preserve-emulation/ppsspp
1
0
Fork 0
mirror of https://github.com/hrydgard/ppsspp.git synced 2026-09-03 11:15:20 +02:00
Code Issues Packages Projects Releases Wiki Activity

Merge pull request #16021 from unknownbrackets/softgpu-compact

Browse Source 
Reduce the size of state and vertex data in softgpu
This commit is contained in:
Henrik Rydgård
2022-09-13 09:45:00 +02:00
committed by GitHub
parent a0515bf5ca 8371091734
commit f18b0277f4
15 changed files with 53 additions and 77 deletions
Download Patch File Download Diff File Expand all files Collapse all files
Common/Data/Collections/TinySet.h
+2 -2
View File
@@ -47,12 +47,12 @@ struct TinySet {
size_t otherSize = other.size();
if (size() + otherSize <= MaxFastSize) {
// Fast case
for (int i = 0; i < otherSize; i++) {
for (size_t i = 0; i < otherSize; i++) {
fastLookup_[fastCount + i] = other.fastLookup_[i];
}
fastCount += other.fastCount;
} else {
for (int i = 0; i < otherSize; i++) {
for (size_t i = 0; i < otherSize; i++) {
push_back(other[i]);
}
}
Common/GPU/OpenGL/thin3d_gl.cpp
+1 -1
View File
@@ -1219,7 +1219,7 @@ bool OpenGLPipeline::LinkShaders() {
}
std::vector<GLRProgram::Initializer> initialize;
for (int i = 0; i < MAX_TEXTURE_SLOTS; ++i) {
if (i < samplers_.size()) {
if (i < (int)samplers_.size()) {
initialize.push_back({ &samplerLocs_[i], 0, i });
} else {
samplerLocs_[i] = -1;
Common/GPU/Vulkan/thin3d_vulkan.cpp
+1 -1
View File
@@ -1086,7 +1086,7 @@ Pipeline *VKContext::CreateGraphicsPipeline(const PipelineDesc &desc, const char
_dbg_assert_((int)input->attributes.size() == (int)input->visc.vertexAttributeDescriptionCount);
gDesc.ibd = input->bindings[0];
for (int i = 0; i < input->attributes.size(); i++) {
for (size_t i = 0; i < input->attributes.size(); i++) {
gDesc.attrs[i] = input->attributes[i];
}
gDesc.vis.vertexAttributeDescriptionCount = input->visc.vertexAttributeDescriptionCount;
GPU/Math3D.h
+3 -24
View File
@@ -57,22 +57,10 @@ inline static T VecClamp(const T &v, const T &low, const T &high)
}
template<typename T>
class Vec2
{
class Vec2 {
public:
union
{
struct
{
T x,y;
};
#if defined(_M_SSE)
__m128i ivec;
__m128 vec;
#elif PPSSPP_ARCH(ARM64_NEON)
int32x4_t ivec;
float32x4_t vec;
#endif
struct {
T x,y;
};
T* AsArray() { return &x; }
@@ -81,15 +69,6 @@ public:
Vec2() {}
Vec2(const T a[2]) : x(a[0]), y(a[1]) {}
Vec2(const T& _x, const T& _y) : x(_x), y(_y) {}
#if defined(_M_SSE)
Vec2(const __m128 &_vec) : vec(_vec) {}
Vec2(const __m128i &_ivec) : ivec(_ivec) {}
#elif PPSSPP_ARCH(ARM64_NEON)
Vec2(const float32x4_t &_vec) : vec(_vec) {}
#if !defined(_MSC_VER)
Vec2(const int32x4_t &_ivec) : ivec(_ivec) {}
#endif
#endif
template<typename T2>
Vec2<T2> Cast() const
GPU/Software/BinManager.cpp
+1 -1
View File
@@ -242,7 +242,7 @@ bool BinManager::HasTextureWrite(const RasterizerState &state) {
if (!state.enableTextures)
return false;
const int textureBits = textureBitsPerPixel[state.samplerID.texfmt];
const uint8_t textureBits = textureBitsPerPixel[state.samplerID.texfmt];
for (int i = 0; i <= state.maxTexLevel; ++i) {
int byteStride = (state.texbufw[i] * textureBits) / 8;
int byteWidth = (state.samplerID.cached.sizes[i].w * textureBits) / 8;
GPU/Software/Rasterizer.cpp
+3 -14
View File
@@ -78,14 +78,6 @@ static inline Vec3<int> Interpolate(const Vec3<int> &c0, const Vec3<int> &c1, co
#endif
}
static inline Vec2<float> Interpolate(const Vec2<float> &c0, const Vec2<float> &c1, const Vec2<float> &c2, int w0, int w1, int w2, float wsum) {
#if defined(_M_SSE) && !PPSSPP_ARCH(X86)
return Vec2<float>(Interpolate(c0.vec, c1.vec, c2.vec, w0, w1, w2, wsum));
#else
return (c0 * w0 + c1 * w1 + c2 * w2) * wsum;
#endif
}
static inline Vec4<float> Interpolate(const float &c0, const float &c1, const float &c2, const Vec4<float> &w0, const Vec4<float> &w1, const Vec4<float> &w2, const Vec4<float> &wsum_recip) {
#if defined(_M_SSE) && !PPSSPP_ARCH(X86)
__m128 v = _mm_mul_ps(w0.vec, _mm_set1_ps(c0));
@@ -124,7 +116,7 @@ void ComputeRasterizerState(RasterizerState *state, bool throughMode) {
for (uint8_t i = 0; i <= state->maxTexLevel; i++) {
u32 texaddr = gstate.getTextureAddress(i);
state->texaddr[i] = texaddr;
state->texbufw[i] = GetTextureBufw(i, texaddr, texfmt);
state->texbufw[i] = (uint16_t)GetTextureBufw(i, texaddr, texfmt);
if (Memory::IsValidAddress(texaddr))
state->texptr[i] = Memory::GetPointerUnchecked(texaddr);
else
@@ -143,9 +135,6 @@ void ComputeRasterizerState(RasterizerState *state, bool throughMode) {
state->throughMode = throughMode;
state->antialiasLines = gstate.isAntiAliasEnabled();
state->screenOffsetX = gstate.getOffsetX16();
state->screenOffsetY = gstate.getOffsetY16();
#if defined(SOFTGPU_MEMORY_TAGGING_DETAILED) || defined(SOFTGPU_MEMORY_TAGGING_BASIC)
DisplayList currentList{};
if (gpuDebug)
@@ -421,7 +410,7 @@ Vec3<int> AlphaBlendingResult(const PixelFuncID &pixelID, const Vec4<int> &sourc
static inline Vec4IntResult SOFTRAST_CALL ApplyTexturing(float s, float t, int x, int y, Vec4IntArg prim_color, int texlevel, int frac_texlevel, bool bilinear, const RasterizerState &state) {
const u8 **tptr0 = const_cast<const u8 **>(&state.texptr[texlevel]);
const int *bufw0 = &state.texbufw[texlevel];
const uint16_t *bufw0 = &state.texbufw[texlevel];
if (!bilinear) {
return state.nearest(s, t, x, y, prim_color, tptr0, bufw0, texlevel, frac_texlevel, state.samplerID);
@@ -1476,7 +1465,7 @@ bool GetCurrentTexture(GPUDebugBuffer &buffer, int level)
GETextureFormat texfmt = gstate.getTextureFormat();
u32 texaddr = gstate.getTextureAddress(level);
int texbufw = GetTextureBufw(level, texaddr, texfmt);
u32 texbufw = GetTextureBufw(level, texaddr, texfmt);
int w = gstate.getTextureWidth(level);
int h = gstate.getTextureHeight(level);
GPU/Software/Rasterizer.h
+1 -3
View File
@@ -39,11 +39,9 @@ struct RasterizerState {
Sampler::LinearFunc linear;
Sampler::NearestFunc nearest;
uint32_t texaddr[8]{};
int texbufw[8]{};
uint16_t texbufw[8]{};
const u8 *texptr[8]{};
float textureLodSlope;
int screenOffsetX;
int screenOffsetY;
struct {
uint8_t maxTexLevel : 3;
GPU/Software/RasterizerRectangle.cpp
+3 -3
View File
@@ -103,7 +103,7 @@ void DrawSprite(const VertexData &v0, const VertexData &v1, const BinCoords &ran
const u8 *texptr = state.texptr[0];
GETextureFormat texfmt = state.samplerID.TexFmt();
int texbufw = state.texbufw[0];
uint16_t texbufw = state.texbufw[0];
Sampler::FetchFunc fetchFunc = Sampler::GetFetchFunc(state.samplerID);
auto &pixelID = state.pixelID;
@@ -300,9 +300,9 @@ bool RectangleFastPath(const VertexData &v0, const VertexData &v1, BinManager &b
if (PSP_CoreParameter().compat.flags().DarkStalkersPresentHack && v0.texturecoords.x == 64.0f && v0.texturecoords.y == 16.0f && v1.texturecoords.x == 448.0f && v1.texturecoords.y == 240.0f) {
// check for save/load dialog.
if (!currentDialogActive) {
if (v0.screenpos.x + state.screenOffsetX == 0x7100 && v0.screenpos.y + state.screenOffsetY == 0x7780 && v1.screenpos.x + state.screenOffsetX == 0x8f00 && v1.screenpos.y + state.screenOffsetY == 0x8880) {
if (v0.screenpos.x + gstate.getOffsetX16() == 0x7100 && v0.screenpos.y + gstate.getOffsetY16() == 0x7780 && v1.screenpos.x + gstate.getOffsetX16() == 0x8f00 && v1.screenpos.y + gstate.getOffsetY16() == 0x8880) {
g_DarkStalkerStretch = DSStretch::Wide;
} else if (v0.screenpos.x + state.screenOffsetX == 0x7400 && v0.screenpos.y + state.screenOffsetY == 0x7780 && v1.screenpos.x + state.screenOffsetX == 0x8C00 && v1.screenpos.y + state.screenOffsetY == 0x8880) {
} else if (v0.screenpos.x + gstate.getOffsetX16() == 0x7400 && v0.screenpos.y + gstate.getOffsetY16() == 0x7780 && v1.screenpos.x + gstate.getOffsetX16() == 0x8C00 && v1.screenpos.y + gstate.getOffsetY16() == 0x8880) {
g_DarkStalkerStretch = DSStretch::Normal;
} else {
return false;
GPU/Software/Sampler.cpp
+6 -6
View File
@@ -38,8 +38,8 @@ using namespace Rasterizer;
namespace Sampler {
static Vec4IntResult SOFTRAST_CALL SampleNearest(float s, float t, int x, int y, Vec4IntArg prim_color, const u8 *const *tptr, const int *bufw, int level, int levelFrac, const SamplerID &samplerID);
static Vec4IntResult SOFTRAST_CALL SampleLinear(float s, float t, int x, int y, Vec4IntArg prim_color, const u8 *const *tptr, const int *bufw, int level, int levelFrac, const SamplerID &samplerID);
static Vec4IntResult SOFTRAST_CALL SampleNearest(float s, float t, int x, int y, Vec4IntArg prim_color, const u8 *const *tptr, const uint16_t *bufw, int level, int levelFrac, const SamplerID &samplerID);
static Vec4IntResult SOFTRAST_CALL SampleLinear(float s, float t, int x, int y, Vec4IntArg prim_color, const u8 *const *tptr, const uint16_t *bufw, int level, int levelFrac, const SamplerID &samplerID);
static Vec4IntResult SOFTRAST_CALL SampleFetch(int u, int v, const u8 *tptr, int bufw, int level, const SamplerID &samplerID);
std::mutex jitCacheLock;
@@ -281,7 +281,7 @@ struct Nearest4 {
};
template <int N>
inline static Nearest4 SOFTRAST_CALL SampleNearest(const int u[N], const int v[N], const u8 *srcptr, int texbufw, int level, const SamplerID &samplerID) {
inline static Nearest4 SOFTRAST_CALL SampleNearest(const int u[N], const int v[N], const u8 *srcptr, uint16_t texbufw, int level, const SamplerID &samplerID) {
Nearest4 res;
if (!srcptr) {
memset(res.v, 0, sizeof(res.v));
@@ -535,7 +535,7 @@ Vec4IntResult SOFTRAST_CALL GetTextureFunctionOutput(Vec4IntArg prim_color_in, V
return ToVec4IntResult(Vec4<int>(out_rgb, out_a));
}
static Vec4IntResult SOFTRAST_CALL SampleNearest(float s, float t, int x, int y, Vec4IntArg prim_color, const u8 *const *tptr, const int *bufw, int level, int levelFrac, const SamplerID &samplerID) {
static Vec4IntResult SOFTRAST_CALL SampleNearest(float s, float t, int x, int y, Vec4IntArg prim_color, const u8 *const *tptr, const uint16_t *bufw, int level, int levelFrac, const SamplerID &samplerID) {
int u, v;
// Nearest filtering only. Round texcoords.
@@ -631,7 +631,7 @@ static inline Vec4IntResult SOFTRAST_CALL GetTexelCoordinatesQuadT(int level, fl
return ApplyTexelClampQuadT(samplerID.clampT, base_v, height);
}
static Vec4IntResult SOFTRAST_CALL SampleLinearLevel(float s, float t, int x, int y, const u8 *const *tptr, const int *bufw, int texlevel, const SamplerID &samplerID) {
static Vec4IntResult SOFTRAST_CALL SampleLinearLevel(float s, float t, int x, int y, const u8 *const *tptr, const uint16_t *bufw, int texlevel, const SamplerID &samplerID) {
int frac_u, frac_v;
const Vec4<int> u = GetTexelCoordinatesQuadS(texlevel, s, frac_u, x, samplerID);
const Vec4<int> v = GetTexelCoordinatesQuadT(texlevel, t, frac_v, y, samplerID);
@@ -646,7 +646,7 @@ static Vec4IntResult SOFTRAST_CALL SampleLinearLevel(float s, float t, int x, in
return ToVec4IntResult((top * (0x10 - frac_v) + bot * frac_v) / (16 * 16));
}
static Vec4IntResult SOFTRAST_CALL SampleLinear(float s, float t, int x, int y, Vec4IntArg prim_color, const u8 *const *tptr, const int *bufw, int texlevel, int levelFrac, const SamplerID &samplerID) {
static Vec4IntResult SOFTRAST_CALL SampleLinear(float s, float t, int x, int y, Vec4IntArg prim_color, const u8 *const *tptr, const uint16_t *bufw, int texlevel, int levelFrac, const SamplerID &samplerID) {
Vec4<int> c0 = SampleLinearLevel(s, t, x, y, tptr, bufw, texlevel, samplerID);
if (levelFrac) {
const Vec4<int> c1 = SampleLinearLevel(s, t, x, y, tptr + 1, bufw + 1, texlevel + 1, samplerID);
GPU/Software/Sampler.h
+2 -2
View File
@@ -36,10 +36,10 @@ namespace Sampler {
typedef Rasterizer::Vec4IntResult(SOFTRAST_CALL *FetchFunc)(int u, int v, const u8 *tptr, int bufw, int level, const SamplerID &samplerID);
FetchFunc GetFetchFunc(SamplerID id);
typedef Rasterizer::Vec4IntResult (SOFTRAST_CALL *NearestFunc)(float s, float t, int x, int y, Rasterizer::Vec4IntArg prim_color, const u8 *const *tptr, const int *bufw, int level, int levelFrac, const SamplerID &samplerID);
typedef Rasterizer::Vec4IntResult (SOFTRAST_CALL *NearestFunc)(float s, float t, int x, int y, Rasterizer::Vec4IntArg prim_color, const u8 *const *tptr, const uint16_t *bufw, int level, int levelFrac, const SamplerID &samplerID);
NearestFunc GetNearestFunc(SamplerID id);
typedef Rasterizer::Vec4IntResult (SOFTRAST_CALL *LinearFunc)(float s, float t, int x, int y, Rasterizer::Vec4IntArg prim_color, const u8 *const *tptr, const int *bufw, int level, int levelFrac, const SamplerID &samplerID);
typedef Rasterizer::Vec4IntResult (SOFTRAST_CALL *LinearFunc)(float s, float t, int x, int y, Rasterizer::Vec4IntArg prim_color, const u8 *const *tptr, const uint16_t *bufw, int level, int levelFrac, const SamplerID &samplerID);
LinearFunc GetLinearFunc(SamplerID id);
void Init();
GPU/Software/SamplerX86.cpp
+17 -14
View File
@@ -246,7 +246,7 @@ NearestFunc SamplerJitCache::CompileNearest(const SamplerID &id) {
auto loadPtrs = [&](bool level1) {
X64Reg bufwReg = regCache_.Alloc(RegCache::GEN_ARG_BUFW);
X64Reg bufwPtrReg = regCache_.Find(RegCache::GEN_ARG_BUFW_PTR);
MOV(32, R(bufwReg), MDisp(bufwPtrReg, level1 ? 4 : 0));
MOVZX(32, 16, bufwReg, MDisp(bufwPtrReg, level1 ? 2 : 0));
regCache_.Unlock(bufwPtrReg, RegCache::GEN_ARG_BUFW_PTR);
regCache_.Unlock(bufwReg, RegCache::GEN_ARG_BUFW);
regCache_.ForceRetain(RegCache::GEN_ARG_BUFW);
@@ -713,7 +713,7 @@ LinearFunc SamplerJitCache::CompileLinear(const SamplerID &id) {
X64Reg srcReg = regCache_.Find(RegCache::GEN_ARG_TEXPTR_PTR);
X64Reg bufwReg = regCache_.Find(RegCache::GEN_ARG_BUFW_PTR);
ADD(64, R(srcArgReg), MDisp(srcReg, level1 ? 8 : 0));
MOV(32, R(bufwArgReg), MDisp(bufwReg, level1 ? 4 : 0));
MOVZX(32, 16, bufwArgReg, MDisp(bufwReg, level1 ? 2 : 0));
// Leave level/levelFrac, we just always load from RAM on Windows and lock on POSIX.
regCache_.Unlock(srcReg, RegCache::GEN_ARG_TEXPTR_PTR);
regCache_.Unlock(bufwReg, RegCache::GEN_ARG_BUFW_PTR);
@@ -2995,12 +2995,13 @@ bool SamplerJitCache::Jit_PrepareDataDirectOffsets(const SamplerID &id, RegCache
if (!id.useStandardBufw || id.hasAnyMips) {
// Spread bufw into each lane.
X64Reg bufwReg = regCache_.Find(RegCache::GEN_ARG_BUFW_PTR);
if (cpu_info.bAVX2) {
VPBROADCASTD(128, bufwVecReg, MDisp(bufwReg, level1 ? 4 : 0));
if (cpu_info.bSSE4_1) {
PMOVZXWD(bufwVecReg, MDisp(bufwReg, level1 ? 2 : 0));
} else {
MOVD_xmm(bufwVecReg, MDisp(bufwReg, level1 ? 4 : 0));
PSHUFD(bufwVecReg, R(bufwVecReg), _MM_SHUFFLE(0, 0, 0, 0));
PXOR(bufwVecReg, R(bufwVecReg));
PINSRW(bufwVecReg, MDisp(bufwReg, level1 ? 2 : 0), 0);
}
PSHUFD(bufwVecReg, R(bufwVecReg), _MM_SHUFFLE(0, 0, 0, 0));
regCache_.Unlock(bufwReg, RegCache::GEN_ARG_BUFW_PTR);
if (bitsPerTexel == 4)
@@ -3070,12 +3071,13 @@ bool SamplerJitCache::Jit_PrepareDataSwizzledOffsets(const SamplerID &id, RegCac
if (!id.useStandardBufw || id.hasAnyMips) {
// Spread bufw into each lane.
X64Reg bufwReg = regCache_.Find(RegCache::GEN_ARG_BUFW_PTR);
if (cpu_info.bAVX2) {
VPBROADCASTD(128, bufwVecReg, MDisp(bufwReg, level1 ? 4 : 0));
if (cpu_info.bSSE4_1) {
PMOVZXWD(bufwVecReg, MDisp(bufwReg, level1 ? 2 : 0));
} else {
MOVD_xmm(bufwVecReg, MDisp(bufwReg, level1 ? 4 : 0));
PSHUFD(bufwVecReg, R(bufwVecReg), _MM_SHUFFLE(0, 0, 0, 0));
PXOR(bufwVecReg, R(bufwVecReg));
PINSRW(bufwVecReg, MDisp(bufwReg, level1 ? 2 : 0), 0);
}
PSHUFD(bufwVecReg, R(bufwVecReg), _MM_SHUFFLE(0, 0, 0, 0));
regCache_.Unlock(bufwReg, RegCache::GEN_ARG_BUFW_PTR);
}
@@ -3162,12 +3164,13 @@ bool SamplerJitCache::Jit_PrepareDataDXTOffsets(const SamplerID &id, Rasterizer:
if (!id.useStandardBufw || id.hasAnyMips) {
// Spread bufw into each lane.
X64Reg bufwReg = regCache_.Find(RegCache::GEN_ARG_BUFW_PTR);
if (cpu_info.bAVX2) {
VPBROADCASTD(128, bufwVecReg, MDisp(bufwReg, level1 ? 4 : 0));
if (cpu_info.bSSE4_1) {
PMOVZXWD(bufwVecReg, MDisp(bufwReg, level1 ? 2 : 0));
} else {
MOVD_xmm(bufwVecReg, MDisp(bufwReg, level1 ? 4 : 0));
PSHUFD(bufwVecReg, R(bufwVecReg), _MM_SHUFFLE(0, 0, 0, 0));
PXOR(bufwVecReg, R(bufwVecReg));
PINSRW(bufwVecReg, MDisp(bufwReg, level1 ? 2 : 0), 0);
}
PSHUFD(bufwVecReg, R(bufwVecReg), _MM_SHUFFLE(0, 0, 0, 0));
regCache_.Unlock(bufwReg, RegCache::GEN_ARG_BUFW_PTR);
// Divide by 4 before the multiply.
GPU/Software/TransformUnit.h
+9
View File
@@ -164,6 +164,15 @@ public:
TransformUnit transformUnit;
#if PPSSPP_ARCH(32BIT)
void *operator new(size_t s) {
return AllocateAlignedMemory(s, 16);
}
void operator delete(void *p) {
FreeAlignedMemory(p);
}
#endif
protected:
bool UpdateUseHWTessellation(bool enable) override { return false; }
};
GPU/ge_constants.h
+2 -4
View File
@@ -374,8 +374,7 @@ enum GEMatrixType {
GE_MTX_TEXGEN,
};
enum GEComparison
{
enum GEComparison : uint8_t {
GE_COMP_NEVER = 0,
GE_COMP_ALWAYS = 1,
GE_COMP_EQUAL = 2,
@@ -578,8 +577,7 @@ enum GEPrimitiveType
GE_PRIM_INVALID = -1,
};
enum GELogicOp
{
enum GELogicOp : uint8_t {
GE_LOGIC_CLEAR = 0,
GE_LOGIC_AND = 1,
GE_LOGIC_AND_REVERSE = 2,
UI/MiscScreens.cpp
+1 -1
View File
@@ -251,7 +251,7 @@ private:
return nullptr;
}
const auto recentIsos = g_Config.RecentIsos();
if (index >= recentIsos.size())
if (index >= (int)recentIsos.size())
return nullptr;
return g_gameInfoCache->GetInfo(dc.GetDrawContext(), Path(recentIsos[index]), GAMEINFO_WANTBG);
}
unittest/TestSoftwareGPUJit.cpp
+1 -1
View File
@@ -48,7 +48,7 @@ static bool TestSamplerJit() {
bool header = false;
u8 **tptr = new u8 *[8];
int *bufw = new int[8];
uint16_t *bufw = new uint16_t[8];
u8 *clut = new u8[1024];
memset(clut, 0, 1024);