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

Correct uv bound pointers in arm vertexjit.

Browse source
This commit is contained in:
Unknown W. Brackets 2015-09-13 14:52:10 -07:00
parent 93e9d6cd56
commit dff4aeb30f
11 changed files with 66 additions and 62 deletions
Download patch file Download diff file Expand all files Collapse all files

10
GPU/Common/VertexDecoderArm.cpp
View file

@ -572,7 +572,7 @@ void VertexDecoderJitCache::Jit_TcU16Through() {
LDRH(tempReg2, srcReg, dec_->tcoff + 2); LDRH(tempReg2, srcReg, dec_->tcoff + 2);
// TODO: Cleanup. // TODO: Cleanup.
MOVP2R(scratchReg, &gstate_c.vertMinU); MOVP2R(scratchReg, &gstate_c.vertBounds.minU);
auto updateSide = [&](ARMReg r, CCFlags cc, u32 off) { auto updateSide = [&](ARMReg r, CCFlags cc, u32 off) {
LDRH(tempReg3, scratchReg, off); LDRH(tempReg3, scratchReg, off);
@ -583,10 +583,10 @@ void VertexDecoderJitCache::Jit_TcU16Through() {
}; };
// TODO: Can this actually be fast? Hmm, floats aren't better. // TODO: Can this actually be fast? Hmm, floats aren't better.
updateSide(tempReg1, CC_LT, 0); updateSide(tempReg1, CC_LT, offsetof(KnownVertexBounds, minU));
updateSide(tempReg1, CC_GT, 2); updateSide(tempReg1, CC_GT, offsetof(KnownVertexBounds, maxU));
updateSide(tempReg2, CC_LT, 4); updateSide(tempReg2, CC_LT, offsetof(KnownVertexBounds, minV));
updateSide(tempReg2, CC_GT, 6); updateSide(tempReg2, CC_GT, offsetof(KnownVertexBounds, maxV));
ORR(tempReg1, tempReg1, Operand2(tempReg2, ST_LSL, 16)); ORR(tempReg1, tempReg1, Operand2(tempReg2, ST_LSL, 16));
STR(tempReg1, dstReg, dec_->decFmt.uvoff); STR(tempReg1, dstReg, dec_->decFmt.uvoff);

10
GPU/Common/VertexDecoderArm64.cpp
View file

@ -563,7 +563,7 @@ void VertexDecoderJitCache::Jit_TcU16Through() {
LDRH(INDEX_UNSIGNED, tempReg2, srcReg, dec_->tcoff + 2); LDRH(INDEX_UNSIGNED, tempReg2, srcReg, dec_->tcoff + 2);
// TODO: Cleanup. // TODO: Cleanup.
MOVP2R(scratchReg64, &gstate_c.vertMinU); MOVP2R(scratchReg64, &gstate_c.vertBounds.minU);
auto updateSide = [&](ARM64Reg r, CCFlags cc, u32 off) { auto updateSide = [&](ARM64Reg r, CCFlags cc, u32 off) {
LDRH(INDEX_UNSIGNED, tempReg3, scratchReg64, off); LDRH(INDEX_UNSIGNED, tempReg3, scratchReg64, off);
@ -574,10 +574,10 @@ void VertexDecoderJitCache::Jit_TcU16Through() {
}; };
// TODO: Can this actually be fast? Hmm, floats aren't better. // TODO: Can this actually be fast? Hmm, floats aren't better.
updateSide(tempReg1, CC_LT, 0); updateSide(tempReg1, CC_LT, offsetof(KnownVertexBounds, minU));
updateSide(tempReg1, CC_GT, 2); updateSide(tempReg1, CC_GT, offsetof(KnownVertexBounds, maxU));
updateSide(tempReg2, CC_LT, 4); updateSide(tempReg2, CC_LT, offsetof(KnownVertexBounds, minV));
updateSide(tempReg2, CC_GT, 6); updateSide(tempReg2, CC_GT, offsetof(KnownVertexBounds, maxV));
ORR(tempReg1, tempReg1, tempReg2, ArithOption(tempReg2, ST_LSL, 16)); ORR(tempReg1, tempReg1, tempReg2, ArithOption(tempReg2, ST_LSL, 16));
STR(INDEX_UNSIGNED, tempReg1, dstReg, dec_->decFmt.uvoff); STR(INDEX_UNSIGNED, tempReg1, dstReg, dec_->decFmt.uvoff);

24
GPU/Common/VertexDecoderCommon.cpp
View file

@ -296,10 +296,10 @@ void VertexDecoder::Step_TcU16Through() const
uv[0] = uvdata[0]; uv[0] = uvdata[0];
uv[1] = uvdata[1]; uv[1] = uvdata[1];
gstate_c.vertMinU = std::min(gstate_c.vertMinU, uvdata[0]); gstate_c.vertBounds.minU = std::min(gstate_c.vertBounds.minU, uvdata[0]);
gstate_c.vertMaxU = std::max(gstate_c.vertMaxU, uvdata[0]); gstate_c.vertBounds.maxU = std::max(gstate_c.vertBounds.maxU, uvdata[0]);
gstate_c.vertMinV = std::min(gstate_c.vertMinV, uvdata[1]); gstate_c.vertBounds.minV = std::min(gstate_c.vertBounds.minV, uvdata[1]);
gstate_c.vertMaxV = std::max(gstate_c.vertMaxV, uvdata[1]); gstate_c.vertBounds.maxV = std::max(gstate_c.vertBounds.maxV, uvdata[1]);
} }
void VertexDecoder::Step_TcU16ThroughDouble() const void VertexDecoder::Step_TcU16ThroughDouble() const
@ -325,10 +325,10 @@ void VertexDecoder::Step_TcU16ThroughToFloat() const
uv[0] = uvdata[0]; uv[0] = uvdata[0];
uv[1] = uvdata[1]; uv[1] = uvdata[1];
gstate_c.vertMinU = std::min(gstate_c.vertMinU, uvdata[0]); gstate_c.vertBounds.minU = std::min(gstate_c.vertBounds.minU, uvdata[0]);
gstate_c.vertMaxU = std::max(gstate_c.vertMaxU, uvdata[0]); gstate_c.vertBounds.maxU = std::max(gstate_c.vertBounds.maxU, uvdata[0]);
gstate_c.vertMinV = std::min(gstate_c.vertMinV, uvdata[1]); gstate_c.vertBounds.minV = std::min(gstate_c.vertBounds.minV, uvdata[1]);
gstate_c.vertMaxV = std::max(gstate_c.vertMaxV, uvdata[1]); gstate_c.vertBounds.maxV = std::max(gstate_c.vertBounds.maxV, uvdata[1]);
} }
void VertexDecoder::Step_TcU16ThroughDoubleToFloat() const void VertexDecoder::Step_TcU16ThroughDoubleToFloat() const
@ -354,10 +354,10 @@ void VertexDecoder::Step_TcFloatThrough() const
uv[0] = uvdata[0]; uv[0] = uvdata[0];
uv[1] = uvdata[1]; uv[1] = uvdata[1];
gstate_c.vertMinU = std::min(gstate_c.vertMinU, (u16)uvdata[0]); gstate_c.vertBounds.minU = std::min(gstate_c.vertBounds.minU, (u16)uvdata[0]);
gstate_c.vertMaxU = std::max(gstate_c.vertMaxU, (u16)uvdata[0]); gstate_c.vertBounds.maxU = std::max(gstate_c.vertBounds.maxU, (u16)uvdata[0]);
gstate_c.vertMinV = std::min(gstate_c.vertMinV, (u16)uvdata[1]); gstate_c.vertBounds.minV = std::min(gstate_c.vertBounds.minV, (u16)uvdata[1]);
gstate_c.vertMaxV = std::max(gstate_c.vertMaxV, (u16)uvdata[1]); gstate_c.vertBounds.maxV = std::max(gstate_c.vertBounds.maxV, (u16)uvdata[1]);
} }
void VertexDecoder::Step_TcU8Prescale() const { void VertexDecoder::Step_TcU8Prescale() const {

8
GPU/Common/VertexDecoderX86.cpp
View file

@ -721,10 +721,10 @@ void VertexDecoderJitCache::Jit_TcU16Through() {
}; };
// TODO: Can this actually be fast? Hmm, floats aren't better. // TODO: Can this actually be fast? Hmm, floats aren't better.
updateSide(tempReg1, CC_GE, &gstate_c.vertMinU); updateSide(tempReg1, CC_GE, &gstate_c.vertBounds.minU);
updateSide(tempReg1, CC_LE, &gstate_c.vertMaxU); updateSide(tempReg1, CC_LE, &gstate_c.vertBounds.maxU);
updateSide(tempReg2, CC_GE, &gstate_c.vertMinV); updateSide(tempReg2, CC_GE, &gstate_c.vertBounds.minV);
updateSide(tempReg2, CC_LE, &gstate_c.vertMaxV); updateSide(tempReg2, CC_LE, &gstate_c.vertBounds.maxV);
} }
void VertexDecoderJitCache::Jit_TcU16ThroughToFloat() { void VertexDecoderJitCache::Jit_TcU16ThroughToFloat() {

10
GPU/Directx9/FramebufferDX9.cpp
View file

@ -639,11 +639,11 @@ namespace DX9 {
// If max is not > min, we probably could not detect it. Skip. // If max is not > min, we probably could not detect it. Skip.
// See the vertex decoder, where this is updated. // See the vertex decoder, where this is updated.
if ((flags & BINDFBCOLOR_MAY_COPY_WITH_UV) != 0 && gstate_c.vertMaxU > gstate_c.vertMinU) { if ((flags & BINDFBCOLOR_MAY_COPY_WITH_UV) != 0 && gstate_c.vertBounds.maxU > gstate_c.vertBounds.minU) {
x = gstate_c.vertMinU; x = gstate_c.vertBounds.minU;
y = gstate_c.vertMinV; y = gstate_c.vertBounds.minV;
w = gstate_c.vertMaxU - x; w = gstate_c.vertBounds.maxU - x;
h = gstate_c.vertMaxV - y; h = gstate_c.vertBounds.maxV - y;
// If we bound a framebuffer, apply the byte offset as pixels to the copy too. // If we bound a framebuffer, apply the byte offset as pixels to the copy too.
if (flags & BINDFBCOLOR_APPLY_TEX_OFFSET) { if (flags & BINDFBCOLOR_APPLY_TEX_OFFSET) {

12
GPU/Directx9/TextureCacheDX9.cpp
View file

@ -926,7 +926,7 @@ void TextureCacheDX9::ApplyTexture() {
// Texture scale/offset and gen modes don't apply in through. // Texture scale/offset and gen modes don't apply in through.
// So we can optimize how much of the texture we look at. // So we can optimize how much of the texture we look at.
if (gstate.isModeThrough()) { if (gstate.isModeThrough()) {
nextTexture_->maxSeenV = std::max(nextTexture_->maxSeenV, gstate_c.vertMaxV); nextTexture_->maxSeenV = std::max(nextTexture_->maxSeenV, gstate_c.vertBounds.maxV);
} else { } else {
// Otherwise, we need to reset to ensure we use the whole thing. // Otherwise, we need to reset to ensure we use the whole thing.
// Can't tell how much is used. // Can't tell how much is used.
@ -987,17 +987,17 @@ void TextureCacheDX9::ApplyTextureFramebuffer(TexCacheEntry *entry, VirtualFrame
}; };
// If min is not < max, then we don't have values (wasn't set during decode.) // If min is not < max, then we don't have values (wasn't set during decode.)
if (gstate_c.vertMinV < gstate_c.vertMaxV) { if (gstate_c.vertBounds.minV < gstate_c.vertBounds.maxV) {
const float invWidth = 1.0f / (float)framebuffer->bufferWidth; const float invWidth = 1.0f / (float)framebuffer->bufferWidth;
const float invHeight = 1.0f / (float)framebuffer->bufferHeight; const float invHeight = 1.0f / (float)framebuffer->bufferHeight;
// Inverse of half = double. // Inverse of half = double.
const float invHalfWidth = invWidth * 2.0f; const float invHalfWidth = invWidth * 2.0f;
const float invHalfHeight = invHeight * 2.0f; const float invHalfHeight = invHeight * 2.0f;
const int u1 = gstate_c.vertMinU + gstate_c.curTextureXOffset; const int u1 = gstate_c.vertBounds.minU + gstate_c.curTextureXOffset;
const int v1 = gstate_c.vertMinV + gstate_c.curTextureYOffset; const int v1 = gstate_c.vertBounds.minV + gstate_c.curTextureYOffset;
const int u2 = gstate_c.vertMaxU + gstate_c.curTextureXOffset; const int u2 = gstate_c.vertBounds.maxU + gstate_c.curTextureXOffset;
const int v2 = gstate_c.vertMaxV + gstate_c.curTextureYOffset; const int v2 = gstate_c.vertBounds.maxV + gstate_c.curTextureYOffset;
const float left = u1 * invHalfWidth - 1.0f + xoff; const float left = u1 * invHalfWidth - 1.0f + xoff;
const float right = u2 * invHalfWidth - 1.0f + xoff; const float right = u2 * invHalfWidth - 1.0f + xoff;

10
GPU/Directx9/TransformPipelineDX9.cpp
View file

@ -695,7 +695,7 @@ void TransformDrawEngineDX9::DoFlush() {
vai->numVerts = indexGen.PureCount(); vai->numVerts = indexGen.PureCount();
} }
_dbg_assert_msg_(G3D, gstate_c.vertMinV >= gstate_c.vertMaxV, "Should not have checked UVs when caching."); _dbg_assert_msg_(G3D, gstate_c.vertBounds.minV >= gstate_c.vertBounds.maxV, "Should not have checked UVs when caching.");
void * pVb; void * pVb;
u32 size = dec_->GetDecVtxFmt().stride * indexGen.MaxIndex(); u32 size = dec_->GetDecVtxFmt().stride * indexGen.MaxIndex();
@ -888,10 +888,10 @@ rotateVBO:
framebufferManager_->SetColorUpdated(gstate_c.skipDrawReason); framebufferManager_->SetColorUpdated(gstate_c.skipDrawReason);
// Now seems as good a time as any to reset the min/max coords, which we may examine later. // Now seems as good a time as any to reset the min/max coords, which we may examine later.
gstate_c.vertMinU = 512; gstate_c.vertBounds.minU = 512;
gstate_c.vertMinV = 512; gstate_c.vertBounds.minV = 512;
gstate_c.vertMaxU = 0; gstate_c.vertBounds.maxU = 0;
gstate_c.vertMaxV = 0; gstate_c.vertBounds.maxV = 0;
host->GPUNotifyDraw(); host->GPUNotifyDraw();
} }

10
GPU/GLES/Framebuffer.cpp
View file

@ -874,11 +874,11 @@ void FramebufferManager::BindFramebufferColor(int stage, u32 fbRawAddress, Virtu
// If max is not > min, we probably could not detect it. Skip. // If max is not > min, we probably could not detect it. Skip.
// See the vertex decoder, where this is updated. // See the vertex decoder, where this is updated.
if ((flags & BINDFBCOLOR_MAY_COPY_WITH_UV) != 0 && gstate_c.vertMaxU > gstate_c.vertMinU) { if ((flags & BINDFBCOLOR_MAY_COPY_WITH_UV) != 0 && gstate_c.vertBounds.maxU > gstate_c.vertBounds.minU) {
x = gstate_c.vertMinU; x = gstate_c.vertBounds.minU;
y = gstate_c.vertMinV; y = gstate_c.vertBounds.minV;
w = gstate_c.vertMaxU - x; w = gstate_c.vertBounds.maxU - x;
h = gstate_c.vertMaxV - y; h = gstate_c.vertBounds.maxV - y;
// If we bound a framebuffer, apply the byte offset as pixels to the copy too. // If we bound a framebuffer, apply the byte offset as pixels to the copy too.
if (flags & BINDFBCOLOR_APPLY_TEX_OFFSET) { if (flags & BINDFBCOLOR_APPLY_TEX_OFFSET) {

12
GPU/GLES/TextureCache.cpp
View file

@ -996,7 +996,7 @@ void TextureCache::ApplyTexture() {
// Texture scale/offset and gen modes don't apply in through. // Texture scale/offset and gen modes don't apply in through.
// So we can optimize how much of the texture we look at. // So we can optimize how much of the texture we look at.
if (gstate.isModeThrough()) { if (gstate.isModeThrough()) {
nextTexture_->maxSeenV = std::max(nextTexture_->maxSeenV, gstate_c.vertMaxV); nextTexture_->maxSeenV = std::max(nextTexture_->maxSeenV, gstate_c.vertBounds.maxV);
} else { } else {
// Otherwise, we need to reset to ensure we use the whole thing. // Otherwise, we need to reset to ensure we use the whole thing.
// Can't tell how much is used. // Can't tell how much is used.
@ -1053,17 +1053,17 @@ void TextureCache::ApplyTextureFramebuffer(TexCacheEntry *entry, VirtualFramebuf
static const GLubyte indices[4] = { 0, 1, 3, 2 }; static const GLubyte indices[4] = { 0, 1, 3, 2 };
// If min is not < max, then we don't have values (wasn't set during decode.) // If min is not < max, then we don't have values (wasn't set during decode.)
if (gstate_c.vertMinV < gstate_c.vertMaxV) { if (gstate_c.vertBounds.minV < gstate_c.vertBounds.maxV) {
const float invWidth = 1.0f / (float)framebuffer->bufferWidth; const float invWidth = 1.0f / (float)framebuffer->bufferWidth;
const float invHeight = 1.0f / (float)framebuffer->bufferHeight; const float invHeight = 1.0f / (float)framebuffer->bufferHeight;
// Inverse of half = double. // Inverse of half = double.
const float invHalfWidth = invWidth * 2.0f; const float invHalfWidth = invWidth * 2.0f;
const float invHalfHeight = invHeight * 2.0f; const float invHalfHeight = invHeight * 2.0f;
const int u1 = gstate_c.vertMinU + gstate_c.curTextureXOffset; const int u1 = gstate_c.vertBounds.minU + gstate_c.curTextureXOffset;
const int v1 = gstate_c.vertMinV + gstate_c.curTextureYOffset; const int v1 = gstate_c.vertBounds.minV + gstate_c.curTextureYOffset;
const int u2 = gstate_c.vertMaxU + gstate_c.curTextureXOffset; const int u2 = gstate_c.vertBounds.maxU + gstate_c.curTextureXOffset;
const int v2 = gstate_c.vertMaxV + gstate_c.curTextureYOffset; const int v2 = gstate_c.vertBounds.maxV + gstate_c.curTextureYOffset;
const float left = u1 * invHalfWidth - 1.0f; const float left = u1 * invHalfWidth - 1.0f;
const float right = u2 * invHalfWidth - 1.0f; const float right = u2 * invHalfWidth - 1.0f;

10
GPU/GLES/TransformPipeline.cpp
View file

@ -684,7 +684,7 @@ void TransformDrawEngine::DoFlush() {
vai->numVerts = indexGen.PureCount(); vai->numVerts = indexGen.PureCount();
} }
_dbg_assert_msg_(G3D, gstate_c.vertMinV >= gstate_c.vertMaxV, "Should not have checked UVs when caching."); _dbg_assert_msg_(G3D, gstate_c.vertBounds.minV >= gstate_c.vertBounds.maxV, "Should not have checked UVs when caching.");
vai->vbo = AllocateBuffer(); vai->vbo = AllocateBuffer();
glstate.arrayBuffer.bind(vai->vbo); glstate.arrayBuffer.bind(vai->vbo);
@ -885,10 +885,10 @@ rotateVBO:
framebufferManager_->SetColorUpdated(gstate_c.skipDrawReason); framebufferManager_->SetColorUpdated(gstate_c.skipDrawReason);
// Now seems as good a time as any to reset the min/max coords, which we may examine later. // Now seems as good a time as any to reset the min/max coords, which we may examine later.
gstate_c.vertMinU = 512; gstate_c.vertBounds.minU = 512;
gstate_c.vertMinV = 512; gstate_c.vertBounds.minV = 512;
gstate_c.vertMaxU = 0; gstate_c.vertBounds.maxU = 0;
gstate_c.vertMaxV = 0; gstate_c.vertBounds.maxV = 0;
#ifndef MOBILE_DEVICE #ifndef MOBILE_DEVICE
host->GPUNotifyDraw(); host->GPUNotifyDraw();

12
GPU/GPUState.h
View file

@ -464,6 +464,13 @@ enum {
GPU_PREFER_REVERSE_COLOR_ORDER = FLAG_BIT(31), GPU_PREFER_REVERSE_COLOR_ORDER = FLAG_BIT(31),
}; };
struct KnownVertexBounds {
u16 minU;
u16 minV;
u16 maxU;
u16 maxV;
};
struct GPUStateCache { struct GPUStateCache {
bool Supports(int flag) { return (featureFlags & flag) != 0; } bool Supports(int flag) { return (featureFlags & flag) != 0; }
@ -507,10 +514,7 @@ struct GPUStateCache {
float vpWidthScale; float vpWidthScale;
float vpHeightScale; float vpHeightScale;
u16 vertMinU; KnownVertexBounds vertBounds;
u16 vertMinV;
u16 vertMaxU;
u16 vertMaxV;
// TODO: These should be accessed from the current VFB object directly. // TODO: These should be accessed from the current VFB object directly.
u32 curRTWidth; u32 curRTWidth;