Emulation/ppsspp
1
0
Fork 0
mirror of https://github.com/hrydgard/ppsspp.git synced 2025-04-02 11:01:50 -04:00
Code Issues Projects Releases Packages Wiki Activity

SSE addstrip: Add the early-outs.

Browse source
This commit is contained in:
Henrik Rydgård 2023-06-13 11:35:26 +02:00
parent 39034586a4
commit 77da36c03f
2 changed files with 22 additions and 15 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

6
GPU/Common/DrawEngineCommon.cpp
View file

@ -104,9 +104,11 @@ int DrawEngineCommon::ComputeNumVertsToDecode() const {
}
void DrawEngineCommon::DecodeVerts(u8 *dest) {
for (; decodeCounter_ < numDrawCalls_; decodeCounter_++) {
DecodeVertsStep(dest, decodeCounter_, decodedVerts_, &drawCalls_[decodeCounter_].uvScale); // NOTE! DecodeVertsStep can modify decodeCounter_!
int decodeCounter = decodeCounter_;
for (; decodeCounter < numDrawCalls_; decodeCounter++) {
DecodeVertsStep(dest, decodeCounter, decodedVerts_, &drawCalls_[decodeCounter].uvScale); // NOTE! DecodeVertsStep can modify decodeCounter_!
}
decodeCounter_ = decodeCounter;
// Sanity check
if (indexGen.Prim() < 0) {

31
GPU/Common/IndexGenerator.cpp
View file

@ -131,23 +131,28 @@ void IndexGenerator::AddStrip(int numVerts, bool clockwise) {
// That's alright as we're appending to a buffer - they will get overwritten anyway.
int numChunks = (numTris + 7) / 8;
__m128i ibase8 = _mm_set1_epi16(index_);
__m128i increment = _mm_set1_epi16(8);
const __m128i *offsets = (const __m128i *)(clockwise ? offsets_clockwise : offsets_counter_clockwise);
__m128i offsets0 = _mm_add_epi16(ibase8, _mm_load_si128(offsets));
__m128i offsets1 = _mm_add_epi16(ibase8, _mm_load_si128(offsets + 1));
__m128i offsets2 = _mm_add_epi16(ibase8, _mm_load_si128(offsets + 2));
__m128i *dst = (__m128i *)inds_;
__m128i offsets0 = _mm_add_epi16(ibase8, _mm_load_si128(offsets));
// A single store is always enough for two triangles, which is a very common case.
_mm_storeu_si128(dst, offsets0);
_mm_storeu_si128(dst + 1, offsets1);
_mm_storeu_si128(dst + 2, offsets2);
for (int i = 1; i < numChunks; i++) {
offsets0 = _mm_add_epi16(offsets0, increment);
offsets1 = _mm_add_epi16(offsets1, increment);
offsets2 = _mm_add_epi16(offsets2, increment);
dst += 3;
_mm_storeu_si128(dst, offsets0);
if (numTris > 2) {
__m128i offsets1 = _mm_add_epi16(ibase8, _mm_load_si128(offsets + 1));
_mm_storeu_si128(dst + 1, offsets1);
_mm_storeu_si128(dst + 2, offsets2);
if (numTris > 5) {
__m128i offsets2 = _mm_add_epi16(ibase8, _mm_load_si128(offsets + 2));
_mm_storeu_si128(dst + 2, offsets2);
__m128i increment = _mm_set1_epi16(8);
for (int i = 1; i < numChunks; i++) {
dst += 3;
offsets0 = _mm_add_epi16(offsets0, increment);
offsets1 = _mm_add_epi16(offsets1, increment);
offsets2 = _mm_add_epi16(offsets2, increment);
_mm_storeu_si128(dst, offsets0);
_mm_storeu_si128(dst + 1, offsets1);
_mm_storeu_si128(dst + 2, offsets2);
}
}
}
inds_ += numTris * 3;
// wind doesn't need to be updated, an even number of triangles have been drawn.