Emulator-Archive/ppsspp
1
0
Fork 0
mirror of https://github.com/hrydgard/ppsspp.git synced 2025-02-18 21:27:52 +00:00
Code Issues Actions 2 Packages Projects Releases Wiki Activity

Merge pull request #18261 from hrydgard/revert-expand-lines-fix

Browse Source 
Revert "Merge pull request #18184 from hrydgard/expand-lines-mem-fix"
This commit is contained in:
Henrik Rydgård 2023-09-27 21:09:43 +02:00 committed by GitHub
commit fde10ed4bf
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
6 changed files with 41 additions and 92 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

86
GPU/Common/SoftwareTransformCommon.cpp
View File

@ -547,7 +547,7 @@ void SoftwareTransform::DetectOffsetTexture(int maxIndex) {
}
// NOTE: The viewport must be up to date!
void SoftwareTransform::BuildDrawingParams(int prim, int vertexCount, u32 vertType, u16 *inds, int &indsOffset, int indexBufferSize, int &maxIndex, SoftwareTransformResult *result) {
void SoftwareTransform::BuildDrawingParams(int prim, int vertexCount, u32 vertType, u16 *&inds, int &maxIndex, SoftwareTransformResult *result) {
TransformedVertex *transformed = params_.transformed;
TransformedVertex *transformedExpanded = params_.transformedExpanded;
bool throughmode = (vertType & GE_VTYPE_THROUGH_MASK) != 0;
@ -560,11 +560,7 @@ void SoftwareTransform::BuildDrawingParams(int prim, int vertexCount, u32 vertTy
bool useBufferedRendering = fbman->UseBufferedRendering();
if (prim == GE_PRIM_RECTANGLES) {
if (!ExpandRectangles(vertexCount, maxIndex, inds, indsOffset, indexBufferSize, transformed, transformedExpanded, numTrans, throughmode)) {
result->drawIndexed = false;
result->drawNumTrans = 0;
return;
}
ExpandRectangles(vertexCount, maxIndex, inds, transformed, transformedExpanded, numTrans, throughmode);
result->drawBuffer = transformedExpanded;
result->drawIndexed = true;
@ -582,19 +578,11 @@ void SoftwareTransform::BuildDrawingParams(int prim, int vertexCount, u32 vertTy
}
}
} else if (prim == GE_PRIM_POINTS) {
if (!ExpandPoints(vertexCount, maxIndex, inds, indsOffset, indexBufferSize, transformed, transformedExpanded, numTrans, throughmode)) {
result->drawIndexed = false;
result->drawNumTrans = 0;
return;
}
ExpandPoints(vertexCount, maxIndex, inds, transformed, transformedExpanded, numTrans, throughmode);
result->drawBuffer = transformedExpanded;
result->drawIndexed = true;
} else if (prim == GE_PRIM_LINES) {
if (!ExpandLines(vertexCount, maxIndex, inds, indsOffset, indexBufferSize, transformed, transformedExpanded, numTrans, throughmode)) {
result->drawIndexed = false;
result->drawNumTrans = 0;
return;
}
ExpandLines(vertexCount, maxIndex, inds, transformed, transformedExpanded, numTrans, throughmode);
result->drawBuffer = transformedExpanded;
result->drawIndexed = true;
} else {
@ -686,28 +674,15 @@ void SoftwareTransform::CalcCullParams(float &minZValue, float &maxZValue) {
std::swap(minZValue, maxZValue);
}
static void ReportExpandOverflow(const char *primName, int vertexCount, int indexSpaceLeft) {
ERROR_LOG_REPORT_ONCE(expandoverflow, G3D, "%s expansion overflow: vertexCount=%d (left: %d)", primName, vertexCount, indexSpaceLeft);
}
// indsOffset is in indices, so half what the byte offset would be.
// indexBufferSize is also in indices, not bytes.
bool SoftwareTransform::ExpandRectangles(int vertexCount, int &maxIndex, u16 *inds, int &indsOffset, int indexBufferSize, const TransformedVertex *transformed, TransformedVertex *transformedExpanded, int &numTrans, bool throughmode) {
// Before we start, do a sanity check - does the output fit?
if ((vertexCount / 2) * 6 > indexBufferSize - indsOffset) {
// Won't fit, report and kill the draw.
ReportExpandOverflow("rects", vertexCount, indexBufferSize - indsOffset);
return false;
}
void SoftwareTransform::ExpandRectangles(int vertexCount, int &maxIndex, u16 *&inds, const TransformedVertex *transformed, TransformedVertex *transformedExpanded, int &numTrans, bool throughmode) {
// Rectangles always need 2 vertices, disregard the last one if there's an odd number.
vertexCount = vertexCount & ~1;
numTrans = 0;
TransformedVertex *trans = &transformedExpanded[0];
const u16 *indsIn = inds + indsOffset;
int newIndsOffset = indsOffset + vertexCount;
u16 *indsOut = inds + newIndsOffset;
const u16 *indsIn = (const u16 *)inds;
u16 *newInds = inds + vertexCount;
u16 *indsOut = newInds;
maxIndex = 4 * (vertexCount / 2);
for (int i = 0; i < vertexCount; i += 2) {
@ -752,35 +727,23 @@ bool SoftwareTransform::ExpandRectangles(int vertexCount, int &maxIndex, u16 *in
indsOut[3] = i * 2 + 3;
indsOut[4] = i * 2 + 0;
indsOut[5] = i * 2 + 2;
trans += 4;
indsOut += 6;
numTrans += 6;
}
indsOffset = newIndsOffset;
return true;
inds = newInds;
}
// indsOffset is in indices, so half what the byte offset would be.
// indexBufferSize is also in indices, not bytes.
bool SoftwareTransform::ExpandLines(int vertexCount, int &maxIndex, u16 *inds, int &indsOffset, int indexBufferSize, const TransformedVertex *transformed, TransformedVertex *transformedExpanded, int &numTrans, bool throughmode) {
// Before we start, do a sanity check - does the output fit?
if ((vertexCount / 2) * 6 > indexBufferSize - indsOffset) {
// Won't fit, report and kill the draw.
ReportExpandOverflow("lines", vertexCount, indexBufferSize - indsOffset);
return false;
}
void SoftwareTransform::ExpandLines(int vertexCount, int &maxIndex, u16 *&inds, const TransformedVertex *transformed, TransformedVertex *transformedExpanded, int &numTrans, bool throughmode) {
// Lines always need 2 vertices, disregard the last one if there's an odd number.
vertexCount = vertexCount & ~1;
numTrans = 0;
TransformedVertex *trans = &transformedExpanded[0];
const u16 *indsIn = inds + indsOffset;
int newIndsOffset = indsOffset + vertexCount;
u16 *indsOut = inds + newIndsOffset;
const u16 *indsIn = (const u16 *)inds;
u16 *newInds = inds + vertexCount;
u16 *indsOut = newInds;
float dx = 1.0f * gstate_c.vpWidthScale * (1.0f / fabsf(gstate.getViewportXScale()));
float dy = 1.0f * gstate_c.vpHeightScale * (1.0f / fabsf(gstate.getViewportYScale()));
@ -893,26 +856,17 @@ bool SoftwareTransform::ExpandLines(int vertexCount, int &maxIndex, u16 *inds, i
}
}
indsOffset = newIndsOffset;
return true;
inds = newInds;
}
// indsOffset is in indices, so half what the byte offset would be.
// indexBufferSize is also in indices, not bytes.
bool SoftwareTransform::ExpandPoints(int vertexCount, int &maxIndex, u16 *inds, int &indsOffset, int indexBufferSize, const TransformedVertex *transformed, TransformedVertex *transformedExpanded, int &numTrans, bool throughmode) {
// Before we start, do a sanity check - does the output fit?
if (vertexCount * 6 > indexBufferSize - indsOffset) {
// Won't fit, report and kill the draw.
ReportExpandOverflow("points", vertexCount, indexBufferSize - indsOffset);
return false;
}
void SoftwareTransform::ExpandPoints(int vertexCount, int &maxIndex, u16 *&inds, const TransformedVertex *transformed, TransformedVertex *transformedExpanded, int &numTrans, bool throughmode) {
numTrans = 0;
TransformedVertex *trans = &transformedExpanded[0];
const u16 *indsIn = (const u16 *)(inds + indsOffset);
int newIndsOffset = indsOffset + vertexCount;
u16 *indsOut = inds + newIndsOffset;
const u16 *indsIn = (const u16 *)inds;
u16 *newInds = inds + vertexCount;
u16 *indsOut = newInds;
float dx = 1.0f * gstate_c.vpWidthScale * (1.0f / gstate.getViewportXScale());
float dy = 1.0f * gstate_c.vpHeightScale * (1.0f / gstate.getViewportYScale());
@ -971,7 +925,5 @@ bool SoftwareTransform::ExpandPoints(int vertexCount, int &maxIndex, u16 *inds,
numTrans += 6;
}
indsOffset = newIndsOffset;
return true;
inds = newInds;
}

11
GPU/Common/SoftwareTransformCommon.h
View File

@ -62,18 +62,19 @@ struct SoftwareTransformParams {
class SoftwareTransform {
public:
SoftwareTransform(SoftwareTransformParams &params) : params_(params) {}
SoftwareTransform(SoftwareTransformParams &params) : params_(params) {
}
void SetProjMatrix(const float mtx[14], bool invertedX, bool invertedY, const Lin::Vec3 &trans, const Lin::Vec3 &scale);
void Decode(int prim, u32 vertexType, const DecVtxFormat &decVtxFormat, int maxIndex, SoftwareTransformResult *result);
void DetectOffsetTexture(int maxIndex);
void BuildDrawingParams(int prim, int vertexCount, u32 vertType, u16 *inds, int &indsOffset, int indexBufferSize, int &maxIndex, SoftwareTransformResult *result);
void BuildDrawingParams(int prim, int vertexCount, u32 vertType, u16 *&inds, int &maxIndex, SoftwareTransformResult *result);
protected:
void CalcCullParams(float &minZValue, float &maxZValue);
bool ExpandRectangles(int vertexCount, int &maxIndex, u16 *inds, int &indsOffset, int indexBufferSize, const TransformedVertex *transformed, TransformedVertex *transformedExpanded, int &numTrans, bool throughmode);
bool ExpandLines(int vertexCount, int &maxIndex, u16 *inds, int &indsOffset, int indexBufferSize, const TransformedVertex *transformed, TransformedVertex *transformedExpanded, int &numTrans, bool throughmode);
bool ExpandPoints(int vertexCount, int &maxIndex, u16 *inds, int &indsOffset, int indexBufferSize, const TransformedVertex *transformed, TransformedVertex *transformedExpanded, int &numTrans, bool throughmode);
void ExpandRectangles(int vertexCount, int &maxIndex, u16 *&inds, const TransformedVertex *transformed, TransformedVertex *transformedExpanded, int &numTrans, bool throughmode);
void ExpandLines(int vertexCount, int &maxIndex, u16 *&inds, const TransformedVertex *transformed, TransformedVertex *transformedExpanded, int &numTrans, bool throughmode);
void ExpandPoints(int vertexCount, int &maxIndex, u16 *&inds, const TransformedVertex *transformed, TransformedVertex *transformedExpanded, int &numTrans, bool throughmode);
const SoftwareTransformParams &params_;
Lin::Matrix4x4 projMatrix_;

9
GPU/D3D11/DrawEngineD3D11.cpp
View File

@ -598,7 +598,7 @@ rotateVBO:
prim = GE_PRIM_TRIANGLES;
VERBOSE_LOG(G3D, "Flush prim %i SW! %i verts in one go", prim, indexGen.VertexCount());
u16 *const inds = decIndex_;
u16 *inds = decIndex_;
SoftwareTransformResult result{};
SoftwareTransformParams params{};
params.decoded = decoded_;
@ -644,9 +644,8 @@ rotateVBO:
// Need to ApplyDrawState after ApplyTexture because depal can launch a render pass and that wrecks the state.
ApplyDrawState(prim);
int indsOffset = 0;
if (result.action == SW_NOT_READY)
swTransform.BuildDrawingParams(prim, indexGen.VertexCount(), dec_->VertexType(), inds, indsOffset, DECODED_INDEX_BUFFER_SIZE / sizeof(uint16_t), maxIndex, &result);
swTransform.BuildDrawingParams(prim, indexGen.VertexCount(), dec_->VertexType(), inds, maxIndex, &result);
if (result.setSafeSize)
framebufferManager_->SetSafeSize(result.safeWidth, result.safeHeight);
@ -684,11 +683,11 @@ rotateVBO:
UINT iOffset;
int iSize = sizeof(uint16_t) * result.drawNumTrans;
uint8_t *iptr = pushInds_->BeginPush(context_, &iOffset, iSize);
memcpy(iptr, inds + indsOffset, iSize);
memcpy(iptr, inds, iSize);
pushInds_->EndPush(context_);
context_->IASetIndexBuffer(pushInds_->Buf(), DXGI_FORMAT_R16_UINT, iOffset);
context_->DrawIndexed(result.drawNumTrans, 0, 0);
} else if (result.drawNumTrans > 0) {
} else {
context_->Draw(result.drawNumTrans, 0);
}
} else if (result.action == SW_CLEAR) {

9
GPU/Directx9/DrawEngineDX9.cpp
View File

@ -558,7 +558,7 @@ rotateVBO:
prim = GE_PRIM_TRIANGLES;
VERBOSE_LOG(G3D, "Flush prim %i SW! %i verts in one go", prim, indexGen.VertexCount());
u16 *const inds = decIndex_;
u16 *inds = decIndex_;
SoftwareTransformResult result{};
SoftwareTransformParams params{};
params.decoded = decoded_;
@ -607,9 +607,8 @@ rotateVBO:
ApplyDrawState(prim);
int indsOffset = 0;
if (result.action == SW_NOT_READY)
swTransform.BuildDrawingParams(prim, indexGen.VertexCount(), dec_->VertexType(), inds, indsOffset, DECODED_INDEX_BUFFER_SIZE / sizeof(uint16_t), maxIndex, &result);
swTransform.BuildDrawingParams(prim, indexGen.VertexCount(), dec_->VertexType(), inds, maxIndex, &result);
if (result.setSafeSize)
framebufferManager_->SetSafeSize(result.safeWidth, result.safeHeight);
@ -629,8 +628,8 @@ rotateVBO:
device_->SetVertexDeclaration(transformedVertexDecl_);
if (result.drawIndexed) {
device_->DrawIndexedPrimitiveUP(d3d_prim[prim], 0, maxIndex, D3DPrimCount(d3d_prim[prim], result.drawNumTrans), inds + indsOffset, D3DFMT_INDEX16, result.drawBuffer, sizeof(TransformedVertex));
} else if (result.drawNumTrans > 0) {
device_->DrawIndexedPrimitiveUP(d3d_prim[prim], 0, maxIndex, D3DPrimCount(d3d_prim[prim], result.drawNumTrans), inds, D3DFMT_INDEX16, result.drawBuffer, sizeof(TransformedVertex));
} else {
device_->DrawPrimitiveUP(d3d_prim[prim], D3DPrimCount(d3d_prim[prim], result.drawNumTrans), result.drawBuffer, sizeof(TransformedVertex));
}
} else if (result.action == SW_CLEAR) {

9
GPU/GLES/DrawEngineGLES.cpp
View File

@ -357,7 +357,7 @@ void DrawEngineGLES::DoFlush() {
if (prim == GE_PRIM_TRIANGLE_STRIP)
prim = GE_PRIM_TRIANGLES;
u16 *const inds = decIndex_;
u16 *inds = decIndex_;
SoftwareTransformResult result{};
// TODO: Keep this static? Faster than repopulating?
SoftwareTransformParams params{};
@ -414,9 +414,8 @@ void DrawEngineGLES::DoFlush() {
// Need to ApplyDrawState after ApplyTexture because depal can launch a render pass and that wrecks the state.
ApplyDrawState(prim);
int indsOffset = 0;
if (result.action == SW_NOT_READY)
swTransform.BuildDrawingParams(prim, vertexCount, dec_->VertexType(), inds, indsOffset, DECODED_INDEX_BUFFER_SIZE / sizeof(uint16_t), maxIndex, &result);
swTransform.BuildDrawingParams(prim, vertexCount, dec_->VertexType(), inds, maxIndex, &result);
if (result.setSafeSize)
framebufferManager_->SetSafeSize(result.safeWidth, result.safeHeight);
@ -431,11 +430,11 @@ void DrawEngineGLES::DoFlush() {
if (result.action == SW_DRAW_PRIMITIVES) {
if (result.drawIndexed) {
vertexBufferOffset = (uint32_t)frameData.pushVertex->Push(result.drawBuffer, maxIndex * sizeof(TransformedVertex), 4, &vertexBuffer);
indexBufferOffset = (uint32_t)frameData.pushIndex->Push(inds + indsOffset, sizeof(uint16_t) * result.drawNumTrans, 2, &indexBuffer);
indexBufferOffset = (uint32_t)frameData.pushIndex->Push(inds, sizeof(uint16_t) * result.drawNumTrans, 2, &indexBuffer);
render_->DrawIndexed(
softwareInputLayout_, vertexBuffer, vertexBufferOffset, indexBuffer, indexBufferOffset,
glprim[prim], result.drawNumTrans, GL_UNSIGNED_SHORT);
} else if (result.drawNumTrans > 0) {
} else {
vertexBufferOffset = (uint32_t)frameData.pushVertex->Push(result.drawBuffer, result.drawNumTrans * sizeof(TransformedVertex), 4, &vertexBuffer);
render_->Draw(
softwareInputLayout_, vertexBuffer, vertexBufferOffset, glprim[prim], 0, result.drawNumTrans);

9
GPU/Vulkan/DrawEngineVulkan.cpp
View File

@ -858,7 +858,7 @@ void DrawEngineVulkan::DoFlush() {
if (prim == GE_PRIM_TRIANGLE_STRIP)
prim = GE_PRIM_TRIANGLES;
u16 *const inds = decIndex_;
u16 *inds = decIndex_;
SoftwareTransformResult result{};
SoftwareTransformParams params{};
params.decoded = decoded_;
@ -898,10 +898,9 @@ void DrawEngineVulkan::DoFlush() {
// Games sometimes expect exact matches (see #12626, for example) for equal comparisons.
if (result.action == SW_CLEAR && everUsedEqualDepth_ && gstate.isClearModeDepthMask() && result.depth > 0.0f && result.depth < 1.0f)
result.action = SW_NOT_READY;
int indsOffset = 0;
if (result.action == SW_NOT_READY) {
swTransform.DetectOffsetTexture(maxIndex);
swTransform.BuildDrawingParams(prim, indexGen.VertexCount(), dec_->VertexType(), inds, indsOffset, DECODED_INDEX_BUFFER_SIZE / sizeof(uint16_t), maxIndex, &result);
swTransform.BuildDrawingParams(prim, indexGen.VertexCount(), dec_->VertexType(), inds, maxIndex, &result);
}
if (result.setSafeSize)
@ -972,9 +971,9 @@ void DrawEngineVulkan::DoFlush() {
if (result.drawIndexed) {
VkBuffer vbuf, ibuf;
vbOffset = (uint32_t)pushVertex_->Push(result.drawBuffer, maxIndex * sizeof(TransformedVertex), 4, &vbuf);
ibOffset = (uint32_t)pushIndex_->Push(inds + indsOffset, sizeof(short) * result.drawNumTrans, 4, &ibuf);
ibOffset = (uint32_t)pushIndex_->Push(inds, sizeof(short) * result.drawNumTrans, 4, &ibuf);
renderManager->DrawIndexed(ds, ARRAY_SIZE(dynamicUBOOffsets), dynamicUBOOffsets, vbuf, vbOffset, ibuf, ibOffset, result.drawNumTrans, 1);
} else if (result.drawNumTrans > 0) {
} else {
VkBuffer vbuf;
vbOffset = (uint32_t)pushVertex_->Push(result.drawBuffer, result.drawNumTrans * sizeof(TransformedVertex), 4, &vbuf);
renderManager->Draw(ds, ARRAY_SIZE(dynamicUBOOffsets), dynamicUBOOffsets, vbuf, vbOffset, result.drawNumTrans);