thirdparty/ppsspp
2
0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

softgpu: Use cached sampler state outside jit.

Browse source
This commit is contained in:
Unknown W. Brackets 2022-01-15 15:20:21 -08:00
parent a2abf9402b
commit a228b2ab6c
5 changed files with 78 additions and 76 deletions
Download patch file Download diff file Expand all files Collapse all files

109
GPU/Software/Sampler.cpp
View file

@ -39,9 +39,9 @@ extern u32 clut[4096];
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);
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);
static Vec4IntResult SOFTRAST_CALL SampleFetch(int u, int v, const u8 *tptr, int bufw, int level);
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 SampleFetch(int u, int v, const u8 *tptr, int bufw, int level, const SamplerID &samplerID);
std::mutex jitCacheLock;
SamplerJitCache *jitCache = nullptr;
@ -242,10 +242,9 @@ FetchFunc SamplerJitCache::GetFetch(const SamplerID &id) {
return nullptr;
}
template <unsigned int texel_size_bits>
static inline int GetPixelDataOffset(unsigned int row_pitch_pixels, unsigned int u, unsigned int v)
{
if (!gstate.isTextureSwizzled())
template <uint32_t texel_size_bits>
static inline int GetPixelDataOffset(uint32_t row_pitch_pixels, uint32_t u, uint32_t v, bool swizzled) {
if (!swizzled)
return (v * (row_pitch_pixels * texel_size_bits >> 3)) + (u * texel_size_bits >> 3);
const int tile_size_bits = 32;
@ -263,12 +262,10 @@ static inline int GetPixelDataOffset(unsigned int row_pitch_pixels, unsigned int
return tile_idx * (tile_size_bits / 8) + ((u % texels_per_tile) * texel_size_bits) / 8;
}
static inline u32 LookupColor(unsigned int index, unsigned int level)
{
const bool mipmapShareClut = gstate.isClutSharedForMipmaps();
const int clutSharingOffset = mipmapShareClut ? 0 : level * 16;
static inline u32 LookupColor(unsigned int index, unsigned int level, const SamplerID &samplerID) {
const int clutSharingOffset = samplerID.useSharedClut ? 0 : level * 16;
switch (gstate.getClutPaletteFormat()) {
switch (samplerID.ClutFmt()) {
case GE_CMODE_16BIT_BGR5650:
return RGB565ToRGBA8888(reinterpret_cast<u16*>(clut)[index + clutSharingOffset]);
@ -282,11 +279,15 @@ static inline u32 LookupColor(unsigned int index, unsigned int level)
return clut[index + clutSharingOffset];
default:
ERROR_LOG_REPORT(G3D, "Software: Unsupported palette format: %x", gstate.getClutPaletteFormat());
ERROR_LOG_REPORT(G3D, "Software: Unsupported palette format: %x", samplerID.ClutFmt());
return 0;
}
}
uint32_t TransformClutIndex(uint32_t index, const SamplerID &samplerID) {
return gstate.transformClutIndex(index);
}
struct Nearest4 {
alignas(16) u32 v[4];
@ -296,76 +297,74 @@ 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) {
inline static Nearest4 SOFTRAST_CALL SampleNearest(const int u[N], const int v[N], const u8 *srcptr, int texbufw, int level, const SamplerID &samplerID) {
Nearest4 res;
if (!srcptr) {
memset(res.v, 0, sizeof(res.v));
return res;
}
GETextureFormat texfmt = gstate.getTextureFormat();
// TODO: Should probably check if textures are aligned properly...
switch (texfmt) {
switch (samplerID.TexFmt()) {
case GE_TFMT_4444:
for (int i = 0; i < N; ++i) {
const u8 *src = srcptr + GetPixelDataOffset<16>(texbufw, u[i], v[i]);
const u8 *src = srcptr + GetPixelDataOffset<16>(texbufw, u[i], v[i], samplerID.swizzle);
res.v[i] = RGBA4444ToRGBA8888(*(const u16 *)src);
}
return res;
case GE_TFMT_5551:
for (int i = 0; i < N; ++i) {
const u8 *src = srcptr + GetPixelDataOffset<16>(texbufw, u[i], v[i]);
const u8 *src = srcptr + GetPixelDataOffset<16>(texbufw, u[i], v[i], samplerID.swizzle);
res.v[i] = RGBA5551ToRGBA8888(*(const u16 *)src);
}
return res;
case GE_TFMT_5650:
for (int i = 0; i < N; ++i) {
const u8 *src = srcptr + GetPixelDataOffset<16>(texbufw, u[i], v[i]);
const u8 *src = srcptr + GetPixelDataOffset<16>(texbufw, u[i], v[i], samplerID.swizzle);
res.v[i] = RGB565ToRGBA8888(*(const u16 *)src);
}
return res;
case GE_TFMT_8888:
for (int i = 0; i < N; ++i) {
const u8 *src = srcptr + GetPixelDataOffset<32>(texbufw, u[i], v[i]);
const u8 *src = srcptr + GetPixelDataOffset<32>(texbufw, u[i], v[i], samplerID.swizzle);
res.v[i] = *(const u32 *)src;
}
return res;
case GE_TFMT_CLUT32:
for (int i = 0; i < N; ++i) {
const u8 *src = srcptr + GetPixelDataOffset<32>(texbufw, u[i], v[i]);
const u8 *src = srcptr + GetPixelDataOffset<32>(texbufw, u[i], v[i], samplerID.swizzle);
u32 val = src[0] + (src[1] << 8) + (src[2] << 16) + (src[3] << 24);
res.v[i] = LookupColor(gstate.transformClutIndex(val), 0);
res.v[i] = LookupColor(TransformClutIndex(val, samplerID), 0, samplerID);
}
return res;
case GE_TFMT_CLUT16:
for (int i = 0; i < N; ++i) {
const u8 *src = srcptr + GetPixelDataOffset<16>(texbufw, u[i], v[i]);
const u8 *src = srcptr + GetPixelDataOffset<16>(texbufw, u[i], v[i], samplerID.swizzle);
u16 val = src[0] + (src[1] << 8);
res.v[i] = LookupColor(gstate.transformClutIndex(val), 0);
res.v[i] = LookupColor(TransformClutIndex(val, samplerID), 0, samplerID);
}
return res;
case GE_TFMT_CLUT8:
for (int i = 0; i < N; ++i) {
const u8 *src = srcptr + GetPixelDataOffset<8>(texbufw, u[i], v[i]);
const u8 *src = srcptr + GetPixelDataOffset<8>(texbufw, u[i], v[i], samplerID.swizzle);
u8 val = *src;
res.v[i] = LookupColor(gstate.transformClutIndex(val), 0);
res.v[i] = LookupColor(TransformClutIndex(val, samplerID), 0, samplerID);
}
return res;
case GE_TFMT_CLUT4:
for (int i = 0; i < N; ++i) {
const u8 *src = srcptr + GetPixelDataOffset<4>(texbufw, u[i], v[i]);
const u8 *src = srcptr + GetPixelDataOffset<4>(texbufw, u[i], v[i], samplerID.swizzle);
u8 val = (u[i] & 1) ? (src[0] >> 4) : (src[0] & 0xF);
// Only CLUT4 uses separate mipmap palettes.
res.v[i] = LookupColor(gstate.transformClutIndex(val), level);
res.v[i] = LookupColor(TransformClutIndex(val, samplerID), level, samplerID);
}
return res;
@ -391,7 +390,7 @@ inline static Nearest4 SOFTRAST_CALL SampleNearest(const int u[N], const int v[N
return res;
default:
ERROR_LOG_REPORT(G3D, "Software: Unsupported texture format: %x", texfmt);
ERROR_LOG_REPORT(G3D, "Software: Unsupported texture format: %x", samplerID.TexFmt());
memset(res.v, 0, sizeof(res.v));
return res;
}
@ -410,8 +409,8 @@ static inline int WrapUV(int v, int height) {
}
template <int N>
static inline void ApplyTexelClamp(int out_u[N], int out_v[N], const int u[N], const int v[N], int width, int height) {
if (gstate.isTexCoordClampedS()) {
static inline void ApplyTexelClamp(int out_u[N], int out_v[N], const int u[N], const int v[N], int width, int height, const SamplerID &samplerID) {
if (samplerID.clampS) {
for (int i = 0; i < N; ++i) {
out_u[i] = ClampUV(u[i], width);
}
@ -420,7 +419,7 @@ static inline void ApplyTexelClamp(int out_u[N], int out_v[N], const int u[N], c
out_u[i] = WrapUV(u[i], width);
}
}
if (gstate.isTexCoordClampedT()) {
if (samplerID.clampT) {
for (int i = 0; i < N; ++i) {
out_v[i] = ClampUV(v[i], height);
}
@ -431,7 +430,7 @@ static inline void ApplyTexelClamp(int out_u[N], int out_v[N], const int u[N], c
}
}
static inline void GetTexelCoordinates(int level, float s, float t, int &out_u, int &out_v, int x, int y) {
static inline void GetTexelCoordinates(int level, float s, float t, int &out_u, int &out_v, int x, int y, const SamplerID &samplerID) {
int width = gstate.getTextureWidth(level);
int height = gstate.getTextureHeight(level);
@ -441,28 +440,28 @@ static inline void GetTexelCoordinates(int level, float s, float t, int &out_u,
base_u >>= 8;
base_v >>= 8;
ApplyTexelClamp<1>(&out_u, &out_v, &base_u, &base_v, width, height);
ApplyTexelClamp<1>(&out_u, &out_v, &base_u, &base_v, width, height, samplerID);
}
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) {
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) {
int u, v;
// Nearest filtering only. Round texcoords.
GetTexelCoordinates(level, s, t, u, v, x, y);
Vec4<int> c0 = Vec4<int>::FromRGBA(SampleNearest<1>(&u, &v, tptr[0], bufw[0], level).v[0]);
GetTexelCoordinates(level, s, t, u, v, x, y, samplerID);
Vec4<int> c0 = Vec4<int>::FromRGBA(SampleNearest<1>(&u, &v, tptr[0], bufw[0], level, samplerID).v[0]);
if (levelFrac) {
GetTexelCoordinates(level + 1, s, t, u, v, x, y);
Vec4<int> c1 = Vec4<int>::FromRGBA(SampleNearest<1>(&u, &v, tptr[1], bufw[1], level + 1).v[0]);
GetTexelCoordinates(level + 1, s, t, u, v, x, y, samplerID);
Vec4<int> c1 = Vec4<int>::FromRGBA(SampleNearest<1>(&u, &v, tptr[1], bufw[1], level + 1, samplerID).v[0]);
c0 = (c1 * levelFrac + c0 * (16 - levelFrac)) / 16;
}
return GetTextureFunctionOutput(prim_color, ToVec4IntArg(c0));
return GetTextureFunctionOutput(prim_color, ToVec4IntArg(c0), samplerID);
}
static Vec4IntResult SOFTRAST_CALL SampleFetch(int u, int v, const u8 *tptr, int bufw, int level) {
Nearest4 c = SampleNearest<1>(&u, &v, tptr, bufw, level);
static Vec4IntResult SOFTRAST_CALL SampleFetch(int u, int v, const u8 *tptr, int bufw, int level, const SamplerID &samplerID) {
Nearest4 c = SampleNearest<1>(&u, &v, tptr, bufw, level, samplerID);
return ToVec4IntResult(Vec4<int>::FromRGBA(c.v[0]));
}
@ -518,7 +517,7 @@ static inline Vec4IntResult SOFTRAST_CALL ApplyTexelClampQuadT(bool clamp, int v
#endif
}
static inline Vec4IntResult SOFTRAST_CALL GetTexelCoordinatesQuadS(int level, float in_s, int &frac_u, int x) {
static inline Vec4IntResult SOFTRAST_CALL GetTexelCoordinatesQuadS(int level, float in_s, int &frac_u, int x, const SamplerID &samplerID) {
int width = gstate.getTextureWidth(level);
int base_u = (int)(in_s * width * 256) + 12 - x - 128;
@ -526,10 +525,10 @@ static inline Vec4IntResult SOFTRAST_CALL GetTexelCoordinatesQuadS(int level, fl
base_u >>= 8;
// Need to generate and individually wrap/clamp the four sample coordinates. Ugh.
return ApplyTexelClampQuadS(gstate.isTexCoordClampedS(), base_u, width);
return ApplyTexelClampQuadS(samplerID.clampS, base_u, width);
}
static inline Vec4IntResult SOFTRAST_CALL GetTexelCoordinatesQuadT(int level, float in_t, int &frac_v, int y) {
static inline Vec4IntResult SOFTRAST_CALL GetTexelCoordinatesQuadT(int level, float in_t, int &frac_v, int y, const SamplerID &samplerID) {
int height = gstate.getTextureHeight(level);
int base_v = (int)(in_t * height * 256) + 12 - y - 128;
@ -537,14 +536,14 @@ static inline Vec4IntResult SOFTRAST_CALL GetTexelCoordinatesQuadT(int level, fl
base_v >>= 8;
// Need to generate and individually wrap/clamp the four sample coordinates. Ugh.
return ApplyTexelClampQuadT(gstate.isTexCoordClampedT(), base_v, height);
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) {
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) {
int frac_u, frac_v;
const Vec4<int> u = GetTexelCoordinatesQuadS(texlevel, s, frac_u, x);
const Vec4<int> v = GetTexelCoordinatesQuadT(texlevel, t, frac_v, y);
Nearest4 c = SampleNearest<4>(u.AsArray(), v.AsArray(), tptr[0], bufw[0], texlevel);
const Vec4<int> u = GetTexelCoordinatesQuadS(texlevel, s, frac_u, x, samplerID);
const Vec4<int> v = GetTexelCoordinatesQuadT(texlevel, t, frac_v, y, samplerID);
Nearest4 c = SampleNearest<4>(u.AsArray(), v.AsArray(), tptr[0], bufw[0], texlevel, samplerID);
Vec4<int> texcolor_tl = Vec4<int>::FromRGBA(c.v[0]);
Vec4<int> texcolor_tr = Vec4<int>::FromRGBA(c.v[1]);
@ -555,13 +554,13 @@ 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) {
Vec4<int> c0 = SampleLinearLevel(s, t, x, y, tptr, bufw, texlevel);
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) {
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);
const Vec4<int> c1 = SampleLinearLevel(s, t, x, y, tptr + 1, bufw + 1, texlevel + 1, samplerID);
c0 = (c1 * levelFrac + c0 * (16 - levelFrac)) / 16;
}
return GetTextureFunctionOutput(prim_color, ToVec4IntArg(c0));
return GetTextureFunctionOutput(prim_color, ToVec4IntArg(c0), samplerID);
}
};