {
return;
}
CommonRenderer_setAttributePointerCacheVboOutput(attributeShaderLoc, vboOutput);
CommonRenderer_setAttributePointerCacheVboStride(attributeShaderLoc, vboStride);
CommonRenderer_setAttributePointerCacheDataFormat(attributeShaderLoc, dataFormat);
CommonRenderer_setAttributePointerCacheNfa(attributeShaderLoc, nfa);
CommonRenderer_setAttributePointerCacheIsSigned(attributeShaderLoc, isSigned);
// setup attribute pointer
if (dataFormat == FMT_32_32_32_32_FLOAT || dataFormat == FMT_32_32_32_32)
{
@ -385,70 +371,6 @@ void LatteDrawGL_prepareIndicesWithGPUCache(MPTR indexDataMPTR, _INDEX_TYPE inde
indexState->indexData = (uint8*)(size_t)cacheEntry->heapEntry->startOffset;
}
void LatteDraw_handleSpecialState8_clearAsDepth()
{
if (LatteGPUState.contextNew.GetSpecialStateValues()[0] == 0)
cemuLog_logDebug(LogType::Force, "Special state 8 requires special state 0 but it is not set?");
// get depth buffer information
uint32 regDepthBuffer = LatteGPUState.contextRegister[mmDB_HTILE_DATA_BASE];
uint32 regDepthSize = LatteGPUState.contextRegister[mmDB_DEPTH_SIZE];
uint32 regDepthBufferInfo = LatteGPUState.contextRegister[mmDB_DEPTH_INFO];
// get format and tileMode from info reg
uint32 depthBufferTileMode = (regDepthBufferInfo >> 15) & 0xF;
MPTR depthBufferPhysMem = regDepthBuffer << 8;
uint32 depthBufferPitch = (((regDepthSize >> 0) & 0x3FF) + 1);
uint32 depthBufferHeight = ((((regDepthSize >> 10) & 0xFFFFF) + 1) / depthBufferPitch);
depthBufferPitch <<= 3;
depthBufferHeight <<= 3;
uint32 depthBufferWidth = depthBufferPitch;
sint32 sliceIndex = 0; // todo
sint32 mipIndex = 0;
// clear all color buffers that match the format of the depth buffer
sint32 searchIndex = 0;
bool targetFound = false;
while (true)
{
LatteTextureView* view = LatteTC_LookupTextureByData(depthBufferPhysMem, depthBufferWidth, depthBufferHeight, depthBufferPitch, 0, 1, sliceIndex, 1, &searchIndex);
if (!view)
{
// should we clear in RAM instead?
break;
}
sint32 effectiveClearWidth = view->baseTexture->width;
sint32 effectiveClearHeight = view->baseTexture->height;
LatteTexture_scaleToEffectiveSize(view->baseTexture, &effectiveClearWidth, &effectiveClearHeight, 0);
// hacky way to get clear color
float* regClearColor = (float*)(LatteGPUState.contextRegister + 0xC000 + 0); // REG_BASE_ALU_CONST
uint8 clearColor[4] = { 0 };
clearColor[0] = (uint8)(regClearColor[0] * 255.0f);
clearColor[1] = (uint8)(regClearColor[1] * 255.0f);
clearColor[2] = (uint8)(regClearColor[2] * 255.0f);
clearColor[3] = (uint8)(regClearColor[3] * 255.0f);
// todo - use fragment shader software emulation (evoke for one pixel) to determine clear color
// todo - dont clear entire slice, use effectiveClearWidth, effectiveClearHeight
if (g_renderer->GetType() == RendererAPI::OpenGL)
{
//cemu_assert_debug(false); // implement g_renderer->texture_clearColorSlice properly for OpenGL renderer
if (glClearTexSubImage)
glClearTexSubImage(((LatteTextureViewGL*)view)->glTexId, mipIndex, 0, 0, 0, effectiveClearWidth, effectiveClearHeight, 1, GL_RGBA, GL_UNSIGNED_BYTE, clearColor);
}
else
{
if (view->baseTexture->isDepth)
g_renderer->texture_clearDepthSlice(view->baseTexture, sliceIndex + view->firstSlice, mipIndex + view->firstMip, true, view->baseTexture->hasStencil, 0.0f, 0);
else
g_renderer->texture_clearColorSlice(view->baseTexture, sliceIndex + view->firstSlice, mipIndex + view->firstMip, clearColor[0], clearColor[1], clearColor[2], clearColor[3]);
}
}
}
void LatteDrawGL_doDraw(_INDEX_TYPE indexType, uint32 baseVertex, uint32 baseInstance, uint32 instanceCount, uint32 count)
{
indexState_t* indexState = CommonRenderer_getIndexState();

1
src/Cafe/HW/Latte/Renderer/Vulkan/VulkanRendererCore.cpp
View File

@ -1221,6 +1221,7 @@ void VulkanRenderer::draw_endRenderPass()
m_state.activeRenderpassFBO = nullptr;
}
// Defined in the Common renderer
void LatteDraw_handleSpecialState8_clearAsDepth();
// transfer depth buffer data to color buffer