piwalker/melonDS
1
0
Fork 0
mirror of https://github.com/melonDS-emu/melonDS.git synced 2024-11-14 13:27:41 -07:00
Code Issues Actions 3 Packages Projects Releases Wiki Activity

more cleanup + "fix" RDLines_Count

Browse Source 
fix feels wrong, but i can't prove it either way yet.
This commit is contained in:
Jaklyy 2024-03-06 08:33:21 -05:00
parent 246fa18ab6
commit b32f519c5a
2 changed files with 69 additions and 30 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

22
src/GPU3D.h
View File

@ -341,19 +341,19 @@ public:
static constexpr int DelayBetweenReads = 809 * TimingFrac;
static constexpr int ScanlineReadSpeed = 256 * TimingFrac;
static constexpr int ScanlineReadInc = DelayBetweenReads + ScanlineReadSpeed;
static constexpr int InitGPU2DTimeout = 51875 * TimingFrac; // 51618? 51874? 52128? | when it finishes reading the first scanline.
static constexpr int FrameLength = ScanlineReadInc * 263; // how long the entire frame is. TODO: Verify if we actually need this?
//static constexpr int GPU2DSpeedFirstInPair = 810 * TimingFrac; // 810 | the delay between finishing reading a pair and beginning reading a new pair.
//static constexpr int GPU2DSpeedSecondInPair = 296 * TimingFrac; // 296 | 295??? | the delay between finishing reading the first scanline
// and beginning reading the second scanline of a scanline pair.
//static constexpr int GPU2DReadScanline = 256 * TimingFrac; // 256 | the time it takes to read a scanline.
//static constexpr int GPU2DReadSLPair = 1618 * TimingFrac; // 1618 | notably the same as the scanline increment.
static constexpr int InitGPU2DTimeout = 51875 * TimingFrac; // 51618? 51874? 52128? | when it finishes reading the first scanline.
//static constexpr int GPU2D48Scanlines = GPU2DReadSLPair * 24; // time to read 48 scanlines.
static constexpr int FrameLength = ScanlineReadInc * 263; // how long the entire frame is. TODO: Verify if we actually need this?
// compile-time list of scanline read times
// these *should* always occur at the same point in each frame, so it shouldn't matter if we make them fixed
constexpr std::array<u32, 192> SLRead = []() constexpr {
static constexpr std::array<u32, 192> SLRead = []() constexpr {
std::array<u32, 192> readtime {};
for (int i = 0, time = InitGPU2DTimeout; i < 192; i++, time += ScanlineReadInc)
@ -363,6 +363,20 @@ public:
return readtime;
}();
static constexpr int Arbitrary = 565; // extra value after the scanline is read at which the cutoff of a scanline should be...?
// idk why this is needed. im probably doing something wrong.
// the point at which rdlines decrements not sure why it's different...?
static constexpr std::array<u32, 192> RDDecrement = []() constexpr {
std::array<u32, 192> dec {};
for (int i = 0; i < 192; i++)
{
dec[i] = SLRead[i] + Arbitrary - 39 - (!(i % 2));
}
return dec;
}();
// GPU 3D Rasterization Timings: For Emulating Scanline Timeout
//static constexpr int ScanlinePairLength = 2130 * TimingFrac;
@ -378,6 +392,8 @@ public:
// (can be interpreted as the minimum amount of cycles for the next scanline
// pair to start after the previous pair began) (related to final pass?)
static constexpr int ScanlinePushDelay = 242 * TimingFrac;
static constexpr int EMGlitchThreshhold = 502 * TimingFrac; // The threshold for the edge marking glitch behavior to change.
static constexpr int EMFixNum = 571 * TimingFrac; // Arbitrary value added to fix edge marking glitch, not sure why it's needed?
//static constexpr int TimeoutIncrement = 2130 * TimingFrac;
//static constexpr int ScanlineIncrementold = 1618 * TimingFrac; // 1618 | how much to regain per scanline pair
//static constexpr int ScanlineIncrement = 2114 * TimingFrac; // 2114 | how much time a scanline pair "gains"

63
src/GPU3D_Soft.cpp
View File

@ -1900,6 +1900,8 @@ void SoftRenderer::RenderPolygons(GPU& gpu, Polygon** polygons, int npolys)
s32 scanlineswaiting = 0;
s32 nextread = 0;
s32 slwaitingrd = 0;
s32 nextreadrd = 0;
u32 timespent;
u32 prevtimespent;
@ -1932,23 +1934,33 @@ void SoftRenderer::RenderPolygons(GPU& gpu, Polygon** polygons, int npolys)
if (ScanlineTimeout <= RasterTiming) gpu.GPU3D.RDLinesUnderflow = true;
scanlineswaiting++;
slwaitingrd++;
// simulate the process of scanlines being read from the 48 scanline buffer
while (RasterTiming >= SLRead[nextread] + 565)
while (RasterTiming >= SLRead[nextread] + Arbitrary)
{
if (RasterTiming < SLRead[nextread] + 565)
if (RasterTiming < SLRead[nextread] + Arbitrary)
{
RasterTiming += timespent = (SLRead[nextread] + 565) - RasterTiming; // why + 565?
timespent += 571; // fixes edge marking bug emulation. not sure why this is needed?
RasterTiming += timespent = (SLRead[nextread] + Arbitrary) - RasterTiming; // why + 565?
timespent += EMFixNum; // fixes edge marking bug emulation. not sure why this is needed?
}
scanlineswaiting--;
nextread++;
// update rdlines_count register
if (gpu.GPU3D.RDLinesTemp > scanlineswaiting) gpu.GPU3D.RDLinesTemp = scanlineswaiting; // TODO: not accurate, rdlines appears to update early in some manner?
//if (gpu.GPU3D.RDLinesTemp > scanlineswaiting) gpu.GPU3D.RDLinesTemp = scanlineswaiting; // TODO: not accurate, rdlines appears to update early in some manner?
}
// feels wrong, needs improvement.
while (RasterTiming >= RDDecrement[nextreadrd])
{
slwaitingrd--;
nextreadrd++;
// update rdlines_count register
if (gpu.GPU3D.RDLinesTemp > slwaitingrd) gpu.GPU3D.RDLinesTemp = slwaitingrd;
}
// final pass pairs are the previous scanline pair offset -1 scanline, thus we start with only building one
ScanlineFinalPass<true>(gpu.GPU3D, 0, true, timespent >= 502);
ScanlineFinalPass<true>(gpu.GPU3D, 0, true, timespent >= EMGlitchThreshhold);
for (int y = 4; y < 192; y+=2)
{
//update sl timeout
@ -1965,45 +1977,56 @@ void SoftRenderer::RenderPolygons(GPU& gpu, Polygon** polygons, int npolys)
if (ScanlineTimeout <= RasterTiming) gpu.GPU3D.RDLinesUnderflow = true;
scanlineswaiting += 2;
slwaitingrd += 2;
// simulate the process of scanlines being read from the 48 scanline buffer
while (scanlineswaiting >= 47 || RasterTiming >= SLRead[nextread] + 565)
while (scanlineswaiting >= 47 || RasterTiming >= SLRead[nextread] + Arbitrary)
{
if (RasterTiming < SLRead[nextread] + 565)
if (RasterTiming < SLRead[nextread] + Arbitrary)
{
RasterTiming += timespent = (SLRead[nextread] + 565) - RasterTiming; // why + 565?
timespent += 571; // fixes edge marking bug emulation. not sure why this is needed?
RasterTiming += timespent = (SLRead[nextread] + Arbitrary) - RasterTiming; // why + 565?
timespent += EMFixNum; // fixes edge marking bug emulation. not sure why this is needed?
}
scanlineswaiting--;
nextread++;
// update rdlines_count register
if (gpu.GPU3D.RDLinesTemp > scanlineswaiting) gpu.GPU3D.RDLinesTemp = scanlineswaiting; // TODO: not accurate, rdlines appears to update early in some manner?
//if (gpu.GPU3D.RDLinesTemp > scanlineswaiting) gpu.GPU3D.RDLinesTemp = scanlineswaiting; // TODO: not accurate, rdlines appears to update early in some manner?
}
ScanlineFinalPass<true>(gpu.GPU3D, y-3, prevtimespent >= 502 || y-3 == 1, timespent >= 502);
ScanlineFinalPass<true>(gpu.GPU3D, y-2, prevtimespent >= 502, timespent >= 502);
// feels wrong, needs improvement.
while (RasterTiming >= RDDecrement[nextreadrd])
{
slwaitingrd--;
nextreadrd++;
// update rdlines_count register
if (gpu.GPU3D.RDLinesTemp > slwaitingrd) gpu.GPU3D.RDLinesTemp = slwaitingrd;
}
ScanlineFinalPass<true>(gpu.GPU3D, y-3, prevtimespent >= EMGlitchThreshhold || y-3 == 1, timespent >= EMGlitchThreshhold);
ScanlineFinalPass<true>(gpu.GPU3D, y-2, prevtimespent >= EMGlitchThreshhold, timespent >= EMGlitchThreshhold);
}
scanlineswaiting += 2;
slwaitingrd += 2;
prevtimespent = timespent;
// emulate read timings one last time, since it shouldn't matter after this
// additionally dont bother tracking rdlines anymore since it shouldn't be able to decrement anymore (CHECKME)
while (scanlineswaiting >= 47 || RasterTiming >= SLRead[nextread] + 565)
while (scanlineswaiting >= 47 || RasterTiming >= SLRead[nextread] + Arbitrary)
{
if (RasterTiming < SLRead[nextread] + 565)
if (RasterTiming < SLRead[nextread] + Arbitrary)
{
RasterTiming += timespent = (SLRead[nextread] + 565) - RasterTiming; // why + 565?
timespent += 571; // fixes edge marking bug emulation. not sure why this is needed?
RasterTiming += timespent = (SLRead[nextread] + Arbitrary) - RasterTiming; // why + 565?
timespent += EMFixNum; // fixes edge marking bug emulation. not sure why this is needed?
}
scanlineswaiting--;
nextread++;
}
// finish the last 3 scanlines
ScanlineFinalPass<true>(gpu.GPU3D, 189, prevtimespent >= 502, timespent >= 502);
ScanlineFinalPass<true>(gpu.GPU3D, 190, prevtimespent >= 502, true);
ScanlineFinalPass<true>(gpu.GPU3D, 189, prevtimespent >= EMGlitchThreshhold, timespent >= EMGlitchThreshhold);
ScanlineFinalPass<true>(gpu.GPU3D, 190, prevtimespent >= EMGlitchThreshhold, true);
ScanlineFinalPass<true>(gpu.GPU3D, 191, timespent >= 502, true);
ScanlineFinalPass<true>(gpu.GPU3D, 191, timespent >= EMGlitchThreshhold, true);
}
/*else
{