diff options
Diffstat (limited to 'llvm/lib/Target/R600')
| -rw-r--r-- | llvm/lib/Target/R600/AMDGPUISelLowering.cpp | 5 | ||||
| -rw-r--r-- | llvm/lib/Target/R600/AMDGPUInstrInfo.td | 7 | ||||
| -rw-r--r-- | llvm/lib/Target/R600/SIInstrInfo.td | 22 | ||||
| -rw-r--r-- | llvm/lib/Target/R600/SIInstructions.td | 19 |
4 files changed, 46 insertions, 7 deletions
diff --git a/llvm/lib/Target/R600/AMDGPUISelLowering.cpp b/llvm/lib/Target/R600/AMDGPUISelLowering.cpp index d577d689f66..d96f03aaeb0 100644 --- a/llvm/lib/Target/R600/AMDGPUISelLowering.cpp +++ b/llvm/lib/Target/R600/AMDGPUISelLowering.cpp @@ -907,10 +907,9 @@ SDValue AMDGPUTargetLowering::LowerINTRINSIC_WO_CHAIN(SDValue Op, } case Intrinsic::AMDGPU_div_fmas: - // FIXME: Dropping bool parameter. Work is needed to support the implicit - // read from VCC. return DAG.getNode(AMDGPUISD::DIV_FMAS, DL, VT, - Op.getOperand(1), Op.getOperand(2), Op.getOperand(3)); + Op.getOperand(1), Op.getOperand(2), Op.getOperand(3), + Op.getOperand(4)); case Intrinsic::AMDGPU_div_fixup: return DAG.getNode(AMDGPUISD::DIV_FIXUP, DL, VT, diff --git a/llvm/lib/Target/R600/AMDGPUInstrInfo.td b/llvm/lib/Target/R600/AMDGPUInstrInfo.td index 0e34392bd50..d657ad05c8c 100644 --- a/llvm/lib/Target/R600/AMDGPUInstrInfo.td +++ b/llvm/lib/Target/R600/AMDGPUInstrInfo.td @@ -35,6 +35,11 @@ def AMDGPUDivScaleOp : SDTypeProfile<2, 3, [SDTCisFP<0>, SDTCisInt<1>, SDTCisSameAs<0, 2>, SDTCisSameAs<0, 3>, SDTCisSameAs<0, 4>] >; +// float, float, float, vcc +def AMDGPUFmasOp : SDTypeProfile<1, 4, + [SDTCisFP<0>, SDTCisSameAs<0, 1>, SDTCisSameAs<0, 2>, SDTCisSameAs<0, 3>, SDTCisInt<4>] +>; + //===----------------------------------------------------------------------===// // AMDGPU DAG Nodes // @@ -153,7 +158,7 @@ def AMDGPUdiv_scale : SDNode<"AMDGPUISD::DIV_SCALE", AMDGPUDivScaleOp>; // Special case divide FMA with scale and flags (src0 = Quotient, // src1 = Denominator, src2 = Numerator). -def AMDGPUdiv_fmas : SDNode<"AMDGPUISD::DIV_FMAS", SDTFPTernaryOp>; +def AMDGPUdiv_fmas : SDNode<"AMDGPUISD::DIV_FMAS", AMDGPUFmasOp>; // Single or double precision division fixup. // Special case divide fixup and flags(src0 = Quotient, src1 = diff --git a/llvm/lib/Target/R600/SIInstrInfo.td b/llvm/lib/Target/R600/SIInstrInfo.td index cb16be5f2f7..4082b570c41 100644 --- a/llvm/lib/Target/R600/SIInstrInfo.td +++ b/llvm/lib/Target/R600/SIInstrInfo.td @@ -1300,6 +1300,28 @@ multiclass VOP3Inst <vop3 op, string opName, VOPProfile P, P.NumSrcArgs, P.HasModifiers >; +// Special case for v_div_fmas_{f32|f64}, since it seems to be the +// only VOP instruction that implicitly reads VCC. +multiclass VOP3_VCC_Inst <vop3 op, string opName, + VOPProfile P, + SDPatternOperator node = null_frag> : VOP3_Helper < + op, opName, + P.Outs, + (ins InputModsNoDefault:$src0_modifiers, P.Src0RC64:$src0, + InputModsNoDefault:$src1_modifiers, P.Src1RC64:$src1, + InputModsNoDefault:$src2_modifiers, P.Src2RC64:$src2, + ClampMod:$clamp, + omod:$omod), + " $dst, $src0_modifiers, $src1_modifiers, $src2_modifiers"#"$clamp"#"$omod", + [(set P.DstVT:$dst, + (node (P.Src0VT (VOP3Mods0 P.Src0VT:$src0, i32:$src0_modifiers, + i1:$clamp, i32:$omod)), + (P.Src1VT (VOP3Mods P.Src1VT:$src1, i32:$src1_modifiers)), + (P.Src2VT (VOP3Mods P.Src2VT:$src2, i32:$src2_modifiers)), + (i1 VCC)))], + 3, 1 +>; + multiclass VOP3b_Helper <vop op, RegisterClass vrc, RegisterOperand arc, string opName, list<dag> pattern> : VOP3b_3_m < diff --git a/llvm/lib/Target/R600/SIInstructions.td b/llvm/lib/Target/R600/SIInstructions.td index 032d6c2abfc..0f14dc31136 100644 --- a/llvm/lib/Target/R600/SIInstructions.td +++ b/llvm/lib/Target/R600/SIInstructions.td @@ -1744,14 +1744,27 @@ let SchedRW = [WriteDouble, WriteSALU] in { defm V_DIV_SCALE_F64 : VOP3b_64 <vop3<0x16e, 0x1e1>, "v_div_scale_f64", []>; } // let SchedRW = [WriteDouble] -let isCommutable = 1 in { -defm V_DIV_FMAS_F32 : VOP3Inst <vop3<0x16f, 0x1e2>, "v_div_fmas_f32", +let isCommutable = 1, Uses = [VCC] in { + +// v_div_fmas_f32: +// result = src0 * src1 + src2 +// if (vcc) +// result *= 2^32 +// +defm V_DIV_FMAS_F32 : VOP3_VCC_Inst <vop3<0x16f, 0x1e2>, "v_div_fmas_f32", VOP_F32_F32_F32_F32, AMDGPUdiv_fmas >; + let SchedRW = [WriteDouble] in { -defm V_DIV_FMAS_F64 : VOP3Inst <vop3<0x170, 0x1e3>, "v_div_fmas_f64", +// v_div_fmas_f64: +// result = src0 * src1 + src2 +// if (vcc) +// result *= 2^64 +// +defm V_DIV_FMAS_F64 : VOP3_VCC_Inst <vop3<0x170, 0x1e3>, "v_div_fmas_f64", VOP_F64_F64_F64_F64, AMDGPUdiv_fmas >; + } // End SchedRW = [WriteDouble] } // End isCommutable = 1 |
