From cfac7324d6646a6e75ab1413553dc6610ef1b4f8 Mon Sep 17 00:00:00 2001 From: "Unknown W. Brackets" Date: Fri, 19 Apr 2013 22:47:49 -0700 Subject: [PATCH 1/4] Implement vscl in the x86 jit. --- Core/MIPS/ARM/ArmCompVFPU.cpp | 4 +++ Core/MIPS/ARM/ArmJit.h | 1 + Core/MIPS/MIPSTables.cpp | 2 +- Core/MIPS/x86/CompVFPU.cpp | 51 +++++++++++++++++++++++++++++++++++ Core/MIPS/x86/Jit.h | 1 + 5 files changed, 58 insertions(+), 1 deletion(-) diff --git a/Core/MIPS/ARM/ArmCompVFPU.cpp b/Core/MIPS/ARM/ArmCompVFPU.cpp index eb670ff0d0..21d294cdb6 100644 --- a/Core/MIPS/ARM/ArmCompVFPU.cpp +++ b/Core/MIPS/ARM/ArmCompVFPU.cpp @@ -714,4 +714,8 @@ namespace MIPSComp DISABLE; } + void Jit::Comp_VScl(u32 op) { + DISABLE; + } + } diff --git a/Core/MIPS/ARM/ArmJit.h b/Core/MIPS/ARM/ArmJit.h index 8eee4db9f1..b2cb72bebd 100644 --- a/Core/MIPS/ARM/ArmJit.h +++ b/Core/MIPS/ARM/ArmJit.h @@ -193,6 +193,7 @@ public: void Comp_Mftv(u32 op); void Comp_Vmtvc(u32 op); void Comp_Vmmov(u32 op); + void Comp_VScl(u32 op); ArmJitBlockCache *GetBlockCache() { return &blocks; } diff --git a/Core/MIPS/MIPSTables.cpp b/Core/MIPS/MIPSTables.cpp index 5fc4208d54..8b347554ea 100644 --- a/Core/MIPS/MIPSTables.cpp +++ b/Core/MIPS/MIPSTables.cpp @@ -490,7 +490,7 @@ const MIPSInstruction tableVFPU1[8] = { INSTR("vmul",&Jit::Comp_VecDo3, Dis_VectorSet3, Int_VecDo3, IS_VFPU|OUT_EAT_PREFIX), INSTR("vdot",&Jit::Comp_VDot, Dis_VectorDot, Int_VDot, IS_VFPU|OUT_EAT_PREFIX), - INSTR("vscl",&Jit::Comp_Generic, Dis_VScl, Int_VScl, IS_VFPU|OUT_EAT_PREFIX), + INSTR("vscl",&Jit::Comp_VScl, Dis_VScl, Int_VScl, IS_VFPU|OUT_EAT_PREFIX), {-2}, INSTR("vhdp",&Jit::Comp_Generic, Dis_Generic, Int_VHdp, IS_VFPU|OUT_EAT_PREFIX), INSTR("vcrs",&Jit::Comp_Generic, Dis_Vcrs, Int_Vcrs, IS_VFPU), diff --git a/Core/MIPS/x86/CompVFPU.cpp b/Core/MIPS/x86/CompVFPU.cpp index d6b3aa1ab9..e8a97249cb 100644 --- a/Core/MIPS/x86/CompVFPU.cpp +++ b/Core/MIPS/x86/CompVFPU.cpp @@ -732,4 +732,55 @@ void Jit::Comp_Vmmov(u32 op) { fpr.ReleaseSpillLocks(); } +void Jit::Comp_VScl(u32 op) { + CONDITIONAL_DISABLE; + + if (js.HasUnknownPrefix()) + DISABLE; + + VectorSize sz = GetVecSize(op); + int n = GetNumVectorElements(sz); + + u8 sregs[4], dregs[4], scale; + GetVectorRegsPrefixS(sregs, sz, _VS); + // TODO: Prefixes seem strange... + GetVectorRegsPrefixT(&scale, V_Single, _VT); + GetVectorRegsPrefixD(dregs, sz, _VD); + + // Move to XMM0 early, so we don't have to worry about overlap with scale. + MOVSS(XMM0, fpr.V(scale)); + + X64Reg tempxregs[4]; + for (int i = 0; i < n; ++i) + { + if (!IsOverlapSafeAllowS(dregs[i], i, n, sregs)) + { + int reg = fpr.GetTempV(); + fpr.MapRegV(reg, MAP_NOINIT | MAP_DIRTY); + fpr.SpillLockV(reg); + tempxregs[i] = fpr.VX(reg); + } + else + { + fpr.MapRegV(dregs[i], (dregs[i] == sregs[i] ? 0 : MAP_NOINIT) | MAP_DIRTY); + fpr.SpillLockV(dregs[i]); + tempxregs[i] = fpr.VX(dregs[i]); + } + } + for (int i = 0; i < n; ++i) + { + if (!fpr.V(sregs[i]).IsSimpleReg(tempxregs[i])) + MOVSS(tempxregs[i], fpr.V(sregs[i])); + MULSS(tempxregs[i], R(XMM0)); + } + for (int i = 0; i < n; ++i) + { + if (!fpr.V(dregs[i]).IsSimpleReg(tempxregs[i])) + MOVSS(fpr.V(dregs[i]), tempxregs[i]); + } + ApplyPrefixD(dregs, sz); + + fpr.ReleaseSpillLocks(); +} + } diff --git a/Core/MIPS/x86/Jit.h b/Core/MIPS/x86/Jit.h index 0f05e32eba..8a1c8cd979 100644 --- a/Core/MIPS/x86/Jit.h +++ b/Core/MIPS/x86/Jit.h @@ -203,6 +203,7 @@ public: void Comp_Mftv(u32 op); void Comp_Vmtvc(u32 op); void Comp_Vmmov(u32 op); + void Comp_VScl(u32 op); void Comp_DoNothing(u32 op); From 29109d25af75d50f2a1cad2ef43eb31187f914f9 Mon Sep 17 00:00:00 2001 From: "Unknown W. Brackets" Date: Sat, 20 Apr 2013 01:11:40 -0700 Subject: [PATCH 2/4] Non-optimal vmmul for x86 jit. It's faster than interpreter anyway, but it could be much better. --- Core/MIPS/ARM/ArmCompVFPU.cpp | 4 +++ Core/MIPS/ARM/ArmJit.h | 1 + Core/MIPS/MIPSTables.cpp | 8 +++--- Core/MIPS/x86/CompVFPU.cpp | 48 +++++++++++++++++++++++++++++++++++ Core/MIPS/x86/Jit.h | 1 + 5 files changed, 58 insertions(+), 4 deletions(-) diff --git a/Core/MIPS/ARM/ArmCompVFPU.cpp b/Core/MIPS/ARM/ArmCompVFPU.cpp index 21d294cdb6..bf59bb6a31 100644 --- a/Core/MIPS/ARM/ArmCompVFPU.cpp +++ b/Core/MIPS/ARM/ArmCompVFPU.cpp @@ -718,4 +718,8 @@ namespace MIPSComp DISABLE; } + void Jit::Comp_Vmmul(u32 op) { + DISABLE; + } + } diff --git a/Core/MIPS/ARM/ArmJit.h b/Core/MIPS/ARM/ArmJit.h index b2cb72bebd..1e726db716 100644 --- a/Core/MIPS/ARM/ArmJit.h +++ b/Core/MIPS/ARM/ArmJit.h @@ -194,6 +194,7 @@ public: void Comp_Vmtvc(u32 op); void Comp_Vmmov(u32 op); void Comp_VScl(u32 op); + void Comp_Vmmul(u32 op); ArmJitBlockCache *GetBlockCache() { return &blocks; } diff --git a/Core/MIPS/MIPSTables.cpp b/Core/MIPS/MIPSTables.cpp index 8b347554ea..f52773591d 100644 --- a/Core/MIPS/MIPSTables.cpp +++ b/Core/MIPS/MIPSTables.cpp @@ -626,10 +626,10 @@ MIPSInstruction tableVFPU5[8] = //110111 xxx const MIPSInstruction tableVFPU6[32] = //111100 xxx { //0 - INSTR("vmmul",&Jit::Comp_Generic, Dis_MatrixMult, Int_Vmmul, IS_VFPU|OUT_EAT_PREFIX), - INSTR("vmmul",&Jit::Comp_Generic, Dis_MatrixMult, Int_Vmmul, IS_VFPU|OUT_EAT_PREFIX), - INSTR("vmmul",&Jit::Comp_Generic, Dis_MatrixMult, Int_Vmmul, IS_VFPU|OUT_EAT_PREFIX), - INSTR("vmmul",&Jit::Comp_Generic, Dis_MatrixMult, Int_Vmmul, IS_VFPU|OUT_EAT_PREFIX), + INSTR("vmmul",&Jit::Comp_Vmmul, Dis_MatrixMult, Int_Vmmul, IS_VFPU|OUT_EAT_PREFIX), + INSTR("vmmul",&Jit::Comp_Vmmul, Dis_MatrixMult, Int_Vmmul, IS_VFPU|OUT_EAT_PREFIX), + INSTR("vmmul",&Jit::Comp_Vmmul, Dis_MatrixMult, Int_Vmmul, IS_VFPU|OUT_EAT_PREFIX), + INSTR("vmmul",&Jit::Comp_Vmmul, Dis_MatrixMult, Int_Vmmul, IS_VFPU|OUT_EAT_PREFIX), INSTR("v(h)tfm2",&Jit::Comp_Generic, Dis_Vtfm, Int_Vtfm, IS_VFPU|OUT_EAT_PREFIX), INSTR("v(h)tfm2",&Jit::Comp_Generic, Dis_Vtfm, Int_Vtfm, IS_VFPU|OUT_EAT_PREFIX), diff --git a/Core/MIPS/x86/CompVFPU.cpp b/Core/MIPS/x86/CompVFPU.cpp index e8a97249cb..2c1c733337 100644 --- a/Core/MIPS/x86/CompVFPU.cpp +++ b/Core/MIPS/x86/CompVFPU.cpp @@ -783,4 +783,52 @@ void Jit::Comp_VScl(u32 op) { fpr.ReleaseSpillLocks(); } +void Jit::Comp_Vmmul(u32 op) { + CONDITIONAL_DISABLE; + + // TODO: This probably ignores prefixes? + if (js.MayHavePrefix()) + DISABLE; + + MatrixSize sz = GetMtxSize(op); + int n = GetMatrixSide(sz); + + u8 sregs[16], tregs[16], dregs[16]; + GetMatrixRegs(sregs, sz, _VS); + GetMatrixRegs(tregs, sz, _VT); + GetMatrixRegs(dregs, sz, _VD); + + // TODO: test overlap, fix non-optimal. + u8 tempregs[16]; + for (int a = 0; a < n; a++) + { + for (int b = 0; b < n; b++) + { + XORPS(XMM0, R(XMM0)); + for (int c = 0; c < n; c++) + { + MOVSS(XMM1, fpr.V(sregs[b * 4 + c])); + MULSS(XMM1, fpr.V(tregs[a * 4 + c])); + ADDSS(XMM0, R(XMM1)); + } + u8 temp = (u8) fpr.GetTempV(); + fpr.MapRegV(temp, MAP_NOINIT | MAP_DIRTY); + MOVSS(fpr.VX(temp), R(XMM0)); + fpr.StoreFromRegisterV(temp); + tempregs[a * 4 + b] = temp; + } + } + for (int a = 0; a < n; a++) + { + for (int b = 0; b < n; b++) + { + u8 temp = tempregs[a * 4 + b]; + fpr.MapRegV(temp, 0); + MOVSS(fpr.V(dregs[a * 4 + b]), fpr.VX(temp)); + } + } + + fpr.ReleaseSpillLocks(); +} + } diff --git a/Core/MIPS/x86/Jit.h b/Core/MIPS/x86/Jit.h index 8a1c8cd979..f7f1121bf5 100644 --- a/Core/MIPS/x86/Jit.h +++ b/Core/MIPS/x86/Jit.h @@ -204,6 +204,7 @@ public: void Comp_Vmtvc(u32 op); void Comp_Vmmov(u32 op); void Comp_VScl(u32 op); + void Comp_Vmmul(u32 op); void Comp_DoNothing(u32 op); From 9245490b533c660c13d7b15cd955572e231ac0e6 Mon Sep 17 00:00:00 2001 From: "Unknown W. Brackets" Date: Sat, 20 Apr 2013 01:34:16 -0700 Subject: [PATCH 3/4] Initial / simple vmscl for x86 jit. --- Core/MIPS/ARM/ArmCompVFPU.cpp | 4 ++++ Core/MIPS/ARM/ArmJit.h | 1 + Core/MIPS/MIPSTables.cpp | 8 +++---- Core/MIPS/x86/CompVFPU.cpp | 45 +++++++++++++++++++++++++++++++++++ Core/MIPS/x86/Jit.h | 1 + 5 files changed, 55 insertions(+), 4 deletions(-) diff --git a/Core/MIPS/ARM/ArmCompVFPU.cpp b/Core/MIPS/ARM/ArmCompVFPU.cpp index bf59bb6a31..221003bc52 100644 --- a/Core/MIPS/ARM/ArmCompVFPU.cpp +++ b/Core/MIPS/ARM/ArmCompVFPU.cpp @@ -722,4 +722,8 @@ namespace MIPSComp DISABLE; } + void Jit::Comp_Vmscl(u32 op) { + DISABLE; + } + } diff --git a/Core/MIPS/ARM/ArmJit.h b/Core/MIPS/ARM/ArmJit.h index 1e726db716..715b01e929 100644 --- a/Core/MIPS/ARM/ArmJit.h +++ b/Core/MIPS/ARM/ArmJit.h @@ -195,6 +195,7 @@ public: void Comp_Vmmov(u32 op); void Comp_VScl(u32 op); void Comp_Vmmul(u32 op); + void Comp_Vmscl(u32 op); ArmJitBlockCache *GetBlockCache() { return &blocks; } diff --git a/Core/MIPS/MIPSTables.cpp b/Core/MIPS/MIPSTables.cpp index f52773591d..e8cc943e17 100644 --- a/Core/MIPS/MIPSTables.cpp +++ b/Core/MIPS/MIPSTables.cpp @@ -646,10 +646,10 @@ const MIPSInstruction tableVFPU6[32] = //111100 xxx INSTR("v(h)tfm4",&Jit::Comp_Generic, Dis_Vtfm, Int_Vtfm, IS_VFPU|OUT_EAT_PREFIX), INSTR("v(h)tfm4",&Jit::Comp_Generic, Dis_Vtfm, Int_Vtfm, IS_VFPU|OUT_EAT_PREFIX), //16 - INSTR("vmscl",&Jit::Comp_Generic, Dis_Generic, Int_Vmscl, IS_VFPU|OUT_EAT_PREFIX), - INSTR("vmscl",&Jit::Comp_Generic, Dis_Generic, Int_Vmscl, IS_VFPU|OUT_EAT_PREFIX), - INSTR("vmscl",&Jit::Comp_Generic, Dis_Generic, Int_Vmscl, IS_VFPU|OUT_EAT_PREFIX), - INSTR("vmscl",&Jit::Comp_Generic, Dis_Generic, Int_Vmscl, IS_VFPU|OUT_EAT_PREFIX), + INSTR("vmscl",&Jit::Comp_Vmscl, Dis_Generic, Int_Vmscl, IS_VFPU|OUT_EAT_PREFIX), + INSTR("vmscl",&Jit::Comp_Vmscl, Dis_Generic, Int_Vmscl, IS_VFPU|OUT_EAT_PREFIX), + INSTR("vmscl",&Jit::Comp_Vmscl, Dis_Generic, Int_Vmscl, IS_VFPU|OUT_EAT_PREFIX), + INSTR("vmscl",&Jit::Comp_Vmscl, Dis_Generic, Int_Vmscl, IS_VFPU|OUT_EAT_PREFIX), INSTR("vcrsp.t/vqmul.q",&Jit::Comp_Generic, Dis_CrossQuat, Int_CrossQuat, IS_VFPU|OUT_EAT_PREFIX), INSTR("vcrsp.t/vqmul.q",&Jit::Comp_Generic, Dis_CrossQuat, Int_CrossQuat, IS_VFPU|OUT_EAT_PREFIX), diff --git a/Core/MIPS/x86/CompVFPU.cpp b/Core/MIPS/x86/CompVFPU.cpp index 2c1c733337..86cd1a013b 100644 --- a/Core/MIPS/x86/CompVFPU.cpp +++ b/Core/MIPS/x86/CompVFPU.cpp @@ -831,4 +831,49 @@ void Jit::Comp_Vmmul(u32 op) { fpr.ReleaseSpillLocks(); } +void Jit::Comp_Vmscl(u32 op) { + CONDITIONAL_DISABLE; + + // TODO: This probably ignores prefixes? + if (js.MayHavePrefix()) + DISABLE; + + MatrixSize sz = GetMtxSize(op); + int n = GetMatrixSide(sz); + + u8 sregs[16], dregs[16], scale; + GetMatrixRegs(sregs, sz, _VS); + GetVectorRegs(&scale, V_Single, _VT); + GetMatrixRegs(dregs, sz, _VD); + + // Move to XMM0 early, so we don't have to worry about overlap with scale. + MOVSS(XMM0, fpr.V(scale)); + + // TODO: test overlap, optimize. + u8 tempregs[16]; + for (int a = 0; a < n; a++) + { + for (int b = 0; b < n; b++) + { + u8 temp = (u8) fpr.GetTempV(); + fpr.MapRegV(temp, MAP_NOINIT | MAP_DIRTY); + MOVSS(fpr.VX(temp), fpr.V(sregs[a * 4 + b])); + MULSS(fpr.VX(temp), R(XMM0)); + fpr.StoreFromRegisterV(temp); + tempregs[a * 4 + b] = temp; + } + } + for (int a = 0; a < n; a++) + { + for (int b = 0; b < n; b++) + { + u8 temp = tempregs[a * 4 + b]; + fpr.MapRegV(temp, 0); + MOVSS(fpr.V(dregs[a * 4 + b]), fpr.VX(temp)); + } + } + + fpr.ReleaseSpillLocks(); +} + } diff --git a/Core/MIPS/x86/Jit.h b/Core/MIPS/x86/Jit.h index f7f1121bf5..fe647a54ec 100644 --- a/Core/MIPS/x86/Jit.h +++ b/Core/MIPS/x86/Jit.h @@ -205,6 +205,7 @@ public: void Comp_Vmmov(u32 op); void Comp_VScl(u32 op); void Comp_Vmmul(u32 op); + void Comp_Vmscl(u32 op); void Comp_DoNothing(u32 op); From 3bb5651ca7fb2ec0ce60434a96ce00e3d9c9d01d Mon Sep 17 00:00:00 2001 From: "Unknown W. Brackets" Date: Sat, 20 Apr 2013 01:52:06 -0700 Subject: [PATCH 4/4] Initial x86 jit for vtfm/vhtfm. --- Core/MIPS/ARM/ArmCompVFPU.cpp | 4 +++ Core/MIPS/ARM/ArmJit.h | 1 + Core/MIPS/MIPSTables.cpp | 24 +++++++-------- Core/MIPS/x86/CompVFPU.cpp | 58 +++++++++++++++++++++++++++++++++++ Core/MIPS/x86/Jit.h | 1 + 5 files changed, 76 insertions(+), 12 deletions(-) diff --git a/Core/MIPS/ARM/ArmCompVFPU.cpp b/Core/MIPS/ARM/ArmCompVFPU.cpp index 221003bc52..b377103130 100644 --- a/Core/MIPS/ARM/ArmCompVFPU.cpp +++ b/Core/MIPS/ARM/ArmCompVFPU.cpp @@ -726,4 +726,8 @@ namespace MIPSComp DISABLE; } + void Jit::Comp_Vtfm(u32 op) { + DISABLE; + } + } diff --git a/Core/MIPS/ARM/ArmJit.h b/Core/MIPS/ARM/ArmJit.h index 715b01e929..8c9c78ffaa 100644 --- a/Core/MIPS/ARM/ArmJit.h +++ b/Core/MIPS/ARM/ArmJit.h @@ -196,6 +196,7 @@ public: void Comp_VScl(u32 op); void Comp_Vmmul(u32 op); void Comp_Vmscl(u32 op); + void Comp_Vtfm(u32 op); ArmJitBlockCache *GetBlockCache() { return &blocks; } diff --git a/Core/MIPS/MIPSTables.cpp b/Core/MIPS/MIPSTables.cpp index e8cc943e17..66dfee4ca2 100644 --- a/Core/MIPS/MIPSTables.cpp +++ b/Core/MIPS/MIPSTables.cpp @@ -631,20 +631,20 @@ const MIPSInstruction tableVFPU6[32] = //111100 xxx INSTR("vmmul",&Jit::Comp_Vmmul, Dis_MatrixMult, Int_Vmmul, IS_VFPU|OUT_EAT_PREFIX), INSTR("vmmul",&Jit::Comp_Vmmul, Dis_MatrixMult, Int_Vmmul, IS_VFPU|OUT_EAT_PREFIX), - INSTR("v(h)tfm2",&Jit::Comp_Generic, Dis_Vtfm, Int_Vtfm, IS_VFPU|OUT_EAT_PREFIX), - INSTR("v(h)tfm2",&Jit::Comp_Generic, Dis_Vtfm, Int_Vtfm, IS_VFPU|OUT_EAT_PREFIX), - INSTR("v(h)tfm2",&Jit::Comp_Generic, Dis_Vtfm, Int_Vtfm, IS_VFPU|OUT_EAT_PREFIX), - INSTR("v(h)tfm2",&Jit::Comp_Generic, Dis_Vtfm, Int_Vtfm, IS_VFPU|OUT_EAT_PREFIX), + INSTR("v(h)tfm2",&Jit::Comp_Vtfm, Dis_Vtfm, Int_Vtfm, IS_VFPU|OUT_EAT_PREFIX), + INSTR("v(h)tfm2",&Jit::Comp_Vtfm, Dis_Vtfm, Int_Vtfm, IS_VFPU|OUT_EAT_PREFIX), + INSTR("v(h)tfm2",&Jit::Comp_Vtfm, Dis_Vtfm, Int_Vtfm, IS_VFPU|OUT_EAT_PREFIX), + INSTR("v(h)tfm2",&Jit::Comp_Vtfm, Dis_Vtfm, Int_Vtfm, IS_VFPU|OUT_EAT_PREFIX), //8 - INSTR("v(h)tfm3",&Jit::Comp_Generic, Dis_Vtfm, Int_Vtfm, IS_VFPU|OUT_EAT_PREFIX), - INSTR("v(h)tfm3",&Jit::Comp_Generic, Dis_Vtfm, Int_Vtfm, IS_VFPU|OUT_EAT_PREFIX), - INSTR("v(h)tfm3",&Jit::Comp_Generic, Dis_Vtfm, Int_Vtfm, IS_VFPU|OUT_EAT_PREFIX), - INSTR("v(h)tfm3",&Jit::Comp_Generic, Dis_Vtfm, Int_Vtfm, IS_VFPU|OUT_EAT_PREFIX), + INSTR("v(h)tfm3",&Jit::Comp_Vtfm, Dis_Vtfm, Int_Vtfm, IS_VFPU|OUT_EAT_PREFIX), + INSTR("v(h)tfm3",&Jit::Comp_Vtfm, Dis_Vtfm, Int_Vtfm, IS_VFPU|OUT_EAT_PREFIX), + INSTR("v(h)tfm3",&Jit::Comp_Vtfm, Dis_Vtfm, Int_Vtfm, IS_VFPU|OUT_EAT_PREFIX), + INSTR("v(h)tfm3",&Jit::Comp_Vtfm, Dis_Vtfm, Int_Vtfm, IS_VFPU|OUT_EAT_PREFIX), - INSTR("v(h)tfm4",&Jit::Comp_Generic, Dis_Vtfm, Int_Vtfm, IS_VFPU|OUT_EAT_PREFIX), - INSTR("v(h)tfm4",&Jit::Comp_Generic, Dis_Vtfm, Int_Vtfm, IS_VFPU|OUT_EAT_PREFIX), - INSTR("v(h)tfm4",&Jit::Comp_Generic, Dis_Vtfm, Int_Vtfm, IS_VFPU|OUT_EAT_PREFIX), - INSTR("v(h)tfm4",&Jit::Comp_Generic, Dis_Vtfm, Int_Vtfm, IS_VFPU|OUT_EAT_PREFIX), + INSTR("v(h)tfm4",&Jit::Comp_Vtfm, Dis_Vtfm, Int_Vtfm, IS_VFPU|OUT_EAT_PREFIX), + INSTR("v(h)tfm4",&Jit::Comp_Vtfm, Dis_Vtfm, Int_Vtfm, IS_VFPU|OUT_EAT_PREFIX), + INSTR("v(h)tfm4",&Jit::Comp_Vtfm, Dis_Vtfm, Int_Vtfm, IS_VFPU|OUT_EAT_PREFIX), + INSTR("v(h)tfm4",&Jit::Comp_Vtfm, Dis_Vtfm, Int_Vtfm, IS_VFPU|OUT_EAT_PREFIX), //16 INSTR("vmscl",&Jit::Comp_Vmscl, Dis_Generic, Int_Vmscl, IS_VFPU|OUT_EAT_PREFIX), INSTR("vmscl",&Jit::Comp_Vmscl, Dis_Generic, Int_Vmscl, IS_VFPU|OUT_EAT_PREFIX), diff --git a/Core/MIPS/x86/CompVFPU.cpp b/Core/MIPS/x86/CompVFPU.cpp index 86cd1a013b..c371ba6915 100644 --- a/Core/MIPS/x86/CompVFPU.cpp +++ b/Core/MIPS/x86/CompVFPU.cpp @@ -876,4 +876,62 @@ void Jit::Comp_Vmscl(u32 op) { fpr.ReleaseSpillLocks(); } +void Jit::Comp_Vtfm(u32 op) { + CONDITIONAL_DISABLE; + + // TODO: This probably ignores prefixes? Or maybe uses D? + if (js.MayHavePrefix()) + DISABLE; + + VectorSize sz = GetVecSize(op); + MatrixSize msz = GetMtxSize(op); + int n = GetNumVectorElements(sz); + int ins = (op >> 23) & 7; + + bool homogenous = false; + if (n == ins) + { + n++; + sz = (VectorSize)((int)(sz) + 1); + msz = (MatrixSize)((int)(msz) + 1); + homogenous = true; + } + // Otherwise, n should already be ins + 1. + else if (n != ins + 1) + DISABLE; + + u8 sregs[16], dregs[4], tregs[4]; + GetMatrixRegs(sregs, msz, _VS); + GetVectorRegs(tregs, sz, _VT); + GetVectorRegs(dregs, sz, _VD); + + // TODO: test overlap, optimize. + u8 tempregs[4]; + for (int i = 0; i < n; i++) + { + XORPS(XMM0, R(XMM0)); + for (int k = 0; k < n; k++) + { + MOVSS(XMM1, fpr.V(sregs[i * 4 + k])); + if (!homogenous || k != n - 1) + MULSS(XMM1, fpr.V(tregs[k])); + ADDSS(XMM0, R(XMM1)); + } + + u8 temp = (u8) fpr.GetTempV(); + fpr.MapRegV(temp, MAP_NOINIT | MAP_DIRTY); + MOVSS(fpr.VX(temp), R(XMM0)); + fpr.StoreFromRegisterV(temp); + tempregs[i] = temp; + } + for (int i = 0; i < n; i++) + { + u8 temp = tempregs[i]; + fpr.MapRegV(temp, 0); + MOVSS(fpr.V(dregs[i]), fpr.VX(temp)); + } + + fpr.ReleaseSpillLocks(); +} + } diff --git a/Core/MIPS/x86/Jit.h b/Core/MIPS/x86/Jit.h index fe647a54ec..2590ac20c3 100644 --- a/Core/MIPS/x86/Jit.h +++ b/Core/MIPS/x86/Jit.h @@ -206,6 +206,7 @@ public: void Comp_VScl(u32 op); void Comp_Vmmul(u32 op); void Comp_Vmscl(u32 op); + void Comp_Vtfm(u32 op); void Comp_DoNothing(u32 op);