Jit64: implement ps_res

This is probably more accurate than it is allowed to be. After all, the instructions are supposed to be "estimates".
dolphin-emu · Nov 7, 2013 · cdc27e4 · cdc27e4
1 parent 0a2a273
commit cdc27e4
Show file tree

Hide file tree

Showing 4 changed files with 24 additions and 13 deletions.
diff --git a/Source/Core/Core/Src/PowerPC/Jit64/Jit.h b/Source/Core/Core/Src/PowerPC/Jit64/Jit.h
@@ -178,7 +178,7 @@ class Jit64 : public Jitx86Base
 	void ps_arith(UGeckoInstruction inst); //aggregate
 	void ps_mergeXX(UGeckoInstruction inst);
 	void ps_maddXX(UGeckoInstruction inst);
-	void ps_rsqrte(UGeckoInstruction inst);
+	void ps_recip(UGeckoInstruction inst);
 	void ps_sum(UGeckoInstruction inst);
 	void ps_muls(UGeckoInstruction inst);
 

diff --git a/Source/Core/Core/Src/PowerPC/Jit64/Jit64_Tables.cpp b/Source/Core/Core/Src/PowerPC/Jit64/Jit64_Tables.cpp
@@ -138,9 +138,9 @@ static GekkoOPTemplate table4_2[] =
 	{20, &Jit64::ps_arith}, //"ps_sub",    OPTYPE_PS, 0}},
 	{21, &Jit64::ps_arith}, //"ps_add",    OPTYPE_PS, 0}},
 	{23, &Jit64::ps_sel}, //"ps_sel",    OPTYPE_PS, 0}},
-	{24, &Jit64::Default}, //"ps_res",    OPTYPE_PS, 0}},
+	{24, &Jit64::ps_recip}, //"ps_res",    OPTYPE_PS, 0}},
 	{25, &Jit64::ps_arith}, //"ps_mul",    OPTYPE_PS, 0}},
-	{26, &Jit64::ps_rsqrte}, //"ps_rsqrte", OPTYPE_PS, 0, 1}},
+	{26, &Jit64::ps_recip}, //"ps_rsqrte", OPTYPE_PS, 0, 1}},
 	{28, &Jit64::ps_maddXX}, //"ps_msub",   OPTYPE_PS, 0}},
 	{29, &Jit64::ps_maddXX}, //"ps_madd",   OPTYPE_PS, 0}},
 	{30, &Jit64::ps_maddXX}, //"ps_nmsub",  OPTYPE_PS, 0}},

diff --git a/Source/Core/Core/Src/PowerPC/Jit64/Jit_FloatingPoint.cpp b/Source/Core/Core/Src/PowerPC/Jit64/Jit_FloatingPoint.cpp
@@ -8,9 +8,10 @@
 #include "JitRegCache.h"
 #include "CPUDetect.h"
 
-const u64 GC_ALIGNED16(psSignBits2[2]) = {0x8000000000000000ULL, 0x8000000000000000ULL};
-const u64 GC_ALIGNED16(psAbsMask2[2])  = {0x7FFFFFFFFFFFFFFFULL, 0x7FFFFFFFFFFFFFFFULL};
-const double GC_ALIGNED16(psOneOne2[2]) = {1.0, 1.0};
+static const u64 GC_ALIGNED16(psSignBits2[2]) = {0x8000000000000000ULL, 0x8000000000000000ULL};
+static const u64 GC_ALIGNED16(psAbsMask2[2])  = {0x7FFFFFFFFFFFFFFFULL, 0x7FFFFFFFFFFFFFFFULL};
+static const double GC_ALIGNED16(psOneOne2[2]) = {1.0, 1.0};
+static const double one_const = 1.0f;
 
 void Jit64::fp_tri_op(int d, int a, int b, bool reversible, bool dupe, void (XEmitter::*op)(Gen::X64Reg, Gen::OpArg))
 {
@@ -59,9 +60,6 @@ void Jit64::fp_tri_op(int d, int a, int b, bool reversible, bool dupe, void (XEm
 	fpr.UnlockAll();
 }
 
-
-static const double one_const = 1.0f;
-
 void Jit64::fp_arith_s(UGeckoInstruction inst)
 {
 	INSTRUCTION_START

diff --git a/Source/Core/Core/Src/PowerPC/Jit64/Jit_Paired.cpp b/Source/Core/Core/Src/PowerPC/Jit64/Jit_Paired.cpp
@@ -100,20 +100,33 @@ void Jit64::ps_sign(UGeckoInstruction inst)
 	fpr.UnlockAll();
 }
 
-void Jit64::ps_rsqrte(UGeckoInstruction inst)
+// ps_res and ps_rsqrte
+void Jit64::ps_recip(UGeckoInstruction inst)
 {
 	INSTRUCTION_START
 	JITDISABLE(bJITPairedOff)
 	if (inst.Rc) {
 		Default(inst); return;
 	}
+	OpArg divisor;
 	int d = inst.FD;
 	int b = inst.FB;
 	fpr.Lock(d, b);
-	fpr.BindToRegister(d, (d == b), true);
-	SQRTPD(XMM0, fpr.R(b));
+	fpr.BindToRegister(d, (d == b));
+	switch (inst.SUBOP5)
+	{
+	case 24:
+		// ps_res
+		divisor = fpr.R(b);
+		break;
+	case 26:
+		// ps_rsqrte
+		SQRTPD(XMM0, fpr.R(b));
+		divisor = R(XMM0);
+		break;
+	}
 	MOVAPD(XMM1, M((void*)&psOneOne));
-	DIVPD(XMM1, R(XMM0));
+	DIVPD(XMM1, divisor);
 	MOVAPD(fpr.R(d), XMM1);
 	fpr.UnlockAll();
 }