Merge pull request #2606 from Sonicadvance1/aarch64_optimize_psq_st

[AArch64] Optimize paired quantized stores.

2 files changed, 26 insertions, 2 deletions

diff --git a/Source/Core/Common/Arm64Emitter.cpp b/Source/Core/Common/Arm64Emitter.cpp
index 0c4d30c5db..600b424d51 100644
--- a/Source/Core/Common/Arm64Emitter.cpp
+++ b/Source/Core/Common/Arm64Emitter.cpp
@@ -3131,10 +3131,29 @@ void ARM64FloatEmitter::UCVTF(u8 size, ARM64Reg Rd, ARM64Reg Rn, int scale)
 	int imm = size * 2 - scale;
 	EmitShiftImm(IsQuad(Rd), 1, imm >> 3, imm & 7, 0x1C, Rd, Rn);
 }
-
+void ARM64FloatEmitter::SQXTN(u8 dest_size, ARM64Reg Rd, ARM64Reg Rn)
+{
+	Emit2RegMisc(false, 0, dest_size >> 4, 0b10100, Rd, Rn);
+}
+void ARM64FloatEmitter::SQXTN2(u8 dest_size, ARM64Reg Rd, ARM64Reg Rn)
+{
+	Emit2RegMisc(true, 0, dest_size >> 4, 0b10100, Rd, Rn);
+}
+void ARM64FloatEmitter::UQXTN(u8 dest_size, ARM64Reg Rd, ARM64Reg Rn)
+{
+	Emit2RegMisc(false, 1, dest_size >> 4, 0b10100, Rd, Rn);
+}
+void ARM64FloatEmitter::UQXTN2(u8 dest_size, ARM64Reg Rd, ARM64Reg Rn)
+{
+	Emit2RegMisc(true, 1, dest_size >> 4, 0b10100, Rd, Rn);
+}
 void ARM64FloatEmitter::XTN(u8 dest_size, ARM64Reg Rd, ARM64Reg Rn)
 {
-	Emit2RegMisc(IsQuad(Rd), 0, dest_size >> 4, 0x12, Rd, Rn);
+	Emit2RegMisc(false, 0, dest_size >> 4, 0b10010, Rd, Rn);
+}
+void ARM64FloatEmitter::XTN2(u8 dest_size, ARM64Reg Rd, ARM64Reg Rn)
+{
+	Emit2RegMisc(true, 0, dest_size >> 4, 0b10010, Rd, Rn);
 }
 
 // Move
diff --git a/Source/Core/Common/Arm64Emitter.h b/Source/Core/Common/Arm64Emitter.h
index 5796b877f6..d4d95ba78d 100644
--- a/Source/Core/Common/Arm64Emitter.h
+++ b/Source/Core/Common/Arm64Emitter.h
@@ -829,7 +829,12 @@ public:
 	void UCVTF(u8 size, ARM64Reg Rd, ARM64Reg Rn);
 	void SCVTF(u8 size, ARM64Reg Rd, ARM64Reg Rn, int scale);
 	void UCVTF(u8 size, ARM64Reg Rd, ARM64Reg Rn, int scale);
+	void SQXTN(u8 dest_size, ARM64Reg Rd, ARM64Reg Rn);
+	void SQXTN2(u8 dest_size, ARM64Reg Rd, ARM64Reg Rn);
+	void UQXTN(u8 dest_size, ARM64Reg Rd, ARM64Reg Rn);
+	void UQXTN2(u8 dest_size, ARM64Reg Rd, ARM64Reg Rn);
 	void XTN(u8 dest_size, ARM64Reg Rd, ARM64Reg Rn);
+	void XTN2(u8 dest_size, ARM64Reg Rd, ARM64Reg Rn);
 
 	// Move
 	void DUP(u8 size, ARM64Reg Rd, ARM64Reg Rn);