preserve-emulation/ppsspp
1
0
Fork 0
mirror of https://github.com/hrydgard/ppsspp.git synced 2026-07-11 01:25:07 +02:00
Code Issues Packages Projects Releases Wiki Activity

Viewport-scissor dirty state cleanup

Browse Source
This commit is contained in:
Henrik Rydgård
2026-05-30 09:38:18 +02:00
parent 0e8ba0d6fe
commit 116d91fcaa
2 changed files with 13 additions and 46 deletions
Download Patch File Download Diff File Expand all files Collapse all files
GPU/Common/GPUStateUtils.cpp
+2 -36
View File
@@ -518,37 +518,9 @@ ReplaceBlendType ReplaceBlendWithShader(GEBufferFormat bufferFormat) {
return REPLACE_BLEND_STANDARD;
}
static const float DEPTH_SLICE_FACTOR_HIGH = 4.0f;
static const float DEPTH_SLICE_FACTOR_16BIT = 256.0f;
// The supported flag combinations. TODO: Maybe they should be distilled down into an enum.
//
// 0 - "Old"-style GL depth.
// Or "Non-accurate depth" : effectively ignore minz / maxz. Map Z values based on viewport, which clamps.
// This skews depth in many instances. Depth can be inverted in this mode if viewport says.
// This is completely wrong, but works in some cases (probably because some game devs assumed it was how it worked)
// and avoids some depth clamp issues.
//
// GPU_USE_ACCURATE_DEPTH:
// Accurate depth: Z in the framebuffer matches the range of Z used on the PSP linearly in some way. We choose
// a centered range, to simulate clamping by letting otherwise out-of-range pixels survive the 0 and 1 cutoffs.
// Clip depth based on minz/maxz, and viewport is just a means to scale and center the value, not clipping or mapping to stored values.
//
// GPU_USE_ACCURATE_DEPTH | GPU_USE_DEPTH_CLAMP:
// Variant of GPU_USE_ACCURATE_DEPTH, just the range is the nice and convenient 0-1 since we can use
// hardware depth clamp. only viable in accurate depth mode, clamps depth and therefore uses the full 0-1 range. Using the full 0-1 range is not what accurate means, it's implied by depth clamp (which also means we're clamping.)
//
// GPU_USE_ACCURATE_DEPTH | GPU_SCALE_DEPTH_FROM_24BIT_TO_16BIT:
// GPU_USE_ACCURATE_DEPTH | GPU_SCALE_DEPTH_FROM_24BIT_TO_16BIT | GPU_USE_DEPTH_CLAMP:
// Only viable in accurate depth mode, means to use a range of the 24-bit depth values available
// from the GPU to represent the 16-bit values the PSP had, to try to make everything round and
// z-fight (close to) the same way as on hardware, cheaply (cheaper than rounding depth in fragment shader).
// We automatically switch to this if Z tests for equality are used.
// Depth clamp has no effect on the depth scaling here if set, though will still be enabled
// and clamp wildly out of line values.
//
// Any other combinations of these particular flags are bogus (like for example a lonely GPU_USE_DEPTH_CLAMP).
// Currently obsolete. We may reintroduce squeezing 24-bit depth into a 16-bit range, although that
// will mess with hardware depth clamp, so likely not worth it anymore.
float DepthSliceFactor(u32 useFlags) {
return 1.0f;
}
@@ -602,12 +574,6 @@ void ConvertViewportAndScissor(const DisplayLayoutConfig &config, bool useBuffer
int curRTWidth = gstate_c.curRTWidth;
int curRTHeight = gstate_c.curRTHeight;
float offsetX = gstate.getOffsetX();
float offsetY = gstate.getOffsetY();
// If renderX/renderY are offset to compensate for a split framebuffer,
// applying the offset to the viewport isn't enough, since the viewport clips.
// We need to apply either directly to the vertices, or to the "through" projection matrix.
out.viewportX = displayOffsetX;
out.viewportY = displayOffsetY;
out.viewportW = curRTWidth * renderWidthFactor;
GPU/GPUCommonHW.cpp
+11 -10
View File
@@ -192,14 +192,14 @@ const CommonCommandTableEntry commonCommandTable[] = {
{ GE_CMD_ANTIALIASENABLE, FLAG_FLUSHBEFOREONCHANGE },
// Viewport.
{ GE_CMD_OFFSETX, FLAG_FLUSHBEFOREONCHANGE, DIRTY_VIEWPORTSCISSOR_STATE | DIRTY_RASTER_OFFSET },
{ GE_CMD_OFFSETY, FLAG_FLUSHBEFOREONCHANGE, DIRTY_VIEWPORTSCISSOR_STATE | DIRTY_RASTER_OFFSET },
{ GE_CMD_VIEWPORTXSCALE, FLAG_FLUSHBEFOREONCHANGE, DIRTY_FRAMEBUF | DIRTY_TEXTURE_PARAMS | DIRTY_PROJMATRIX | DIRTY_VIEWPORTSCISSOR_STATE | DIRTY_VIEWPORT_UNIFORMS },
{ GE_CMD_VIEWPORTYSCALE, FLAG_FLUSHBEFOREONCHANGE, DIRTY_FRAMEBUF | DIRTY_TEXTURE_PARAMS | DIRTY_PROJMATRIX | DIRTY_VIEWPORTSCISSOR_STATE | DIRTY_VIEWPORT_UNIFORMS },
{ GE_CMD_VIEWPORTXCENTER, FLAG_FLUSHBEFOREONCHANGE, DIRTY_FRAMEBUF | DIRTY_TEXTURE_PARAMS | DIRTY_PROJMATRIX | DIRTY_VIEWPORTSCISSOR_STATE | DIRTY_VIEWPORT_UNIFORMS },
{ GE_CMD_VIEWPORTYCENTER, FLAG_FLUSHBEFOREONCHANGE, DIRTY_FRAMEBUF | DIRTY_TEXTURE_PARAMS | DIRTY_PROJMATRIX | DIRTY_VIEWPORTSCISSOR_STATE | DIRTY_VIEWPORT_UNIFORMS },
{ GE_CMD_VIEWPORTZSCALE, FLAG_FLUSHBEFOREONCHANGE, DIRTY_FRAMEBUF | DIRTY_TEXTURE_PARAMS | DIRTY_PROJMATRIX | DIRTY_VIEWPORTSCISSOR_STATE | DIRTY_VIEWPORT_UNIFORMS },
{ GE_CMD_VIEWPORTZCENTER, FLAG_FLUSHBEFOREONCHANGE, DIRTY_FRAMEBUF | DIRTY_TEXTURE_PARAMS | DIRTY_PROJMATRIX | DIRTY_VIEWPORTSCISSOR_STATE | DIRTY_VIEWPORT_UNIFORMS },
{ GE_CMD_OFFSETX, FLAG_FLUSHBEFOREONCHANGE, DIRTY_RASTER_OFFSET },
{ GE_CMD_OFFSETY, FLAG_FLUSHBEFOREONCHANGE, DIRTY_RASTER_OFFSET },
{ GE_CMD_VIEWPORTXSCALE, FLAG_FLUSHBEFOREONCHANGE, DIRTY_FRAMEBUF | DIRTY_TEXTURE_PARAMS | DIRTY_PROJMATRIX | DIRTY_VIEWPORT_UNIFORMS | DIRTY_VIEWPORTSCISSOR_STATE }, // GetFramebufferHeuristicInputs still uses viewport scale
{ GE_CMD_VIEWPORTYSCALE, FLAG_FLUSHBEFOREONCHANGE, DIRTY_FRAMEBUF | DIRTY_TEXTURE_PARAMS | DIRTY_PROJMATRIX | DIRTY_VIEWPORT_UNIFORMS | DIRTY_VIEWPORTSCISSOR_STATE },
{ GE_CMD_VIEWPORTXCENTER, FLAG_FLUSHBEFOREONCHANGE, DIRTY_FRAMEBUF | DIRTY_TEXTURE_PARAMS | DIRTY_PROJMATRIX | DIRTY_VIEWPORT_UNIFORMS },
{ GE_CMD_VIEWPORTYCENTER, FLAG_FLUSHBEFOREONCHANGE, DIRTY_FRAMEBUF | DIRTY_TEXTURE_PARAMS | DIRTY_PROJMATRIX | DIRTY_VIEWPORT_UNIFORMS },
{ GE_CMD_VIEWPORTZSCALE, FLAG_FLUSHBEFOREONCHANGE, DIRTY_FRAMEBUF | DIRTY_TEXTURE_PARAMS | DIRTY_PROJMATRIX | DIRTY_VIEWPORT_UNIFORMS },
{ GE_CMD_VIEWPORTZCENTER, FLAG_FLUSHBEFOREONCHANGE, DIRTY_FRAMEBUF | DIRTY_TEXTURE_PARAMS | DIRTY_PROJMATRIX | DIRTY_VIEWPORT_UNIFORMS },
// Z range and clipping
{ GE_CMD_DEPTHCLIPENABLE, FLAG_FLUSHBEFOREONCHANGE, DIRTY_VERTEXSHADER_STATE | DIRTY_FRAGMENTSHADER_STATE | DIRTY_RASTER_STATE },
@@ -854,7 +854,7 @@ void GPUCommonHW::Execute_VertexType(u32 op, u32 diff) {
if (diff & GE_VTYPE_THROUGH_MASK) {
// Switching between through and non-through, we need to invalidate a bunch of stuff.
gstate_c.Dirty(DIRTY_RASTER_STATE | DIRTY_VIEWPORTSCISSOR_STATE | DIRTY_FRAGMENTSHADER_STATE);
gstate_c.Dirty(DIRTY_RASTER_STATE | DIRTY_VERTEXSHADER_STATE | DIRTY_VIEWPORTSCISSOR_STATE | DIRTY_FRAGMENTSHADER_STATE);
}
}
}
@@ -874,8 +874,9 @@ void GPUCommonHW::Execute_VertexTypeSkinning(u32 op, u32 diff) {
}
gstate_c.Dirty(DIRTY_VERTEXSHADER_STATE);
}
if (diff & GE_VTYPE_THROUGH_MASK)
gstate_c.Dirty(DIRTY_RASTER_STATE | DIRTY_VIEWPORTSCISSOR_STATE | DIRTY_FRAGMENTSHADER_STATE);
gstate_c.Dirty(DIRTY_RASTER_STATE | DIRTY_VERTEXSHADER_STATE | DIRTY_VIEWPORTSCISSOR_STATE | DIRTY_FRAGMENTSHADER_STATE);
}
void GPUCommonHW::Execute_Prim(u32 op, u32 diff) {