Emulator-Archive/ppsspp
1
0
Fork 0
mirror of https://github.com/hrydgard/ppsspp.git synced 2024-11-27 15:30:35 +00:00
Code Issues Actions 2 Packages Projects Releases Wiki Activity
14ead27f8d
ppsspp/GPU/GLES/TextureCache.cpp
Unknown W. Brackets e30ab95179 d3d9: Fix maxSeenV when not known. 
And just make the code reused between both.  Fixes #8478.
2016-01-22 18:48:54 -08:00

1949 lines
64 KiB
C++
Raw Blame History

// Copyright (c) 2012- PPSSPP Project.
// This program is free software: you can redistribute it and/or modify
// it under the terms of the GNU General Public License as published by
// the Free Software Foundation, version 2.0 or later versions.
// This program is distributed in the hope that it will be useful,
// but WITHOUT ANY WARRANTY; without even the implied warranty of
// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
// GNU General Public License 2.0 for more details.
// A copy of the GPL 2.0 should have been included with the program.
// If not, see http://www.gnu.org/licenses/
// Official git repository and contact information can be found at
// https://github.com/hrydgard/ppsspp and http://www.ppsspp.org/.
#include <algorithm>
#include <cstring>
#include "ext/xxhash.h"
#include "i18n/i18n.h"
#include "math/math_util.h"
#include "profiler/profiler.h"
#include "Common/ColorConv.h"
#include "Core/Config.h"
#include "Core/Host.h"
#include "Core/MemMap.h"
#include "Core/Reporting.h"
#include "GPU/ge_constants.h"
#include "GPU/GPUState.h"
#include "GPU/GLES/GLStateCache.h"
#include "GPU/GLES/TextureCache.h"
#include "GPU/GLES/Framebuffer.h"
#include "GPU/GLES/FragmentShaderGenerator.h"
#include "GPU/GLES/DepalettizeShader.h"
#include "GPU/GLES/ShaderManager.h"
#include "GPU/GLES/TransformPipeline.h"
#include "GPU/Common/TextureDecoder.h"
#include "UI/OnScreenDisplay.h"
#ifdef _M_SSE
#include <xmmintrin.h>
#endif
// If a texture hasn't been seen for this many frames, get rid of it.
#define TEXTURE_KILL_AGE 200
#define TEXTURE_KILL_AGE_LOWMEM 60
// Not used in lowmem mode.
#define TEXTURE_SECOND_KILL_AGE 100
// Try to be prime to other decimation intervals.
#define TEXCACHE_DECIMATION_INTERVAL 13
// Changes more frequent than this will be considered "frequent" and prevent texture scaling.
#define TEXCACHE_FRAME_CHANGE_FREQUENT 6
// Note: only used when hash backoff is disabled.
#define TEXCACHE_FRAME_CHANGE_FREQUENT_REGAIN_TRUST 33
#define TEXCACHE_NAME_CACHE_SIZE 16
#define TEXCACHE_MAX_TEXELS_SCALED (256*256) // Per frame
#define TEXCACHE_MIN_PRESSURE 16 * 1024 * 1024 // Total in GL
#define TEXCACHE_SECOND_MIN_PRESSURE 4 * 1024 * 1024
#ifndef GL_UNPACK_ROW_LENGTH
#define GL_UNPACK_ROW_LENGTH 0x0CF2
#endif
#define INVALID_TEX -1
// Hack!
extern int g_iNumVideos;
TextureCache::TextureCache() : cacheSizeEstimate_(0), secondCacheSizeEstimate_(0), clearCacheNextFrame_(false), lowMemoryMode_(false), clutBuf_(NULL), texelsScaledThisFrame_(0) {
timesInvalidatedAllThisFrame_ = 0;
lastBoundTexture = INVALID_TEX;
decimationCounter_ = TEXCACHE_DECIMATION_INTERVAL;
glGetFloatv(GL_MAX_TEXTURE_MAX_ANISOTROPY_EXT, &maxAnisotropyLevel);
SetupTextureDecoder();
nextTexture_ = nullptr;
}
TextureCache::~TextureCache() {
Clear(true);
}
static u32 EstimateTexMemoryUsage(const TextureCache::TexCacheEntry *entry) {
const u16 dim = entry->dim;
const u8 dimW = ((dim >> 0) & 0xf);
const u8 dimH = ((dim >> 8) & 0xf);
u32 pixelSize = 2;
switch (entry->format) {
case GE_TFMT_CLUT4:
case GE_TFMT_CLUT8:
case GE_TFMT_CLUT16:
case GE_TFMT_CLUT32:
// We assume cluts always point to 8888 for simplicity.
pixelSize = 4;
break;
case GE_TFMT_4444:
case GE_TFMT_5551:
case GE_TFMT_5650:
break;
case GE_TFMT_8888:
case GE_TFMT_DXT1:
case GE_TFMT_DXT3:
case GE_TFMT_DXT5:
default:
pixelSize = 4;
break;
}
// This in other words multiplies by w and h.
return pixelSize << (dimW + dimH);
}
void TextureCache::Clear(bool delete_them) {
glBindTexture(GL_TEXTURE_2D, 0);
lastBoundTexture = INVALID_TEX;
if (delete_them) {
for (TexCache::iterator iter = cache.begin(); iter != cache.end(); ++iter) {
DEBUG_LOG(G3D, "Deleting texture %i", iter->second.textureName);
glDeleteTextures(1, &iter->second.textureName);
}
for (TexCache::iterator iter = secondCache.begin(); iter != secondCache.end(); ++iter) {
DEBUG_LOG(G3D, "Deleting texture %i", iter->second.textureName);
glDeleteTextures(1, &iter->second.textureName);
}
if (!nameCache_.empty()) {
glDeleteTextures((GLsizei)nameCache_.size(), &nameCache_[0]);
nameCache_.clear();
}
}
if (cache.size() + secondCache.size()) {
INFO_LOG(G3D, "Texture cached cleared from %i textures", (int)(cache.size() + secondCache.size()));
cache.clear();
secondCache.clear();
cacheSizeEstimate_ = 0;
secondCacheSizeEstimate_ = 0;
}
fbTexInfo_.clear();
}
void TextureCache::DeleteTexture(TexCache::iterator it) {
glDeleteTextures(1, &it->second.textureName);
auto fbInfo = fbTexInfo_.find(it->second.addr);
if (fbInfo != fbTexInfo_.end()) {
fbTexInfo_.erase(fbInfo);
}
cacheSizeEstimate_ -= EstimateTexMemoryUsage(&it->second);
cache.erase(it);
}
// Removes old textures.
void TextureCache::Decimate() {
if (--decimationCounter_ <= 0) {
decimationCounter_ = TEXCACHE_DECIMATION_INTERVAL;
} else {
return;
}
if (cacheSizeEstimate_ >= TEXCACHE_MIN_PRESSURE) {
const u32 had = cacheSizeEstimate_;
glBindTexture(GL_TEXTURE_2D, 0);
lastBoundTexture = INVALID_TEX;
int killAge = lowMemoryMode_ ? TEXTURE_KILL_AGE_LOWMEM : TEXTURE_KILL_AGE;
for (TexCache::iterator iter = cache.begin(); iter != cache.end(); ) {
if (iter->second.lastFrame + killAge < gpuStats.numFlips) {
DeleteTexture(iter++);
} else {
++iter;
}
}
VERBOSE_LOG(G3D, "Decimated texture cache, saved %d estimated bytes - now %d bytes", had - cacheSizeEstimate_, cacheSizeEstimate_);
}
if (g_Config.bTextureSecondaryCache && secondCacheSizeEstimate_ >= TEXCACHE_SECOND_MIN_PRESSURE) {
const u32 had = secondCacheSizeEstimate_;
for (TexCache::iterator iter = secondCache.begin(); iter != secondCache.end(); ) {
// In low memory mode, we kill them all.
if (lowMemoryMode_ || iter->second.lastFrame + TEXTURE_SECOND_KILL_AGE < gpuStats.numFlips) {
glDeleteTextures(1, &iter->second.textureName);
secondCacheSizeEstimate_ -= EstimateTexMemoryUsage(&iter->second);
secondCache.erase(iter++);
} else {
++iter;
}
}
VERBOSE_LOG(G3D, "Decimated second texture cache, saved %d estimated bytes - now %d bytes", had - secondCacheSizeEstimate_, secondCacheSizeEstimate_);
}
}
void TextureCache::Invalidate(u32 addr, int size, GPUInvalidationType type) {
// If we're hashing every use, without backoff, then this isn't needed.
if (!g_Config.bTextureBackoffCache) {
return;
}
addr &= 0x3FFFFFFF;
const u32 addr_end = addr + size;
// They could invalidate inside the texture, let's just give a bit of leeway.
const int LARGEST_TEXTURE_SIZE = 512 * 512 * 4;
const u64 startKey = (u64)(addr - LARGEST_TEXTURE_SIZE) << 32;
u64 endKey = (u64)(addr + size + LARGEST_TEXTURE_SIZE) << 32;
if (endKey < startKey) {
endKey = (u64)-1;
}
for (TexCache::iterator iter = cache.lower_bound(startKey), end = cache.upper_bound(endKey); iter != end; ++iter) {
u32 texAddr = iter->second.addr;
u32 texEnd = iter->second.addr + iter->second.sizeInRAM;
if (texAddr < addr_end && addr < texEnd) {
if (iter->second.GetHashStatus() == TexCacheEntry::STATUS_RELIABLE) {
iter->second.SetHashStatus(TexCacheEntry::STATUS_HASHING);
}
if (type != GPU_INVALIDATE_ALL) {
gpuStats.numTextureInvalidations++;
// Start it over from 0 (unless it's safe.)
iter->second.numFrames = type == GPU_INVALIDATE_SAFE ? 256 : 0;
if (type == GPU_INVALIDATE_SAFE) {
u32 diff = gpuStats.numFlips - iter->second.lastFrame;
// We still need to mark if the texture is frequently changing, even if it's safely changing.
if (diff < TEXCACHE_FRAME_CHANGE_FREQUENT) {
iter->second.status |= TexCacheEntry::STATUS_CHANGE_FREQUENT;
}
}
iter->second.framesUntilNextFullHash = 0;
} else if (!iter->second.framebuffer) {
iter->second.invalidHint++;
}
}
}
}
void TextureCache::InvalidateAll(GPUInvalidationType /*unused*/) {
// If we're hashing every use, without backoff, then this isn't needed.
if (!g_Config.bTextureBackoffCache) {
return;
}
if (timesInvalidatedAllThisFrame_ > 5) {
return;
}
timesInvalidatedAllThisFrame_++;
for (TexCache::iterator iter = cache.begin(), end = cache.end(); iter != end; ++iter) {
if (iter->second.GetHashStatus() == TexCacheEntry::STATUS_RELIABLE) {
iter->second.SetHashStatus(TexCacheEntry::STATUS_HASHING);
}
if (!iter->second.framebuffer) {
iter->second.invalidHint++;
}
}
}
void TextureCache::ClearNextFrame() {
clearCacheNextFrame_ = true;
}
void TextureCache::AttachFramebufferValid(TexCacheEntry *entry, VirtualFramebuffer *framebuffer, const AttachedFramebufferInfo &fbInfo) {
const bool hasInvalidFramebuffer = entry->framebuffer == nullptr || entry->invalidHint == -1;
const bool hasOlderFramebuffer = entry->framebuffer != nullptr && entry->framebuffer->last_frame_render < framebuffer->last_frame_render;
bool hasFartherFramebuffer = false;
if (!hasInvalidFramebuffer && !hasOlderFramebuffer) {
// If it's valid, but the offset is greater, then we still win.
if (fbTexInfo_[entry->addr].yOffset == fbInfo.yOffset)
hasFartherFramebuffer = fbTexInfo_[entry->addr].xOffset > fbInfo.xOffset;
else
hasFartherFramebuffer = fbTexInfo_[entry->addr].yOffset > fbInfo.yOffset;
}
if (hasInvalidFramebuffer || hasOlderFramebuffer || hasFartherFramebuffer) {
if (entry->framebuffer == nullptr) {
cacheSizeEstimate_ -= EstimateTexMemoryUsage(entry);
}
entry->framebuffer = framebuffer;
entry->invalidHint = 0;
entry->status &= ~TextureCache::TexCacheEntry::STATUS_DEPALETTIZE;
entry->maxLevel = 0;
fbTexInfo_[entry->addr] = fbInfo;
framebuffer->last_frame_attached = gpuStats.numFlips;
host->GPUNotifyTextureAttachment(entry->addr);
} else if (entry->framebuffer == framebuffer) {
framebuffer->last_frame_attached = gpuStats.numFlips;
}
}
void TextureCache::AttachFramebufferInvalid(TexCacheEntry *entry, VirtualFramebuffer *framebuffer, const AttachedFramebufferInfo &fbInfo) {
if (entry->framebuffer == nullptr || entry->framebuffer == framebuffer) {
if (entry->framebuffer == nullptr) {
cacheSizeEstimate_ -= EstimateTexMemoryUsage(entry);
}
entry->framebuffer = framebuffer;
entry->invalidHint = -1;
entry->status &= ~TextureCache::TexCacheEntry::STATUS_DEPALETTIZE;
entry->maxLevel = 0;
fbTexInfo_[entry->addr] = fbInfo;
host->GPUNotifyTextureAttachment(entry->addr);
}
}
bool TextureCache::AttachFramebuffer(TexCacheEntry *entry, u32 address, VirtualFramebuffer *framebuffer, u32 texaddrOffset) {
static const u32 MAX_SUBAREA_Y_OFFSET_SAFE = 32;
AttachedFramebufferInfo fbInfo = {0};
const u64 mirrorMask = 0x00600000;
// Must be in VRAM so | 0x04000000 it is. Also, ignore memory mirrors.
const u32 addr = (address | 0x04000000) & 0x3FFFFFFF & ~mirrorMask;
const u32 texaddr = ((entry->addr + texaddrOffset) & ~mirrorMask);
const bool noOffset = texaddr == addr;
const bool exactMatch = noOffset && entry->format < 4;
const u32 h = 1 << ((entry->dim >> 8) & 0xf);
// 512 on a 272 framebuffer is sane, so let's be lenient.
const u32 minSubareaHeight = h / 4;
// If they match exactly, it's non-CLUT and from the top left.
if (exactMatch) {
// Apply to non-buffered and buffered mode only.
if (!(g_Config.iRenderingMode == FB_NON_BUFFERED_MODE || g_Config.iRenderingMode == FB_BUFFERED_MODE))
return false;
DEBUG_LOG(G3D, "Render to texture detected at %08x!", address);
if (framebuffer->fb_stride != entry->bufw) {
WARN_LOG_REPORT_ONCE(diffStrides1, G3D, "Render to texture with different strides %d != %d", entry->bufw, framebuffer->fb_stride);
}
if (entry->format != (GETextureFormat)framebuffer->format) {
WARN_LOG_REPORT_ONCE(diffFormat1, G3D, "Render to texture with different formats %d != %d", entry->format, framebuffer->format);
// Let's avoid using it when we know the format is wrong. May be a video/etc. updating memory.
// However, some games use a different format to clear the buffer.
if (framebuffer->last_frame_attached + 1 < gpuStats.numFlips) {
DetachFramebuffer(entry, address, framebuffer);
}
} else {
AttachFramebufferValid(entry, framebuffer, fbInfo);
return true;
}
} else {
// Apply to buffered mode only.
if (!(g_Config.iRenderingMode == FB_BUFFERED_MODE))
return false;
const bool clutFormat =
(framebuffer->format == GE_FORMAT_8888 && entry->format == GE_TFMT_CLUT32) ||
(framebuffer->format != GE_FORMAT_8888 && entry->format == GE_TFMT_CLUT16);
const u32 bitOffset = (texaddr - addr) * 8;
const u32 pixelOffset = bitOffset / std::max(1U, (u32)textureBitsPerPixel[entry->format]);
fbInfo.yOffset = entry->bufw == 0 ? 0 : pixelOffset / entry->bufw;
fbInfo.xOffset = entry->bufw == 0 ? 0 : pixelOffset % entry->bufw;
if (framebuffer->fb_stride != entry->bufw) {
if (noOffset) {
WARN_LOG_REPORT_ONCE(diffStrides2, G3D, "Render to texture using CLUT with different strides %d != %d", entry->bufw, framebuffer->fb_stride);
} else {
// Assume any render-to-tex with different bufw + offset is a render from ram.
DetachFramebuffer(entry, address, framebuffer);
return false;
}
}
if (fbInfo.yOffset + minSubareaHeight >= framebuffer->height) {
// Can't be inside the framebuffer then, ram. Detach to be safe.
DetachFramebuffer(entry, address, framebuffer);
return false;
}
// Trying to play it safe. Below 0x04110000 is almost always framebuffers.
// TODO: Maybe we can reduce this check and find a better way above 0x04110000?
if (fbInfo.yOffset > MAX_SUBAREA_Y_OFFSET_SAFE && addr > 0x04110000) {
WARN_LOG_REPORT_ONCE(subareaIgnored, G3D, "Ignoring possible render to texture at %08x +%dx%d / %dx%d", address, fbInfo.xOffset, fbInfo.yOffset, framebuffer->width, framebuffer->height);
DetachFramebuffer(entry, address, framebuffer);
return false;
}
// Check for CLUT. The framebuffer is always RGB, but it can be interpreted as a CLUT texture.
// 3rd Birthday (and a bunch of other games) render to a 16 bit clut texture.
if (clutFormat) {
if (!noOffset) {
WARN_LOG_REPORT_ONCE(subareaClut, G3D, "Render to texture using CLUT with offset at %08x +%dx%d", address, fbInfo.xOffset, fbInfo.yOffset);
}
AttachFramebufferValid(entry, framebuffer, fbInfo);
entry->status |= TexCacheEntry::STATUS_DEPALETTIZE;
// We'll validate it compiles later.
return true;
} else if (entry->format == GE_TFMT_CLUT8 || entry->format == GE_TFMT_CLUT4) {
ERROR_LOG_REPORT_ONCE(fourEightBit, G3D, "4 and 8-bit CLUT format not supported for framebuffers");
}
// This is either normal or we failed to generate a shader to depalettize
if (framebuffer->format == entry->format || clutFormat) {
if (framebuffer->format != entry->format) {
WARN_LOG_REPORT_ONCE(diffFormat2, G3D, "Render to texture with different formats %d != %d at %08x", entry->format, framebuffer->format, address);
AttachFramebufferValid(entry, framebuffer, fbInfo);
return true;
} else {
WARN_LOG_REPORT_ONCE(subarea, G3D, "Render to area containing texture at %08x +%dx%d", address, fbInfo.xOffset, fbInfo.yOffset);
// If "AttachFramebufferValid" , God of War Ghost of Sparta/Chains of Olympus will be missing special effect.
AttachFramebufferInvalid(entry, framebuffer, fbInfo);
return true;
}
} else {
WARN_LOG_REPORT_ONCE(diffFormat2, G3D, "Render to texture with incompatible formats %d != %d at %08x", entry->format, framebuffer->format, address);
}
}
return false;
}
inline void TextureCache::DetachFramebuffer(TexCacheEntry *entry, u32 address, VirtualFramebuffer *framebuffer) {
if (entry->framebuffer == framebuffer) {
cacheSizeEstimate_ += EstimateTexMemoryUsage(entry);
entry->framebuffer = 0;
host->GPUNotifyTextureAttachment(entry->addr);
}
}
void *TextureCache::ReadIndexedTex(int level, const u8 *texptr, int bytesPerIndex, GLuint dstFmt, int bufw) {
int w = gstate.getTextureWidth(level);
int h = gstate.getTextureHeight(level);
int length = bufw * h;
void *buf = NULL;
switch (gstate.getClutPaletteFormat()) {
case GE_CMODE_16BIT_BGR5650:
case GE_CMODE_16BIT_ABGR5551:
case GE_CMODE_16BIT_ABGR4444:
{
tmpTexBuf16.resize(std::max(bufw, w) * h);
tmpTexBufRearrange.resize(std::max(bufw, w) * h);
const u16 *clut = GetCurrentClut<u16>();
if (!gstate.isTextureSwizzled()) {
switch (bytesPerIndex) {
case 1:
DeIndexTexture(tmpTexBuf16.data(), (const u8 *)texptr, length, clut);
break;
case 2:
DeIndexTexture(tmpTexBuf16.data(), (const u16_le *)texptr, length, clut);
break;
case 4:
DeIndexTexture(tmpTexBuf16.data(), (const u32_le *)texptr, length, clut);
break;
}
} else {
tmpTexBuf32.resize(std::max(bufw, w) * h);
UnswizzleFromMem(texptr, bufw, h, bytesPerIndex);
switch (bytesPerIndex) {
case 1:
DeIndexTexture(tmpTexBuf16.data(), (u8 *) tmpTexBuf32.data(), length, clut);
break;
case 2:
DeIndexTexture(tmpTexBuf16.data(), (u16 *) tmpTexBuf32.data(), length, clut);
break;
case 4:
DeIndexTexture(tmpTexBuf16.data(), (u32 *) tmpTexBuf32.data(), length, clut);
break;
}
}
buf = tmpTexBuf16.data();
}
break;
case GE_CMODE_32BIT_ABGR8888:
{
tmpTexBuf32.resize(std::max(bufw, w) * h);
tmpTexBufRearrange.resize(std::max(bufw, w) * h);
const u32 *clut = GetCurrentClut<u32>();
if (!gstate.isTextureSwizzled()) {
switch (bytesPerIndex) {
case 1:
DeIndexTexture(tmpTexBuf32.data(), (const u8 *)texptr, length, clut);
break;
case 2:
DeIndexTexture(tmpTexBuf32.data(), (const u16_le *)texptr, length, clut);
break;
case 4:
DeIndexTexture(tmpTexBuf32.data(), (const u32_le *)texptr, length, clut);
break;
}
buf = tmpTexBuf32.data();
} else {
UnswizzleFromMem(texptr, bufw, h, bytesPerIndex);
// Since we had to unswizzle to tmpTexBuf32, let's output to tmpTexBuf16.
tmpTexBuf16.resize(std::max(bufw, w) * h * 2);
u32 *dest32 = (u32 *) tmpTexBuf16.data();
switch (bytesPerIndex) {
case 1:
DeIndexTexture(dest32, (u8 *) tmpTexBuf32.data(), length, clut);
buf = dest32;
break;
case 2:
DeIndexTexture(dest32, (u16 *) tmpTexBuf32.data(), length, clut);
buf = dest32;
break;
case 4:
// TODO: If a game actually uses this mode, check if using dest32 or tmpTexBuf32 is faster.
DeIndexTexture(tmpTexBuf32.data(), tmpTexBuf32.data(), length, clut);
buf = tmpTexBuf32.data();
break;
}
}
}
break;
default:
ERROR_LOG_REPORT(G3D, "Unhandled clut texture mode %d!!!", (gstate.clutformat & 3));
break;
}
return buf;
}
GLenum getClutDestFormat(GEPaletteFormat format) {
switch (format) {
case GE_CMODE_16BIT_ABGR4444:
return GL_UNSIGNED_SHORT_4_4_4_4;
case GE_CMODE_16BIT_ABGR5551:
return GL_UNSIGNED_SHORT_5_5_5_1;
case GE_CMODE_16BIT_BGR5650:
return GL_UNSIGNED_SHORT_5_6_5;
case GE_CMODE_32BIT_ABGR8888:
return GL_UNSIGNED_BYTE;
}
return 0;
}
static const u8 texByteAlignMap[] = {2, 2, 2, 4};
static const GLuint MinFiltGL[8] = {
GL_NEAREST,
GL_LINEAR,
GL_NEAREST,
GL_LINEAR,
GL_NEAREST_MIPMAP_NEAREST,
GL_LINEAR_MIPMAP_NEAREST,
GL_NEAREST_MIPMAP_LINEAR,
GL_LINEAR_MIPMAP_LINEAR,
};
static const GLuint MagFiltGL[2] = {
GL_NEAREST,
GL_LINEAR
};
// This should not have to be done per texture! OpenGL is silly yo
void TextureCache::UpdateSamplingParams(TexCacheEntry &entry, bool force) {
int minFilt;
int magFilt;
bool sClamp;
bool tClamp;
float lodBias;
GetSamplingParams(minFilt, magFilt, sClamp, tClamp, lodBias, entry.maxLevel);
if (entry.maxLevel != 0) {
if (force || entry.lodBias != lodBias) {
if (gstate_c.Supports(GPU_SUPPORTS_TEXTURE_LOD_CONTROL)) {
GETexLevelMode mode = gstate.getTexLevelMode();
switch (mode) {
case GE_TEXLEVEL_MODE_AUTO:
// TODO
break;
case GE_TEXLEVEL_MODE_CONST:
// Sigh, LOD_BIAS is not even in ES 3.0..
#ifndef USING_GLES2
glTexParameterf(GL_TEXTURE_2D, GL_TEXTURE_LOD_BIAS, lodBias);
#endif
break;
case GE_TEXLEVEL_MODE_SLOPE:
// TODO
break;
}
}
entry.lodBias = lodBias;
}
}
if (force || entry.minFilt != minFilt) {
glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, MinFiltGL[minFilt]);
entry.minFilt = minFilt;
}
if (force || entry.magFilt != magFilt) {
glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, MagFiltGL[magFilt]);
entry.magFilt = magFilt;
}
if (entry.framebuffer) {
WARN_LOG_REPORT_ONCE(wrongFramebufAttach, G3D, "Framebuffer still attached in UpdateSamplingParams()?");
}
if (force || entry.sClamp != sClamp) {
glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, sClamp ? GL_CLAMP_TO_EDGE : GL_REPEAT);
entry.sClamp = sClamp;
}
if (force || entry.tClamp != tClamp) {
glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, tClamp ? GL_CLAMP_TO_EDGE : GL_REPEAT);
entry.tClamp = tClamp;
}
}
void TextureCache::SetFramebufferSamplingParams(u16 bufferWidth, u16 bufferHeight) {
int minFilt;
int magFilt;
bool sClamp;
bool tClamp;
float lodBias;
GetSamplingParams(minFilt, magFilt, sClamp, tClamp, lodBias, 0);
minFilt &= 1; // framebuffers can't mipmap.
glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, MinFiltGL[minFilt]);
glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, MagFiltGL[magFilt]);
// Often the framebuffer will not match the texture size. We'll wrap/clamp in the shader in that case.
// This happens whether we have OES_texture_npot or not.
int w = gstate.getTextureWidth(0);
int h = gstate.getTextureHeight(0);
if (w != bufferWidth || h != bufferHeight) {
return;
}
glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, sClamp ? GL_CLAMP_TO_EDGE : GL_REPEAT);
glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, tClamp ? GL_CLAMP_TO_EDGE : GL_REPEAT);
}
static void ConvertColors(void *dstBuf, const void *srcBuf, GLuint dstFmt, int numPixels) {
const u32 *src = (const u32 *)srcBuf;
u32 *dst = (u32 *)dstBuf;
switch (dstFmt) {
case GL_UNSIGNED_SHORT_4_4_4_4:
ConvertRGBA4444ToABGR4444((u16 *)dst, (const u16 *)src, numPixels);
break;
// Final Fantasy 2 uses this heavily in animated textures.
case GL_UNSIGNED_SHORT_5_5_5_1:
ConvertRGBA5551ToABGR1555((u16 *)dst, (const u16 *)src, numPixels);
break;
case GL_UNSIGNED_SHORT_5_6_5:
ConvertRGB565ToBGR565((u16 *)dst, (const u16 *)src, numPixels);
break;
default:
if (UseBGRA8888()) {
ConvertRGBA8888ToBGRA8888(dst, src, numPixels);
} else {
// No need to convert RGBA8888, right order already
if (dst != src)
memcpy(dst, src, numPixels * sizeof(u32));
}
break;
}
}
void TextureCache::StartFrame() {
lastBoundTexture = INVALID_TEX;
timesInvalidatedAllThisFrame_ = 0;
if (texelsScaledThisFrame_) {
// INFO_LOG(G3D, "Scaled %i texels", texelsScaledThisFrame_);
}
texelsScaledThisFrame_ = 0;
if (clearCacheNextFrame_) {
Clear(true);
clearCacheNextFrame_ = false;
} else {
Decimate();
}
}
static inline u32 MiniHash(const u32 *ptr) {
return ptr[0];
}
static inline u32 QuickTexHash(u32 addr, int bufw, int w, int h, GETextureFormat format, TextureCache::TexCacheEntry *entry) {
if (h == 512 && entry->maxSeenV < 512 && entry->maxSeenV != 0) {
h = (int)entry->maxSeenV;
}
const u32 sizeInRAM = (textureBitsPerPixel[format] * bufw * h) / 8;
const u32 *checkp = (const u32 *) Memory::GetPointer(addr);
return DoQuickTexHash(checkp, sizeInRAM);
}
void TextureCache::UpdateCurrentClut(GEPaletteFormat clutFormat, u32 clutBase, bool clutIndexIsSimple) {
const u32 clutBaseBytes = clutFormat == GE_CMODE_32BIT_ABGR8888 ? (clutBase * sizeof(u32)) : (clutBase * sizeof(u16));
// Technically, these extra bytes weren't loaded, but hopefully it was loaded earlier.
// If not, we're going to hash random data, which hopefully doesn't cause a performance issue.
//
// TODO: Actually, this seems like a hack. The game can upload part of a CLUT and reference other data.
// clutTotalBytes_ is the last amount uploaded. We should hash clutMaxBytes_, but this will often hash
// unrelated old entries for small palettes.
// Adding clutBaseBytes may just be mitigating this for some usage patterns.
const u32 clutExtendedBytes = std::min(clutTotalBytes_ + clutBaseBytes, clutMaxBytes_);
clutHash_ = DoReliableHash32((const char *)clutBufRaw_, clutExtendedBytes, 0xC0108888);
// Avoid a copy when we don't need to convert colors.
if (UseBGRA8888() || clutFormat != GE_CMODE_32BIT_ABGR8888) {
const int numColors = clutFormat == GE_CMODE_32BIT_ABGR8888 ? (clutMaxBytes_ / sizeof(u32)) : (clutMaxBytes_ / sizeof(u16));
ConvertColors(clutBufConverted_, clutBufRaw_, getClutDestFormat(clutFormat), numColors);
clutBuf_ = clutBufConverted_;
} else {
clutBuf_ = clutBufRaw_;
}
// Special optimization: fonts typically draw clut4 with just alpha values in a single color.
clutAlphaLinear_ = false;
clutAlphaLinearColor_ = 0;
if (clutFormat == GE_CMODE_16BIT_ABGR4444 && clutIndexIsSimple) {
const u16_le *clut = GetCurrentClut<u16_le>();
clutAlphaLinear_ = true;
clutAlphaLinearColor_ = clut[15] & 0xFFF0;
for (int i = 0; i < 16; ++i) {
u16 step = clutAlphaLinearColor_ | i;
if (clut[i] != step) {
clutAlphaLinear_ = false;
break;
}
}
}
clutLastFormat_ = gstate.clutformat;
}
template <typename T>
inline const T *TextureCache::GetCurrentClut() {
return (const T *)clutBuf_;
}
inline u32 TextureCache::GetCurrentClutHash() {
return clutHash_;
}
// #define DEBUG_TEXTURES
#ifdef DEBUG_TEXTURES
bool SetDebugTexture() {
static const int highlightFrames = 30;
static int numTextures = 0;
static int lastFrames = 0;
static int mostTextures = 1;
if (lastFrames != gpuStats.numFlips) {
mostTextures = std::max(mostTextures, numTextures);
numTextures = 0;
lastFrames = gpuStats.numFlips;
}
static GLuint solidTexture = 0;
bool changed = false;
if (((gpuStats.numFlips / highlightFrames) % mostTextures) == numTextures) {
if (gpuStats.numFlips % highlightFrames == 0) {
NOTICE_LOG(G3D, "Highlighting texture # %d / %d", numTextures, mostTextures);
}
static const u32 solidTextureData[] = {0x99AA99FF};
if (solidTexture == 0) {
glGenTextures(1, &solidTexture);
glBindTexture(GL_TEXTURE_2D, solidTexture);
glPixelStorei(GL_UNPACK_ALIGNMENT, 1);
glPixelStorei(GL_PACK_ALIGNMENT, 1);
glTexImage2D(GL_TEXTURE_2D, 0, GL_RGBA, 1, 1, 0, GL_RGBA, GL_UNSIGNED_BYTE, solidTextureData);
} else {
glBindTexture(GL_TEXTURE_2D, solidTexture);
}
changed = true;
}
++numTextures;
return changed;
}
#endif
void TextureCache::SetTextureFramebuffer(TexCacheEntry *entry, VirtualFramebuffer *framebuffer) {
_dbg_assert_msg_(G3D, framebuffer != nullptr, "Framebuffer must not be null.");
framebuffer->usageFlags |= FB_USAGE_TEXTURE;
bool useBufferedRendering = g_Config.iRenderingMode != FB_NON_BUFFERED_MODE;
if (useBufferedRendering) {
const GEPaletteFormat clutFormat = gstate.getClutPaletteFormat();
DepalShader *depal = nullptr;
if ((entry->status & TexCacheEntry::STATUS_DEPALETTIZE) && !g_Config.bDisableSlowFramebufEffects) {
depal = depalShaderCache_->GetDepalettizeShader(clutFormat, framebuffer->drawnFormat);
}
if (depal) {
const u32 bytesPerColor = clutFormat == GE_CMODE_32BIT_ABGR8888 ? sizeof(u32) : sizeof(u16);
const u32 clutTotalColors = clutMaxBytes_ / bytesPerColor;
TexCacheEntry::Status alphaStatus = CheckAlpha(clutBuf_, getClutDestFormat(clutFormat), clutTotalColors, clutTotalColors, 1);
gstate_c.textureFullAlpha = alphaStatus == TexCacheEntry::STATUS_ALPHA_FULL;
gstate_c.textureSimpleAlpha = alphaStatus == TexCacheEntry::STATUS_ALPHA_SIMPLE;
} else {
entry->status &= ~TexCacheEntry::STATUS_DEPALETTIZE;
gstate_c.textureFullAlpha = gstate.getTextureFormat() == GE_TFMT_5650;
gstate_c.textureSimpleAlpha = gstate_c.textureFullAlpha;
}
// Keep the framebuffer alive.
framebuffer->last_frame_used = gpuStats.numFlips;
// We need to force it, since we may have set it on a texture before attaching.
gstate_c.curTextureWidth = framebuffer->bufferWidth;
gstate_c.curTextureHeight = framebuffer->bufferHeight;
gstate_c.curTextureXOffset = fbTexInfo_[entry->addr].xOffset;
gstate_c.curTextureYOffset = fbTexInfo_[entry->addr].yOffset;
gstate_c.needShaderTexClamp = gstate_c.curTextureWidth != (u32)gstate.getTextureWidth(0) || gstate_c.curTextureHeight != (u32)gstate.getTextureHeight(0);
if (gstate_c.curTextureXOffset != 0 || gstate_c.curTextureYOffset != 0) {
gstate_c.needShaderTexClamp = true;
}
nextTexture_ = entry;
} else {
if (framebuffer->fbo) {
fbo_destroy(framebuffer->fbo);
framebuffer->fbo = 0;
}
glBindTexture(GL_TEXTURE_2D, 0);
gstate_c.needShaderTexClamp = false;
}
}
void TextureCache::ApplyTexture() {
if (nextTexture_ == nullptr) {
return;
}
if (nextTexture_->framebuffer) {
ApplyTextureFramebuffer(nextTexture_, nextTexture_->framebuffer);
} else {
UpdateMaxSeenV(gstate.isModeThrough());
if (nextTexture_->textureName != lastBoundTexture) {
glBindTexture(GL_TEXTURE_2D, nextTexture_->textureName);
lastBoundTexture = nextTexture_->textureName;
}
UpdateSamplingParams(*nextTexture_, false);
}
nextTexture_ = nullptr;
}
void TextureCache::DownloadFramebufferForClut(u32 clutAddr, u32 bytes) {
framebufferManager_->DownloadFramebufferForClut(clutAddr, bytes);
}
class TextureShaderApplier {
public:
struct Pos {
Pos(float x_, float y_, float z_) : x(x_), y(y_), z(z_) {
}
Pos() {
}
float x;
float y;
float z;
};
struct UV {
UV(float u_, float v_) : u(u_), v(v_) {
}
UV() {
}
float u;
float v;
};
TextureShaderApplier(DepalShader *shader, float bufferW, float bufferH, int renderW, int renderH)
: shader_(shader), bufferW_(bufferW), bufferH_(bufferH), renderW_(renderW), renderH_(renderH) {
static const Pos pos[4] = {
{-1, -1, -1},
{ 1, -1, -1},
{ 1, 1, -1},
{-1, 1, -1},
};
memcpy(pos_, pos, sizeof(pos_));
static const UV uv[4] = {
{0, 0},
{1, 0},
{1, 1},
{0, 1},
};
memcpy(uv_, uv, sizeof(uv_));
}
void ApplyBounds(const KnownVertexBounds &bounds, u32 uoff, u32 voff) {
// If min is not < max, then we don't have values (wasn't set during decode.)
if (bounds.minV < bounds.maxV) {
const float invWidth = 1.0f / bufferW_;
const float invHeight = 1.0f / bufferH_;
// Inverse of half = double.
const float invHalfWidth = invWidth * 2.0f;
const float invHalfHeight = invHeight * 2.0f;
const int u1 = bounds.minU + uoff;
const int v1 = bounds.minV + voff;
const int u2 = bounds.maxU + uoff;
const int v2 = bounds.maxV + voff;
const float left = u1 * invHalfWidth - 1.0f;
const float right = u2 * invHalfWidth - 1.0f;
const float top = v1 * invHalfHeight - 1.0f;
const float bottom = v2 * invHalfHeight - 1.0f;
// Points are: BL, BR, TR, TL.
pos_[0] = Pos(left, bottom, -1.0f);
pos_[1] = Pos(right, bottom, -1.0f);
pos_[2] = Pos(right, top, -1.0f);
pos_[3] = Pos(left, top, -1.0f);
// And also the UVs, same order.
const float uvleft = u1 * invWidth;
const float uvright = u2 * invWidth;
const float uvtop = v1 * invHeight;
const float uvbottom = v2 * invHeight;
uv_[0] = UV(uvleft, uvbottom);
uv_[1] = UV(uvright, uvbottom);
uv_[2] = UV(uvright, uvtop);
uv_[3] = UV(uvleft, uvtop);
}
}
void Use(TransformDrawEngine *transformDraw) {
glUseProgram(shader_->program);
// Restore will rebind all of the state below.
if (gstate_c.Supports(GPU_SUPPORTS_VAO)) {
static const GLubyte indices[4] = { 0, 1, 3, 2 };
transformDraw->BindBuffer(pos_, sizeof(pos_), uv_, sizeof(uv_));
transformDraw->BindElementBuffer(indices, sizeof(indices));
} else {
glBindBuffer(GL_ARRAY_BUFFER, 0);
glBindBuffer(GL_ELEMENT_ARRAY_BUFFER, 0);
}
glEnableVertexAttribArray(shader_->a_position);
glEnableVertexAttribArray(shader_->a_texcoord0);
}
void Shade() {
static const GLubyte indices[4] = { 0, 1, 3, 2 };
glstate.blend.force(false);
glstate.colorMask.force(true, true, true, true);
glstate.scissorTest.force(false);
glstate.cullFace.force(false);
glstate.depthTest.force(false);
glstate.stencilTest.force(false);
#if !defined(USING_GLES2)
glstate.colorLogicOp.force(false);
#endif
glViewport(0, 0, renderW_, renderH_);
if (gstate_c.Supports(GPU_SUPPORTS_VAO)) {
glVertexAttribPointer(shader_->a_position, 3, GL_FLOAT, GL_FALSE, 12, 0);
glVertexAttribPointer(shader_->a_texcoord0, 2, GL_FLOAT, GL_FALSE, 8, (void *)sizeof(pos_));
glDrawElements(GL_TRIANGLE_STRIP, 4, GL_UNSIGNED_BYTE, 0);
} else {
glVertexAttribPointer(shader_->a_position, 3, GL_FLOAT, GL_FALSE, 12, pos_);
glVertexAttribPointer(shader_->a_texcoord0, 2, GL_FLOAT, GL_FALSE, 8, uv_);
glDrawElements(GL_TRIANGLE_STRIP, 4, GL_UNSIGNED_BYTE, indices);
}
glDisableVertexAttribArray(shader_->a_position);
glDisableVertexAttribArray(shader_->a_texcoord0);
glstate.Restore();
}
protected:
DepalShader *shader_;
Pos pos_[4];
UV uv_[4];
float bufferW_;
float bufferH_;
int renderW_;
int renderH_;
};
void TextureCache::ApplyTextureFramebuffer(TexCacheEntry *entry, VirtualFramebuffer *framebuffer) {
DepalShader *depal = nullptr;
const GEPaletteFormat clutFormat = gstate.getClutPaletteFormat();
if ((entry->status & TexCacheEntry::STATUS_DEPALETTIZE) && !g_Config.bDisableSlowFramebufEffects) {
depal = depalShaderCache_->GetDepalettizeShader(clutFormat, framebuffer->drawnFormat);
}
if (depal) {
GLuint clutTexture = depalShaderCache_->GetClutTexture(clutFormat, clutHash_, clutBuf_);
FBO *depalFBO = framebufferManager_->GetTempFBO(framebuffer->renderWidth, framebuffer->renderHeight, FBO_8888);
fbo_bind_as_render_target(depalFBO);
shaderManager_->DirtyLastShader();
TextureShaderApplier shaderApply(depal, framebuffer->bufferWidth, framebuffer->bufferHeight, framebuffer->renderWidth, framebuffer->renderHeight);
shaderApply.ApplyBounds(gstate_c.vertBounds, gstate_c.curTextureXOffset, gstate_c.curTextureYOffset);
shaderApply.Use(transformDraw_);
glActiveTexture(GL_TEXTURE3);
glBindTexture(GL_TEXTURE_2D, clutTexture);
glActiveTexture(GL_TEXTURE0);
framebufferManager_->BindFramebufferColor(GL_TEXTURE0, gstate.getFrameBufRawAddress(), framebuffer, BINDFBCOLOR_SKIP_COPY);
glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_NEAREST);
glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_NEAREST);
shaderApply.Shade();
fbo_bind_color_as_texture(depalFBO, 0);
} else {
framebufferManager_->BindFramebufferColor(GL_TEXTURE0, gstate.getFrameBufRawAddress(), framebuffer, BINDFBCOLOR_MAY_COPY_WITH_UV | BINDFBCOLOR_APPLY_TEX_OFFSET);
}
framebufferManager_->RebindFramebuffer();
SetFramebufferSamplingParams(framebuffer->bufferWidth, framebuffer->bufferHeight);
lastBoundTexture = INVALID_TEX;
}
bool TextureCache::SetOffsetTexture(u32 offset) {
if (g_Config.iRenderingMode != FB_BUFFERED_MODE) {
return false;
}
u32 texaddr = gstate.getTextureAddress(0);
if (!Memory::IsValidAddress(texaddr) || !Memory::IsValidAddress(texaddr + offset)) {
return false;
}
const u16 dim = gstate.getTextureDimension(0);
u64 cachekey = ((u64)(texaddr & 0x3FFFFFFF) << 32) | dim;
TexCache::iterator iter = cache.find(cachekey);
if (iter == cache.end()) {
return false;
}
TexCacheEntry *entry = &iter->second;
bool success = false;
for (size_t i = 0, n = fbCache_.size(); i < n; ++i) {
auto framebuffer = fbCache_[i];
if (AttachFramebuffer(entry, framebuffer->fb_address, framebuffer, offset)) {
success = true;
}
}
if (success && entry->framebuffer) {
// This will not apply the texture immediately.
SetTextureFramebuffer(entry, entry->framebuffer);
entry->lastFrame = gpuStats.numFlips;
return true;
}
return false;
}
void TextureCache::SetTexture(bool force) {
#ifdef DEBUG_TEXTURES
if (SetDebugTexture()) {
// A different texture was bound, let's rebind next time.
lastBoundTexture = INVALID_TEX;
return;
}
#endif
if (force) {
lastBoundTexture = INVALID_TEX;
}
u32 texaddr = gstate.getTextureAddress(0);
if (!Memory::IsValidAddress(texaddr)) {
// Bind a null texture and return.
glBindTexture(GL_TEXTURE_2D, 0);
lastBoundTexture = INVALID_TEX;
return;
}
const u16 dim = gstate.getTextureDimension(0);
int w = gstate.getTextureWidth(0);
int h = gstate.getTextureHeight(0);
GETextureFormat format = gstate.getTextureFormat();
if (format >= 11) {
ERROR_LOG_REPORT(G3D, "Unknown texture format %i", format);
// TODO: Better assumption?
format = GE_TFMT_5650;
}
bool hasClut = gstate.isTextureFormatIndexed();
// Ignore uncached/kernel when caching.
u64 cachekey = ((u64)(texaddr & 0x3FFFFFFF) << 32) | dim;
u32 cluthash;
if (hasClut) {
if (clutLastFormat_ != gstate.clutformat) {
// We update here because the clut format can be specified after the load.
UpdateCurrentClut(gstate.getClutPaletteFormat(), gstate.getClutIndexStartPos(), gstate.isClutIndexSimple());
}
cluthash = GetCurrentClutHash() ^ gstate.clutformat;
cachekey ^= cluthash;
} else {
cluthash = 0;
}
int bufw = GetTextureBufw(0, texaddr, format);
u8 maxLevel = gstate.getTextureMaxLevel();
u32 texhash = MiniHash((const u32 *)Memory::GetPointerUnchecked(texaddr));
u32 fullhash = 0;
TexCache::iterator iter = cache.find(cachekey);
TexCacheEntry *entry = NULL;
gstate_c.needShaderTexClamp = false;
gstate_c.skipDrawReason &= ~SKIPDRAW_BAD_FB_TEXTURE;
bool replaceImages = false;
if (iter != cache.end()) {
entry = &iter->second;
// Validate the texture still matches the cache entry.
bool match = entry->Matches(dim, format, maxLevel);
const char *reason = "different params";
// Check for FBO - slow!
if (entry->framebuffer) {
if (match) {
if (hasClut && clutRenderAddress_ != 0xFFFFFFFF) {
WARN_LOG_REPORT_ONCE(clutAndTexRender, G3D, "Using rendered texture with rendered CLUT: texfmt=%d, clutfmt=%d", gstate.getTextureFormat(), gstate.getClutPaletteFormat());
}
SetTextureFramebuffer(entry, entry->framebuffer);
entry->lastFrame = gpuStats.numFlips;
return;
} else {
// Make sure we re-evaluate framebuffers.
DetachFramebuffer(entry, texaddr, entry->framebuffer);
reason = "detached framebuf";
match = false;
}
}
bool rehash = entry->GetHashStatus() == TexCacheEntry::STATUS_UNRELIABLE;
bool doDelete = true;
// First let's see if another texture with the same address had a hashfail.
if (entry->status & TexCacheEntry::STATUS_CLUT_RECHECK) {
// Always rehash in this case, if one changed the rest all probably did.
rehash = true;
entry->status &= ~TexCacheEntry::STATUS_CLUT_RECHECK;
} else if ((gstate_c.textureChanged & TEXCHANGE_UPDATED) == 0) {
// Okay, just some parameter change - the data didn't change, no need to rehash.
rehash = false;
}
if (match) {
if (entry->lastFrame != gpuStats.numFlips) {
u32 diff = gpuStats.numFlips - entry->lastFrame;
entry->numFrames++;
if (entry->framesUntilNextFullHash < diff) {
// Exponential backoff up to 512 frames. Textures are often reused.
if (entry->numFrames > 32) {
// Also, try to add some "randomness" to avoid rehashing several textures the same frame.
entry->framesUntilNextFullHash = std::min(512, entry->numFrames) + (entry->textureName & 15);
} else {
entry->framesUntilNextFullHash = entry->numFrames;
}
rehash = true;
} else {
entry->framesUntilNextFullHash -= diff;
}
}
// If it's not huge or has been invalidated many times, recheck the whole texture.
if (entry->invalidHint > 180 || (entry->invalidHint > 15 && (dim >> 8) < 9 && (dim & 0xF) < 9)) {
entry->invalidHint = 0;
rehash = true;
}
bool hashFail = false;
if (texhash != entry->hash) {
fullhash = QuickTexHash(texaddr, bufw, w, h, format, entry);
hashFail = true;
rehash = false;
}
if (rehash && entry->GetHashStatus() != TexCacheEntry::STATUS_RELIABLE) {
fullhash = QuickTexHash(texaddr, bufw, w, h, format, entry);
if (fullhash != entry->fullhash) {
hashFail = true;
} else {
if (g_Config.bTextureBackoffCache) {
if (entry->GetHashStatus() != TexCacheEntry::STATUS_HASHING && entry->numFrames > TexCacheEntry::FRAMES_REGAIN_TRUST) {
// Reset to STATUS_HASHING.
entry->SetHashStatus(TexCacheEntry::STATUS_HASHING);
entry->status &= ~TexCacheEntry::STATUS_CHANGE_FREQUENT;
}
} else if (entry->numFrames > TEXCACHE_FRAME_CHANGE_FREQUENT_REGAIN_TRUST) {
entry->status &= ~TexCacheEntry::STATUS_CHANGE_FREQUENT;
}
}
}
if (hashFail) {
match = false;
reason = "hash fail";
entry->status |= TexCacheEntry::STATUS_UNRELIABLE;
if (entry->numFrames < TEXCACHE_FRAME_CHANGE_FREQUENT) {
if (entry->status & TexCacheEntry::STATUS_FREE_CHANGE) {
entry->status &= ~TexCacheEntry::STATUS_FREE_CHANGE;
} else {
entry->status |= TexCacheEntry::STATUS_CHANGE_FREQUENT;
}
}
entry->numFrames = 0;
// Don't give up just yet. Let's try the secondary cache if it's been invalidated before.
// If it's failed a bunch of times, then the second cache is just wasting time and VRAM.
if (g_Config.bTextureSecondaryCache) {
if (entry->numInvalidated > 2 && entry->numInvalidated < 128 && !lowMemoryMode_) {
u64 secondKey = fullhash | (u64)cluthash << 32;
TexCache::iterator secondIter = secondCache.find(secondKey);
if (secondIter != secondCache.end()) {
TexCacheEntry *secondEntry = &secondIter->second;
if (secondEntry->Matches(dim, format, maxLevel)) {
// Reset the numInvalidated value lower, we got a match.
if (entry->numInvalidated > 8) {
--entry->numInvalidated;
}
entry = secondEntry;
match = true;
}
} else {
secondKey = entry->fullhash | ((u64)entry->cluthash << 32);
secondCacheSizeEstimate_ += EstimateTexMemoryUsage(entry);
secondCache[secondKey] = *entry;
doDelete = false;
}
}
}
}
}
if (match && (entry->status & TexCacheEntry::STATUS_TO_SCALE) && standardScaleFactor_ != 1 && texelsScaledThisFrame_ < TEXCACHE_MAX_TEXELS_SCALED) {
if ((entry->status & TexCacheEntry::STATUS_CHANGE_FREQUENT) == 0) {
// INFO_LOG(G3D, "Reloading texture to do the scaling we skipped..");
match = false;
reason = "scaling";
}
}
if (match) {
// TODO: Mark the entry reliable if it's been safe for long enough?
//got one!
entry->lastFrame = gpuStats.numFlips;
if (entry->textureName != lastBoundTexture) {
gstate_c.textureFullAlpha = entry->GetAlphaStatus() == TexCacheEntry::STATUS_ALPHA_FULL;
gstate_c.textureSimpleAlpha = entry->GetAlphaStatus() != TexCacheEntry::STATUS_ALPHA_UNKNOWN;
}
nextTexture_ = entry;
VERBOSE_LOG(G3D, "Texture at %08x Found in Cache, applying", texaddr);
return; //Done!
} else {
cacheSizeEstimate_ -= EstimateTexMemoryUsage(entry);
entry->numInvalidated++;
gpuStats.numTextureInvalidations++;
DEBUG_LOG(G3D, "Texture different or overwritten, reloading at %08x: %s", texaddr, reason);
if (doDelete) {
if (entry->maxLevel == maxLevel && entry->dim == gstate.getTextureDimension(0) && entry->format == format && standardScaleFactor_ == 1) {
// Actually, if size and number of levels match, let's try to avoid deleting and recreating.
// Instead, let's use glTexSubImage to replace the images.
replaceImages = true;
} else {
if (entry->textureName == lastBoundTexture) {
lastBoundTexture = INVALID_TEX;
}
glDeleteTextures(1, &entry->textureName);
}
}
// Clear the reliable bit if set.
if (entry->GetHashStatus() == TexCacheEntry::STATUS_RELIABLE) {
entry->SetHashStatus(TexCacheEntry::STATUS_HASHING);
}
// Also, mark any textures with the same address but different clut. They need rechecking.
if (cluthash != 0) {
const u64 cachekeyMin = (u64)(texaddr & 0x3FFFFFFF) << 32;
const u64 cachekeyMax = cachekeyMin + (1ULL << 32);
for (auto it = cache.lower_bound(cachekeyMin), end = cache.upper_bound(cachekeyMax); it != end; ++it) {
if (it->second.cluthash != cluthash) {
it->second.status |= TexCacheEntry::STATUS_CLUT_RECHECK;
}
}
}
}
} else {
VERBOSE_LOG(G3D, "No texture in cache, decoding...");
TexCacheEntry entryNew = {0};
cache[cachekey] = entryNew;
if (hasClut && clutRenderAddress_ != 0xFFFFFFFF) {
WARN_LOG_REPORT_ONCE(clutUseRender, G3D, "Using texture with rendered CLUT: texfmt=%d, clutfmt=%d", gstate.getTextureFormat(), gstate.getClutPaletteFormat());
}
entry = &cache[cachekey];
if (g_Config.bTextureBackoffCache) {
entry->status = TexCacheEntry::STATUS_HASHING;
} else {
entry->status = TexCacheEntry::STATUS_UNRELIABLE;
}
}
if ((bufw == 0 || (gstate.texbufwidth[0] & 0xf800) != 0) && texaddr >= PSP_GetKernelMemoryEnd()) {
ERROR_LOG_REPORT(G3D, "Texture with unexpected bufw (full=%d)", gstate.texbufwidth[0] & 0xffff);
// Proceeding here can cause a crash.
return;
}
// We have to decode it, let's setup the cache entry first.
entry->addr = texaddr;
entry->hash = texhash;
entry->format = format;
entry->lastFrame = gpuStats.numFlips;
entry->framebuffer = 0;
entry->maxLevel = maxLevel;
entry->lodBias = 0.0f;
entry->dim = gstate.getTextureDimension(0);
entry->bufw = bufw;
// This would overestimate the size in many case so we underestimate instead
// to avoid excessive clearing caused by cache invalidations.
entry->sizeInRAM = (textureBitsPerPixel[format] * bufw * h / 2) / 8;
entry->fullhash = fullhash == 0 ? QuickTexHash(texaddr, bufw, w, h, format, entry) : fullhash;
entry->cluthash = cluthash;
entry->status &= ~TexCacheEntry::STATUS_ALPHA_MASK;
gstate_c.curTextureWidth = w;
gstate_c.curTextureHeight = h;
// For the estimate, we assume cluts always point to 8888 for simplicity.
cacheSizeEstimate_ += EstimateTexMemoryUsage(entry);
// Always generate a texture name, we might need it if the texture is replaced later.
if (!replaceImages) {
entry->textureName = AllocTextureName();
}
// Before we go reading the texture from memory, let's check for render-to-texture.
for (size_t i = 0, n = fbCache_.size(); i < n; ++i) {
auto framebuffer = fbCache_[i];
AttachFramebuffer(entry, framebuffer->fb_address, framebuffer);
}
// If we ended up with a framebuffer, attach it - no texture decoding needed.
if (entry->framebuffer) {
SetTextureFramebuffer(entry, entry->framebuffer);
entry->lastFrame = gpuStats.numFlips;
return;
}
glBindTexture(GL_TEXTURE_2D, entry->textureName);
lastBoundTexture = entry->textureName;
// Adjust maxLevel to actually present levels..
bool badMipSizes = false;
for (u32 i = 0; i <= maxLevel; i++) {
// If encountering levels pointing to nothing, adjust max level.
u32 levelTexaddr = gstate.getTextureAddress(i);
if (!Memory::IsValidAddress(levelTexaddr)) {
maxLevel = i - 1;
break;
}
if (i > 0 && gstate_c.Supports(GPU_SUPPORTS_TEXTURE_LOD_CONTROL)) {
int tw = gstate.getTextureWidth(i);
int th = gstate.getTextureHeight(i);
if (tw != 1 && tw != (gstate.getTextureWidth(i - 1) >> 1))
badMipSizes = true;
else if (th != 1 && th != (gstate.getTextureHeight(i - 1) >> 1))
badMipSizes = true;
}
}
// In addition, simply don't load more than level 0 if g_Config.bMipMap is false.
if (!g_Config.bMipMap) {
maxLevel = 0;
}
// If GLES3 is available, we can preallocate the storage, which makes texture loading more efficient.
GLenum dstFmt = GetDestFormat(format, gstate.getClutPaletteFormat());
int scaleFactor = standardScaleFactor_;
// Rachet down scale factor in low-memory mode.
if (lowMemoryMode_) {
// Keep it even, though, just in case of npot troubles.
scaleFactor = scaleFactor > 4 ? 4 : (scaleFactor > 2 ? 2 : 1);
}
// Don't scale the PPGe texture.
if (entry->addr > 0x05000000 && entry->addr < 0x08800000)
scaleFactor = 1;
if ((entry->status & TexCacheEntry::STATUS_CHANGE_FREQUENT) != 0) {
// Remember for later that we /wanted/ to scale this texture.
entry->status |= TexCacheEntry::STATUS_TO_SCALE;
scaleFactor = 1;
}
if (scaleFactor != 1) {
if (texelsScaledThisFrame_ >= TEXCACHE_MAX_TEXELS_SCALED) {
entry->status |= TexCacheEntry::STATUS_TO_SCALE;
scaleFactor = 1;
} else {
entry->status &= ~TexCacheEntry::STATUS_TO_SCALE;
texelsScaledThisFrame_ += w * h;
}
}
// Disabled this due to issue #6075: https://github.com/hrydgard/ppsspp/issues/6075
// This breaks Dangan Ronpa 2 with mipmapping enabled. Why? No idea, it shouldn't.
// glTexStorage2D probably has few benefits for us anyway.
if (false && gl_extensions.GLES3 && maxLevel > 0) {
// glTexStorage2D requires the use of sized formats.
GLenum storageFmt = GL_RGBA8;
switch (dstFmt) {
case GL_UNSIGNED_BYTE: storageFmt = GL_RGBA8; break;
case GL_UNSIGNED_SHORT_5_6_5: storageFmt = GL_RGB565; break;
case GL_UNSIGNED_SHORT_4_4_4_4: storageFmt = GL_RGBA4; break;
case GL_UNSIGNED_SHORT_5_5_5_1: storageFmt = GL_RGB5_A1; break;
default:
ERROR_LOG(G3D, "Unknown dstfmt %i", (int)dstFmt);
break;
}
// TODO: This may cause bugs, since it hard-sets the texture w/h, and we might try to reuse it later with a different size.
glTexStorage2D(GL_TEXTURE_2D, maxLevel + 1, storageFmt, w * scaleFactor, h * scaleFactor);
// Make sure we don't use glTexImage2D after glTexStorage2D.
replaceImages = true;
}
// GLES2 doesn't have support for a "Max lod" which is critical as PSP games often
// don't specify mips all the way down. As a result, we either need to manually generate
// the bottom few levels or rely on OpenGL's autogen mipmaps instead, which might not
// be as good quality as the game's own (might even be better in some cases though).
// Always load base level texture here
LoadTextureLevel(*entry, 0, replaceImages, scaleFactor, dstFmt);
// Mipmapping only enable when texture scaling disable
if (maxLevel > 0 && scaleFactor == 1) {
if (gstate_c.Supports(GPU_SUPPORTS_TEXTURE_LOD_CONTROL)) {
if (badMipSizes) {
// WARN_LOG(G3D, "Bad mipmap for texture sized %dx%dx%d - autogenerating", w, h, (int)format);
glGenerateMipmap(GL_TEXTURE_2D);
} else {
for (int i = 1; i <= maxLevel; i++) {
LoadTextureLevel(*entry, i, replaceImages, scaleFactor, dstFmt);
}
glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAX_LEVEL, maxLevel);
glTexParameterf(GL_TEXTURE_2D, GL_TEXTURE_MAX_LOD, (float)maxLevel);
}
} else {
// Avoid PowerVR driver bug
if (w > 1 && h > 1 && !(h > w && (gl_extensions.bugs & BUG_PVR_GENMIPMAP_HEIGHT_GREATER))) { // Really! only seems to fail if height > width
// NOTICE_LOG(G3D, "Generating mipmap for texture sized %dx%d%d", w, h, (int)format);
glGenerateMipmap(GL_TEXTURE_2D);
} else {
entry->maxLevel = 0;
}
}
} else if (gstate_c.Supports(GPU_SUPPORTS_TEXTURE_LOD_CONTROL)) {
glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAX_LEVEL, 0);
}
int aniso = 1 << g_Config.iAnisotropyLevel;
float anisotropyLevel = (float) aniso > maxAnisotropyLevel ? maxAnisotropyLevel : (float) aniso;
glTexParameterf(GL_TEXTURE_2D, GL_TEXTURE_MAX_ANISOTROPY_EXT, anisotropyLevel);
gstate_c.textureFullAlpha = entry->GetAlphaStatus() == TexCacheEntry::STATUS_ALPHA_FULL;
gstate_c.textureSimpleAlpha = entry->GetAlphaStatus() != TexCacheEntry::STATUS_ALPHA_UNKNOWN;
// This will rebind it, but that's okay.
nextTexture_ = entry;
UpdateSamplingParams(*entry, true);
//glPixelStorei(GL_UNPACK_ROW_LENGTH, 0);
glPixelStorei(GL_UNPACK_ALIGNMENT, 1);
//glPixelStorei(GL_PACK_ROW_LENGTH, 0);
glPixelStorei(GL_PACK_ALIGNMENT, 1);
}
u32 TextureCache::AllocTextureName() {
if (nameCache_.empty()) {
nameCache_.resize(TEXCACHE_NAME_CACHE_SIZE);
glGenTextures(TEXCACHE_NAME_CACHE_SIZE, &nameCache_[0]);
}
u32 name = nameCache_.back();
nameCache_.pop_back();
return name;
}
GLenum TextureCache::GetDestFormat(GETextureFormat format, GEPaletteFormat clutFormat) const {
switch (format) {
case GE_TFMT_CLUT4:
case GE_TFMT_CLUT8:
case GE_TFMT_CLUT16:
case GE_TFMT_CLUT32:
return getClutDestFormat(clutFormat);
case GE_TFMT_4444:
return GL_UNSIGNED_SHORT_4_4_4_4;
case GE_TFMT_5551:
return GL_UNSIGNED_SHORT_5_5_5_1;
case GE_TFMT_5650:
return GL_UNSIGNED_SHORT_5_6_5;
case GE_TFMT_8888:
case GE_TFMT_DXT1:
case GE_TFMT_DXT3:
case GE_TFMT_DXT5:
default:
return GL_UNSIGNED_BYTE;
}
}
void *TextureCache::DecodeTextureLevel(GETextureFormat format, GEPaletteFormat clutformat, int level, u32 &texByteAlign, GLenum dstFmt, int scaleFactor, int *bufwout) {
void *finalBuf = NULL;
u32 texaddr = gstate.getTextureAddress(level);
bool swizzled = gstate.isTextureSwizzled();
if ((texaddr & 0x00600000) != 0 && Memory::IsVRAMAddress(texaddr)) {
// This means it's in a mirror, possibly a swizzled mirror. Let's report.
WARN_LOG_REPORT_ONCE(texmirror, G3D, "Decoding texture from VRAM mirror at %08x swizzle=%d", texaddr, swizzled ? 1 : 0);
if ((texaddr & 0x00200000) == 0x00200000) {
// Technically 2 and 6 are slightly different, but this is better than nothing probably.
swizzled = !swizzled;
}
// Note that (texaddr & 0x00600000) == 0x00600000 is very likely to be depth texturing.
}
int bufw = GetTextureBufw(level, texaddr, format);
if (bufwout)
*bufwout = bufw;
int w = gstate.getTextureWidth(level);
int h = gstate.getTextureHeight(level);
const u8 *texptr = Memory::GetPointer(texaddr);
switch (format) {
case GE_TFMT_CLUT4:
{
const bool mipmapShareClut = gstate.isClutSharedForMipmaps();
const int clutSharingOffset = mipmapShareClut ? 0 : level * 16;
switch (clutformat) {
case GE_CMODE_16BIT_BGR5650:
case GE_CMODE_16BIT_ABGR5551:
case GE_CMODE_16BIT_ABGR4444:
{
tmpTexBuf16.resize(std::max(bufw, w) * h);
tmpTexBufRearrange.resize(std::max(bufw, w) * h);
const u16 *clut = GetCurrentClut<u16>() + clutSharingOffset;
texByteAlign = 2;
if (!swizzled) {
if (clutAlphaLinear_ && mipmapShareClut) {
DeIndexTexture4Optimal(tmpTexBuf16.data(), texptr, bufw * h, clutAlphaLinearColor_);
} else {
DeIndexTexture4(tmpTexBuf16.data(), texptr, bufw * h, clut);
}
} else {
tmpTexBuf32.resize(std::max(bufw, w) * h);
UnswizzleFromMem(texptr, bufw, h, 0);
if (clutAlphaLinear_ && mipmapShareClut) {
DeIndexTexture4Optimal(tmpTexBuf16.data(), (const u8 *)tmpTexBuf32.data(), bufw * h, clutAlphaLinearColor_);
} else {
DeIndexTexture4(tmpTexBuf16.data(), (const u8 *)tmpTexBuf32.data(), bufw * h, clut);
}
}
finalBuf = tmpTexBuf16.data();
}
break;
case GE_CMODE_32BIT_ABGR8888:
{
tmpTexBuf32.resize(std::max(bufw, w) * h);
tmpTexBufRearrange.resize(std::max(bufw, w) * h);
const u32 *clut = GetCurrentClut<u32>() + clutSharingOffset;
if (!swizzled) {
DeIndexTexture4(tmpTexBuf32.data(), texptr, bufw * h, clut);
finalBuf = tmpTexBuf32.data();
} else {
UnswizzleFromMem(texptr, bufw, h, 0);
// Let's reuse tmpTexBuf16, just need double the space.
tmpTexBuf16.resize(std::max(bufw, w) * h * 2);
DeIndexTexture4((u32 *)tmpTexBuf16.data(), (u8 *)tmpTexBuf32.data(), bufw * h, clut);
finalBuf = tmpTexBuf16.data();
}
}
break;
default:
ERROR_LOG_REPORT(G3D, "Unknown CLUT4 texture mode %d", gstate.getClutPaletteFormat());
return NULL;
}
}
break;
case GE_TFMT_CLUT8:
texByteAlign = texByteAlignMap[gstate.getClutPaletteFormat()];
finalBuf = ReadIndexedTex(level, texptr, 1, dstFmt, bufw);
break;
case GE_TFMT_CLUT16:
texByteAlign = texByteAlignMap[gstate.getClutPaletteFormat()];
finalBuf = ReadIndexedTex(level, texptr, 2, dstFmt, bufw);
break;
case GE_TFMT_CLUT32:
texByteAlign = texByteAlignMap[gstate.getClutPaletteFormat()];
finalBuf = ReadIndexedTex(level, texptr, 4, dstFmt, bufw);
break;
case GE_TFMT_4444:
case GE_TFMT_5551:
case GE_TFMT_5650:
texByteAlign = 2;
if (!swizzled) {
int len = std::max(bufw, w) * h;
tmpTexBuf16.resize(len);
tmpTexBufRearrange.resize(len);
finalBuf = tmpTexBuf16.data();
ConvertColors(finalBuf, texptr, dstFmt, bufw * h);
} else {
tmpTexBuf32.resize(std::max(bufw, w) * h);
finalBuf = UnswizzleFromMem(texptr, bufw, h, 2);
ConvertColors(finalBuf, finalBuf, dstFmt, bufw * h);
}
break;
case GE_TFMT_8888:
if (!swizzled) {
// Special case: if we don't need to deal with packing, we don't need to copy.
if ((scaleFactor == 1 && gstate_c.Supports(GPU_SUPPORTS_UNPACK_SUBIMAGE)) || w == bufw) {
if (UseBGRA8888()) {
tmpTexBuf32.resize(std::max(bufw, w) * h);
finalBuf = tmpTexBuf32.data();
ConvertColors(finalBuf, texptr, dstFmt, bufw * h);
} else {
finalBuf = (void *)texptr;
}
} else {
tmpTexBuf32.resize(std::max(bufw, w) * h);
tmpTexBufRearrange.resize(std::max(bufw, w) * h);
finalBuf = tmpTexBuf32.data();
ConvertColors(finalBuf, texptr, dstFmt, bufw * h);
}
} else {
tmpTexBuf32.resize(std::max(bufw, w) * h);
finalBuf = UnswizzleFromMem(texptr, bufw, h, 4);
ConvertColors(finalBuf, finalBuf, dstFmt, bufw * h);
}
break;
case GE_TFMT_DXT1:
{
int minw = std::min(bufw, w);
tmpTexBuf32.resize(std::max(bufw, w) * h);
tmpTexBufRearrange.resize(std::max(bufw, w) * h);
u32 *dst = tmpTexBuf32.data();
DXT1Block *src = (DXT1Block*)texptr;
for (int y = 0; y < h; y += 4) {
u32 blockIndex = (y / 4) * (bufw / 4);
for (int x = 0; x < minw; x += 4) {
DecodeDXT1Block(dst + bufw * y + x, src + blockIndex, bufw);
blockIndex++;
}
}
finalBuf = tmpTexBuf32.data();
ConvertColors(finalBuf, finalBuf, dstFmt, bufw * h);
w = (w + 3) & ~3;
}
break;
case GE_TFMT_DXT3:
{
int minw = std::min(bufw, w);
tmpTexBuf32.resize(std::max(bufw, w) * h);
tmpTexBufRearrange.resize(std::max(bufw, w) * h);
u32 *dst = tmpTexBuf32.data();
DXT3Block *src = (DXT3Block*)texptr;
for (int y = 0; y < h; y += 4) {
u32 blockIndex = (y / 4) * (bufw / 4);
for (int x = 0; x < minw; x += 4) {
DecodeDXT3Block(dst + bufw * y + x, src + blockIndex, bufw);
blockIndex++;
}
}
w = (w + 3) & ~3;
finalBuf = tmpTexBuf32.data();
ConvertColors(finalBuf, finalBuf, dstFmt, bufw * h);
}
break;
case GE_TFMT_DXT5:
{
int minw = std::min(bufw, w);
tmpTexBuf32.resize(std::max(bufw, w) * h);
tmpTexBufRearrange.resize(std::max(bufw, w) * h);
u32 *dst = tmpTexBuf32.data();
DXT5Block *src = (DXT5Block*)texptr;
for (int y = 0; y < h; y += 4) {
u32 blockIndex = (y / 4) * (bufw / 4);
for (int x = 0; x < minw; x += 4) {
DecodeDXT5Block(dst + bufw * y + x, src + blockIndex, bufw);
blockIndex++;
}
}
w = (w + 3) & ~3;
finalBuf = tmpTexBuf32.data();
ConvertColors(finalBuf, finalBuf, dstFmt, bufw * h);
}
break;
default:
ERROR_LOG_REPORT(G3D, "Unknown Texture Format %d!!!", format);
return NULL;
}
if (!finalBuf) {
ERROR_LOG_REPORT(G3D, "NO finalbuf! Will crash!");
}
if (!(scaleFactor == 1 && gstate_c.Supports(GPU_SUPPORTS_UNPACK_SUBIMAGE)) && w != bufw) {
int pixelSize;
switch (dstFmt) {
case GL_UNSIGNED_SHORT_4_4_4_4:
case GL_UNSIGNED_SHORT_5_5_5_1:
case GL_UNSIGNED_SHORT_5_6_5:
pixelSize = 2;
break;
default:
pixelSize = 4;
break;
}
// Need to rearrange the buffer to simulate GL_UNPACK_ROW_LENGTH etc.
finalBuf = RearrangeBuf(finalBuf, bufw * pixelSize, w * pixelSize, h);
}
return finalBuf;
}
TextureCache::TexCacheEntry::Status TextureCache::CheckAlpha(const u32 *pixelData, GLenum dstFmt, int stride, int w, int h) {
CheckAlphaResult res;
switch (dstFmt) {
case GL_UNSIGNED_SHORT_4_4_4_4:
res = CheckAlphaABGR4444Basic(pixelData, stride, w, h);
break;
case GL_UNSIGNED_SHORT_5_5_5_1:
res = CheckAlphaABGR1555Basic(pixelData, stride, w, h);
break;
case GL_UNSIGNED_SHORT_5_6_5:
// Never has any alpha.
res = CHECKALPHA_FULL;
break;
default:
res = CheckAlphaRGBA8888Basic(pixelData, stride, w, h);
break;
}
return (TexCacheEntry::Status)res;
}
void TextureCache::LoadTextureLevel(TexCacheEntry &entry, int level, bool replaceImages, int scaleFactor, GLenum dstFmt) {
int w = gstate.getTextureWidth(level);
int h = gstate.getTextureHeight(level);
bool useUnpack = false;
bool useBGRA;
u32 *pixelData;
{
PROFILE_THIS_SCOPE("decodetex");
// TODO: only do this once
u32 texByteAlign = 1;
GEPaletteFormat clutformat = gstate.getClutPaletteFormat();
int bufw;
void *finalBuf = DecodeTextureLevel(GETextureFormat(entry.format), clutformat, level, texByteAlign, dstFmt, scaleFactor, &bufw);
if (finalBuf == NULL) {
return;
}
gpuStats.numTexturesDecoded++;
// Can restore these and remove the fixup at the end of DecodeTextureLevel on desktop GL and GLES 3.
if (scaleFactor == 1 && gstate_c.Supports(GPU_SUPPORTS_UNPACK_SUBIMAGE) && w != bufw) {
glPixelStorei(GL_UNPACK_ROW_LENGTH, bufw);
useUnpack = true;
}
glPixelStorei(GL_UNPACK_ALIGNMENT, texByteAlign);
useBGRA = UseBGRA8888() && dstFmt == GL_UNSIGNED_BYTE;
pixelData = (u32 *)finalBuf;
if (scaleFactor > 1)
scaler.Scale(pixelData, dstFmt, w, h, scaleFactor);
if ((entry.status & TexCacheEntry::STATUS_CHANGE_FREQUENT) == 0) {
TexCacheEntry::Status alphaStatus = CheckAlpha(pixelData, dstFmt, useUnpack ? bufw : w, w, h);
entry.SetAlphaStatus(alphaStatus, level);
} else {
entry.SetAlphaStatus(TexCacheEntry::STATUS_ALPHA_UNKNOWN);
}
}
GLuint components = dstFmt == GL_UNSIGNED_SHORT_5_6_5 ? GL_RGB : GL_RGBA;
GLuint components2 = components;
if (useBGRA) {
components2 = GL_BGRA_EXT;
}
if (replaceImages) {
PROFILE_THIS_SCOPE("repltex");
glTexSubImage2D(GL_TEXTURE_2D, level, 0, 0, w, h, components2, dstFmt, pixelData);
} else {
PROFILE_THIS_SCOPE("loadtex");
glTexImage2D(GL_TEXTURE_2D, level, components, w, h, 0, components2, dstFmt, pixelData);
if (!lowMemoryMode_) {
GLenum err = glGetError();
if (err == GL_OUT_OF_MEMORY) {
WARN_LOG_REPORT(G3D, "Texture cache ran out of GPU memory; switching to low memory mode");
lowMemoryMode_ = true;
decimationCounter_ = 0;
Decimate();
// Try again, now that we've cleared out textures in lowMemoryMode_.
glTexImage2D(GL_TEXTURE_2D, level, components, w, h, 0, components2, dstFmt, pixelData);
I18NCategory *err = GetI18NCategory("Error");
if (scaleFactor > 1) {
osm.Show(err->T("Warning: Video memory FULL, reducing upscaling and switching to slow caching mode"), 2.0f);
} else {
osm.Show(err->T("Warning: Video memory FULL, switching to slow caching mode"), 2.0f);
}
} else if (err != GL_NO_ERROR) {
// We checked the err anyway, might as well log if there is one.
WARN_LOG(G3D, "Got an error in texture upload: %08x", err);
}
}
}
if (useUnpack) {
glPixelStorei(GL_UNPACK_ROW_LENGTH, 0);
}
}
// Only used by Qt UI?
bool TextureCache::DecodeTexture(u8* output, const GPUgstate &state) {
GPUgstate oldState = gstate;
gstate = state;
u32 texaddr = gstate.getTextureAddress(0);
if (!Memory::IsValidAddress(texaddr)) {
return false;
}
u32 texByteAlign = 1;
GLenum dstFmt = 0;
GETextureFormat format = gstate.getTextureFormat();
GEPaletteFormat clutformat = gstate.getClutPaletteFormat();
u8 level = 0;
int bufw = GetTextureBufw(level, texaddr, format);
int w = gstate.getTextureWidth(level);
int h = gstate.getTextureHeight(level);
void *finalBuf = DecodeTextureLevel(format, clutformat, level, texByteAlign, dstFmt, 1);
if (finalBuf == NULL) {
return false;
}
switch (dstFmt) {
case GL_UNSIGNED_SHORT_4_4_4_4:
for (int y = 0; y < h; y++)
for (int x = 0; x < bufw; x++) {
u32 val = ((u16*)finalBuf)[y*bufw + x];
u32 r = ((val>>12) & 0xF) * 17;
u32 g = ((val>> 8) & 0xF) * 17;
u32 b = ((val>> 4) & 0xF) * 17;
u32 a = ((val>> 0) & 0xF) * 17;
((u32*)output)[y*w + x] = (a << 24) | (r << 16) | (g << 8) | b;
}
break;
case GL_UNSIGNED_SHORT_5_5_5_1:
for (int y = 0; y < h; y++)
for (int x = 0; x < bufw; x++) {
u32 val = ((u16*)finalBuf)[y*bufw + x];
u32 r = Convert5To8((val>>11) & 0x1F);
u32 g = Convert5To8((val>> 6) & 0x1F);
u32 b = Convert5To8((val>> 1) & 0x1F);
u32 a = (val & 0x1) * 255;
((u32*)output)[y*w + x] = (a << 24) | (r << 16) | (g << 8) | b;
}
break;
case GL_UNSIGNED_SHORT_5_6_5:
for (int y = 0; y < h; y++)
for (int x = 0; x < bufw; x++) {
u32 val = ((u16*)finalBuf)[y*bufw + x];
u32 a = 0xFF;
u32 r = Convert5To8((val>>11) & 0x1F);
u32 g = Convert6To8((val>> 5) & 0x3F);
u32 b = Convert5To8((val ) & 0x1F);
((u32*)output)[y*w + x] = (a << 24) | (r << 16) | (g << 8) | b;
}
break;
default:
for (int y = 0; y < h; y++)
for (int x = 0; x < bufw; x++) {
u32 val = ((u32*)finalBuf)[y*bufw + x];
((u32*)output)[y*w + x] = ((val & 0xFF000000)) | ((val & 0x00FF0000)>>16) | ((val & 0x0000FF00)) | ((val & 0x000000FF)<<16);
}
break;
}
gstate = oldState;
return true;
}
Reference in New Issue View Git Blame Copy Permalink