Vertex decoder: Add fallback for non-SSE4.1

See hrydgard#16157
unknownbrackets · Oct 3, 2022 · 1469a32 · 1469a32
1 parent 973d043
commit 1469a32
Showing 1 changed file with 7 additions and 1 deletion.
diff --git a/GPU/Common/VertexDecoderX86.cpp b/GPU/Common/VertexDecoderX86.cpp
@@ -1391,7 +1391,13 @@ void VertexDecoderJitCache::Jit_PosFloatThrough() {
 
 	CVTTPS2DQ(fpScratchReg, R(fpScratchReg));
 	// Use pack to saturate to 0,65535.
-	PACKUSDW(fpScratchReg, R(fpScratchReg));
+	if (cpu_info.bSSE4_1) {
+		PACKUSDW(fpScratchReg, R(fpScratchReg));
+	} else {
+		PSLLD(fpScratchReg, 16);
+		PSRAD(fpScratchReg, 16);
+		PACKSSDW(fpScratchReg, R(fpScratchReg));
+	}
 	PUNPCKLWD(fpScratchReg, R(fpScratchReg2));
 	CVTDQ2PS(fpScratchReg, R(fpScratchReg));