preserve-emulation/ppsspp
1
0 Fork 0
mirror of https://github.com/hrydgard/ppsspp.git synced 2026-10-11 05:26:24 +02:00
Code Issues Packages Projects Releases Wiki Activity

samplerjit: Refactor and reuse constant pool.

Browse Source 
It's just here to be rip accessible, the fixed values can be output just
once.
This commit is contained in:
Unknown W. Brackets committed 2022-01-01 16:58:05 -08:00
1 parent 40240be91c
commit 91c9343e87
3 files changed
+90 -51
Download Patch File Download Diff File Expand all files Collapse all files

No files matched your search

GPU/Software/Sampler.cpp
+14
View File
@@ -119,6 +119,20 @@ void SamplerJitCache::Clear() {
ClearCodeSpace(0);
cache_.clear();
addresses_.clear();
const10All16_ = nullptr;
const10Low_ = nullptr;
const10All8_ = nullptr;
constWidth256f_ = nullptr;
constHeight256f_ = nullptr;
constWidthMinus1i_ = nullptr;
constHeightMinus1i_ = nullptr;
constOnes32_ = nullptr;
constOnes16_ = nullptr;
constUNext_ = nullptr;
constVNext_ = nullptr;
}
void SamplerJitCache::Describe(const std::string &message) {
GPU/Software/Sampler.h
+2
View File
@@ -82,6 +82,8 @@ private:
Rasterizer::RegCache::Reg GetZeroVec();
Rasterizer::RegCache::Reg GetGState();
void WriteConstantPool(const SamplerID &id);
bool Jit_ReadTextureFormat(const SamplerID &id);
bool Jit_GetTexData(const SamplerID &id, int bitsPerTexel);
bool Jit_GetTexDataSwizzled(const SamplerID &id, int bitsPerTexel);
GPU/Software/SamplerX86.cpp
+74 -51
View File
@@ -152,6 +152,9 @@ LinearFunc SamplerJitCache::CompileLinear(const SamplerID &id) {
lockReg(XMM9, RegCache::VEC_ARG_COLOR);
#endif
// Let's drop some helpful constants here.
WriteConstantPool(id);
// We'll first write the nearest sampler, which we will CALL.
// This may differ slightly based on the "linear" flag.
const u8 *nearest = AlignCode16();
@@ -182,57 +185,6 @@ LinearFunc SamplerJitCache::CompileLinear(const SamplerID &id) {
unlockOptReg(RegCache::VEC_ARG_COLOR);
regCache_.Reset(true);
// Let's drop some helpful constants here.
const10All16_ = AlignCode16();
Write16(0x10); Write16(0x10); Write16(0x10); Write16(0x10);
Write16(0x10); Write16(0x10); Write16(0x10); Write16(0x10);
const10Low_ = AlignCode16();
Write16(0x10); Write16(0x10); Write16(0x10); Write16(0x10);
Write16(0); Write16(0); Write16(0); Write16(0);
const10All8_ = AlignCode16();
for (int i = 0; i < 16; ++i)
Write8(0x10);
if (!id.hasAnyMips) {
constWidth256f_ = AlignCode16();
float w256f = (1 << id.width0Shift) * 256;
Write32(*(uint32_t *)&w256f); Write32(*(uint32_t *)&w256f);
Write32(*(uint32_t *)&w256f); Write32(*(uint32_t *)&w256f);
constHeight256f_ = AlignCode16();
float h256f = (1 << id.height0Shift) * 256;
Write32(*(uint32_t *)&h256f); Write32(*(uint32_t *)&h256f);
Write32(*(uint32_t *)&h256f); Write32(*(uint32_t *)&h256f);
constWidthMinus1i_ = AlignCode16();
Write32((1 << id.width0Shift) - 1); Write32((1 << id.width0Shift) - 1);
Write32((1 << id.width0Shift) - 1); Write32((1 << id.width0Shift) - 1);
constHeightMinus1i_ = AlignCode16();
Write32((1 << id.height0Shift) - 1); Write32((1 << id.height0Shift) - 1);
Write32((1 << id.height0Shift) - 1); Write32((1 << id.height0Shift) - 1);
} else {
constWidth256f_ = nullptr;
constHeight256f_ = nullptr;
constWidthMinus1i_ = nullptr;
constHeightMinus1i_ = nullptr;
}
constOnes32_ = AlignCode16();
Write32(1); Write32(1); Write32(1); Write32(1);
constOnes16_ = AlignCode16();
Write16(1); Write16(1); Write16(1); Write16(1);
Write16(1); Write16(1); Write16(1); Write16(1);
constUNext_ = AlignCode16();
Write32(0); Write32(1); Write32(0); Write32(1);
constVNext_ = AlignCode16();
Write32(0); Write32(0); Write32(1); Write32(1);
// Now the actual linear func, which is exposed externally.
const u8 *start = AlignCode16();
Describe("Init");
@@ -579,6 +531,77 @@ LinearFunc SamplerJitCache::CompileLinear(const SamplerID &id) {
return (LinearFunc)start;
}
void SamplerJitCache::WriteConstantPool(const SamplerID &id) {
// We reuse constants in any pool, because our code space is small.
if (const10All16_ == nullptr) {
const10All16_ = AlignCode16();
for (int i = 0; i < 8; ++i)
Write16(0x10);
}
if (const10Low_ == nullptr) {
const10Low_ = AlignCode16();
for (int i = 0; i < 4; ++i)
Write16(0x10);
for (int i = 0; i < 4; ++i)
Write16(0);
}
if (const10All8_ == nullptr) {
const10All8_ = AlignCode16();
for (int i = 0; i < 16; ++i)
Write8(0x10);
}
if (constOnes32_ == nullptr) {
constOnes32_ = AlignCode16();
for (int i = 0; i < 4; ++i)
Write32(1);
}
if (constOnes16_ == nullptr) {
constOnes16_ = AlignCode16();
for (int i = 0; i < 8; ++i)
Write16(1);
}
if (constUNext_ == nullptr) {
constUNext_ = AlignCode16();
Write32(0); Write32(1); Write32(0); Write32(1);
}
if (constVNext_ == nullptr) {
constVNext_ = AlignCode16();
Write32(0); Write32(0); Write32(1); Write32(1);
}
// These are unique to the sampler ID.
if (!id.hasAnyMips) {
constWidth256f_ = AlignCode16();
float w256f = (1 << id.width0Shift) * 256;
Write32(*(uint32_t *)&w256f); Write32(*(uint32_t *)&w256f);
Write32(*(uint32_t *)&w256f); Write32(*(uint32_t *)&w256f);
constHeight256f_ = AlignCode16();
float h256f = (1 << id.height0Shift) * 256;
Write32(*(uint32_t *)&h256f); Write32(*(uint32_t *)&h256f);
Write32(*(uint32_t *)&h256f); Write32(*(uint32_t *)&h256f);
constWidthMinus1i_ = AlignCode16();
Write32((1 << id.width0Shift) - 1); Write32((1 << id.width0Shift) - 1);
Write32((1 << id.width0Shift) - 1); Write32((1 << id.width0Shift) - 1);
constHeightMinus1i_ = AlignCode16();
Write32((1 << id.height0Shift) - 1); Write32((1 << id.height0Shift) - 1);
Write32((1 << id.height0Shift) - 1); Write32((1 << id.height0Shift) - 1);
} else {
constWidth256f_ = nullptr;
constHeight256f_ = nullptr;
constWidthMinus1i_ = nullptr;
constHeightMinus1i_ = nullptr;
}
}
RegCache::Reg SamplerJitCache::GetZeroVec() {
if (!regCache_.Has(RegCache::VEC_ZERO)) {
X64Reg r = regCache_.Alloc(RegCache::VEC_ZERO);