diff --git a/llvm/lib/Target/RISCV/RISCVISelLowering.h b/llvm/lib/Target/RISCV/RISCVISelLowering.h --- a/llvm/lib/Target/RISCV/RISCVISelLowering.h +++ b/llvm/lib/Target/RISCV/RISCVISelLowering.h @@ -194,6 +194,8 @@ SMAX_VL, UMIN_VL, UMAX_VL, + FMINNUM_VL, + FMAXNUM_VL, MULHS_VL, MULHU_VL, FP_TO_SINT_VL, diff --git a/llvm/lib/Target/RISCV/RISCVISelLowering.cpp b/llvm/lib/Target/RISCV/RISCVISelLowering.cpp --- a/llvm/lib/Target/RISCV/RISCVISelLowering.cpp +++ b/llvm/lib/Target/RISCV/RISCVISelLowering.cpp @@ -555,6 +555,9 @@ for (auto CC : VFPCCToExpand) setCondCodeAction(CC, VT, Expand); + setOperationAction(ISD::FMINNUM, VT, Legal); + setOperationAction(ISD::FMAXNUM, VT, Legal); + setOperationAction(ISD::VECREDUCE_FADD, VT, Custom); setOperationAction(ISD::VECREDUCE_SEQ_FADD, VT, Custom); setOperationAction(ISD::FCOPYSIGN, VT, Legal); @@ -729,6 +732,8 @@ setOperationAction(ISD::FCOPYSIGN, VT, Custom); setOperationAction(ISD::FSQRT, VT, Custom); setOperationAction(ISD::FMA, VT, Custom); + setOperationAction(ISD::FMINNUM, VT, Custom); + setOperationAction(ISD::FMAXNUM, VT, Custom); setOperationAction(ISD::FP_ROUND, VT, Custom); setOperationAction(ISD::FP_EXTEND, VT, Custom); @@ -2205,6 +2210,10 @@ return lowerToScalableOp(Op, DAG, RISCVISD::UMIN_VL); case ISD::UMAX: return lowerToScalableOp(Op, DAG, RISCVISD::UMAX_VL); + case ISD::FMINNUM: + return lowerToScalableOp(Op, DAG, RISCVISD::FMINNUM_VL); + case ISD::FMAXNUM: + return lowerToScalableOp(Op, DAG, RISCVISD::FMAXNUM_VL); case ISD::ABS: return lowerABS(Op, DAG); case ISD::VSELECT: @@ -7457,6 +7466,8 @@ NODE_NAME_CASE(SMAX_VL) NODE_NAME_CASE(UMIN_VL) NODE_NAME_CASE(UMAX_VL) + NODE_NAME_CASE(FMINNUM_VL) + NODE_NAME_CASE(FMAXNUM_VL) NODE_NAME_CASE(MULHS_VL) NODE_NAME_CASE(MULHU_VL) NODE_NAME_CASE(FP_TO_SINT_VL) diff --git a/llvm/lib/Target/RISCV/RISCVInstrInfoVSDPatterns.td b/llvm/lib/Target/RISCV/RISCVInstrInfoVSDPatterns.td --- a/llvm/lib/Target/RISCV/RISCVInstrInfoVSDPatterns.td +++ b/llvm/lib/Target/RISCV/RISCVInstrInfoVSDPatterns.td @@ -635,6 +635,10 @@ vti.RegClass:$rs1, vti.ScalarRegClass:$rs2, vti.AVL, vti.SEW)>; } +// 14.11. Vector Floating-Point MIN/MAX Instructions +defm : VPatBinaryFPSDNode_VV_VF; +defm : VPatBinaryFPSDNode_VV_VF; + // 14.13. Vector Floating-Point Compare Instructions defm : VPatFPSetCCSDNode_VV_VF_FV; defm : VPatFPSetCCSDNode_VV_VF_FV; diff --git a/llvm/lib/Target/RISCV/RISCVInstrInfoVVLPatterns.td b/llvm/lib/Target/RISCV/RISCVInstrInfoVVLPatterns.td --- a/llvm/lib/Target/RISCV/RISCVInstrInfoVVLPatterns.td +++ b/llvm/lib/Target/RISCV/RISCVInstrInfoVVLPatterns.td @@ -97,6 +97,8 @@ def riscv_fabs_vl : SDNode<"RISCVISD::FABS_VL", SDT_RISCVFPUnOp_VL>; def riscv_fsqrt_vl : SDNode<"RISCVISD::FSQRT_VL", SDT_RISCVFPUnOp_VL>; def riscv_fcopysign_vl : SDNode<"RISCVISD::FCOPYSIGN_VL", SDT_RISCVFPBinOp_VL>; +def riscv_fminnum_vl : SDNode<"RISCVISD::FMINNUM_VL", SDT_RISCVFPBinOp_VL>; +def riscv_fmaxnum_vl : SDNode<"RISCVISD::FMAXNUM_VL", SDT_RISCVFPBinOp_VL>; def SDT_RISCVVecFMA_VL : SDTypeProfile<1, 5, [SDTCisSameAs<0, 1>, SDTCisSameAs<0, 2>, @@ -856,6 +858,10 @@ GPR:$vl, vti.SEW)>; } +// 14.11. Vector Floating-Point MIN/MAX Instructions +defm : VPatBinaryFPVL_VV_VF; +defm : VPatBinaryFPVL_VV_VF; + // 14.13. Vector Floating-Point Compare Instructions defm : VPatFPSetCCVL_VV_VF_FV; defm : VPatFPSetCCVL_VV_VF_FV; diff --git a/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-vfmax.ll b/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-vfmax.ll new file mode 100644 --- /dev/null +++ b/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-vfmax.ll @@ -0,0 +1,293 @@ +; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py +; RUN: llc -mtriple=riscv32 -mattr=+d,+experimental-zfh,+experimental-v -target-abi=ilp32d \ +; RUN: -riscv-v-vector-bits-min=128 -verify-machineinstrs < %s | FileCheck %s +; RUN: llc -mtriple=riscv64 -mattr=+d,+experimental-zfh,+experimental-v -target-abi=lp64d \ +; RUN: -riscv-v-vector-bits-min=128 -verify-machineinstrs < %s | FileCheck %s + +declare <2 x half> @llvm.maxnum.v2f16(<2 x half>, <2 x half>) + +define <2 x half> @vfmax_v2f16_vv(<2 x half> %a, <2 x half> %b) { +; CHECK-LABEL: vfmax_v2f16_vv: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetivli a0, 2, e16,m1,ta,mu +; CHECK-NEXT: vfmax.vv v8, v8, v9 +; CHECK-NEXT: ret + %v = call <2 x half> @llvm.maxnum.v2f16(<2 x half> %a, <2 x half> %b) + ret <2 x half> %v +} + +define <2 x half> @vfmax_v2f16_vf(<2 x half> %a, half %b) { +; CHECK-LABEL: vfmax_v2f16_vf: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetivli a0, 2, e16,m1,ta,mu +; CHECK-NEXT: vfmax.vf v8, v8, fa0 +; CHECK-NEXT: ret + %head = insertelement <2 x half> undef, half %b, i32 0 + %splat = shufflevector <2 x half> %head, <2 x half> undef, <2 x i32> zeroinitializer + %v = call <2 x half> @llvm.maxnum.v2f16(<2 x half> %a, <2 x half> %splat) + ret <2 x half> %v +} + +declare <4 x half> @llvm.maxnum.v4f16(<4 x half>, <4 x half>) + +define <4 x half> @vfmax_v4f16_vv(<4 x half> %a, <4 x half> %b) { +; CHECK-LABEL: vfmax_v4f16_vv: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetivli a0, 4, e16,m1,ta,mu +; CHECK-NEXT: vfmax.vv v8, v8, v9 +; CHECK-NEXT: ret + %v = call <4 x half> @llvm.maxnum.v4f16(<4 x half> %a, <4 x half> %b) + ret <4 x half> %v +} + +define <4 x half> @vfmax_v4f16_vf(<4 x half> %a, half %b) { +; CHECK-LABEL: vfmax_v4f16_vf: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetivli a0, 4, e16,m1,ta,mu +; CHECK-NEXT: vfmax.vf v8, v8, fa0 +; CHECK-NEXT: ret + %head = insertelement <4 x half> undef, half %b, i32 0 + %splat = shufflevector <4 x half> %head, <4 x half> undef, <4 x i32> zeroinitializer + %v = call <4 x half> @llvm.maxnum.v4f16(<4 x half> %a, <4 x half> %splat) + ret <4 x half> %v +} + +declare <8 x half> @llvm.maxnum.v8f16(<8 x half>, <8 x half>) + +define <8 x half> @vfmax_v8f16_vv(<8 x half> %a, <8 x half> %b) { +; CHECK-LABEL: vfmax_v8f16_vv: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetivli a0, 8, e16,m1,ta,mu +; CHECK-NEXT: vfmax.vv v8, v8, v9 +; CHECK-NEXT: ret + %v = call <8 x half> @llvm.maxnum.v8f16(<8 x half> %a, <8 x half> %b) + ret <8 x half> %v +} + +define <8 x half> @vfmax_v8f16_vf(<8 x half> %a, half %b) { +; CHECK-LABEL: vfmax_v8f16_vf: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetivli a0, 8, e16,m1,ta,mu +; CHECK-NEXT: vfmax.vf v8, v8, fa0 +; CHECK-NEXT: ret + %head = insertelement <8 x half> undef, half %b, i32 0 + %splat = shufflevector <8 x half> %head, <8 x half> undef, <8 x i32> zeroinitializer + %v = call <8 x half> @llvm.maxnum.v8f16(<8 x half> %a, <8 x half> %splat) + ret <8 x half> %v +} + +declare <16 x half> @llvm.maxnum.v16f16(<16 x half>, <16 x half>) + +define <16 x half> @vfmax_v16f16_vv(<16 x half> %a, <16 x half> %b) { +; CHECK-LABEL: vfmax_v16f16_vv: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetivli a0, 16, e16,m2,ta,mu +; CHECK-NEXT: vfmax.vv v8, v8, v10 +; CHECK-NEXT: ret + %v = call <16 x half> @llvm.maxnum.v16f16(<16 x half> %a, <16 x half> %b) + ret <16 x half> %v +} + +define <16 x half> @vfmax_v16f16_vf(<16 x half> %a, half %b) { +; CHECK-LABEL: vfmax_v16f16_vf: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetivli a0, 16, e16,m2,ta,mu +; CHECK-NEXT: vfmax.vf v8, v8, fa0 +; CHECK-NEXT: ret + %head = insertelement <16 x half> undef, half %b, i32 0 + %splat = shufflevector <16 x half> %head, <16 x half> undef, <16 x i32> zeroinitializer + %v = call <16 x half> @llvm.maxnum.v16f16(<16 x half> %a, <16 x half> %splat) + ret <16 x half> %v +} + +declare <2 x float> @llvm.maxnum.v2f32(<2 x float>, <2 x float>) + +define <2 x float> @vfmax_v2f32_vv(<2 x float> %a, <2 x float> %b) { +; CHECK-LABEL: vfmax_v2f32_vv: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetivli a0, 2, e32,m1,ta,mu +; CHECK-NEXT: vfmax.vv v8, v8, v9 +; CHECK-NEXT: ret + %v = call <2 x float> @llvm.maxnum.v2f32(<2 x float> %a, <2 x float> %b) + ret <2 x float> %v +} + +define <2 x float> @vfmax_v2f32_vf(<2 x float> %a, float %b) { +; CHECK-LABEL: vfmax_v2f32_vf: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetivli a0, 2, e32,m1,ta,mu +; CHECK-NEXT: vfmax.vf v8, v8, fa0 +; CHECK-NEXT: ret + %head = insertelement <2 x float> undef, float %b, i32 0 + %splat = shufflevector <2 x float> %head, <2 x float> undef, <2 x i32> zeroinitializer + %v = call <2 x float> @llvm.maxnum.v2f32(<2 x float> %a, <2 x float> %splat) + ret <2 x float> %v +} + +declare <4 x float> @llvm.maxnum.v4f32(<4 x float>, <4 x float>) + +define <4 x float> @vfmax_v4f32_vv(<4 x float> %a, <4 x float> %b) { +; CHECK-LABEL: vfmax_v4f32_vv: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetivli a0, 4, e32,m1,ta,mu +; CHECK-NEXT: vfmax.vv v8, v8, v9 +; CHECK-NEXT: ret + %v = call <4 x float> @llvm.maxnum.v4f32(<4 x float> %a, <4 x float> %b) + ret <4 x float> %v +} + +define <4 x float> @vfmax_v4f32_vf(<4 x float> %a, float %b) { +; CHECK-LABEL: vfmax_v4f32_vf: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetivli a0, 4, e32,m1,ta,mu +; CHECK-NEXT: vfmax.vf v8, v8, fa0 +; CHECK-NEXT: ret + %head = insertelement <4 x float> undef, float %b, i32 0 + %splat = shufflevector <4 x float> %head, <4 x float> undef, <4 x i32> zeroinitializer + %v = call <4 x float> @llvm.maxnum.v4f32(<4 x float> %a, <4 x float> %splat) + ret <4 x float> %v +} + +declare <8 x float> @llvm.maxnum.v8f32(<8 x float>, <8 x float>) + +define <8 x float> @vfmax_v8f32_vv(<8 x float> %a, <8 x float> %b) { +; CHECK-LABEL: vfmax_v8f32_vv: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetivli a0, 8, e32,m2,ta,mu +; CHECK-NEXT: vfmax.vv v8, v8, v10 +; CHECK-NEXT: ret + %v = call <8 x float> @llvm.maxnum.v8f32(<8 x float> %a, <8 x float> %b) + ret <8 x float> %v +} + +define <8 x float> @vfmax_v8f32_vf(<8 x float> %a, float %b) { +; CHECK-LABEL: vfmax_v8f32_vf: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetivli a0, 8, e32,m2,ta,mu +; CHECK-NEXT: vfmax.vf v8, v8, fa0 +; CHECK-NEXT: ret + %head = insertelement <8 x float> undef, float %b, i32 0 + %splat = shufflevector <8 x float> %head, <8 x float> undef, <8 x i32> zeroinitializer + %v = call <8 x float> @llvm.maxnum.v8f32(<8 x float> %a, <8 x float> %splat) + ret <8 x float> %v +} + +declare <16 x float> @llvm.maxnum.v16f32(<16 x float>, <16 x float>) + +define <16 x float> @vfmax_v16f32_vv(<16 x float> %a, <16 x float> %b) { +; CHECK-LABEL: vfmax_v16f32_vv: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetivli a0, 16, e32,m4,ta,mu +; CHECK-NEXT: vfmax.vv v8, v8, v12 +; CHECK-NEXT: ret + %v = call <16 x float> @llvm.maxnum.v16f32(<16 x float> %a, <16 x float> %b) + ret <16 x float> %v +} + +define <16 x float> @vfmax_v16f32_vf(<16 x float> %a, float %b) { +; CHECK-LABEL: vfmax_v16f32_vf: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetivli a0, 16, e32,m4,ta,mu +; CHECK-NEXT: vfmax.vf v8, v8, fa0 +; CHECK-NEXT: ret + %head = insertelement <16 x float> undef, float %b, i32 0 + %splat = shufflevector <16 x float> %head, <16 x float> undef, <16 x i32> zeroinitializer + %v = call <16 x float> @llvm.maxnum.v16f32(<16 x float> %a, <16 x float> %splat) + ret <16 x float> %v +} + +declare <2 x double> @llvm.maxnum.v2f64(<2 x double>, <2 x double>) + +define <2 x double> @vfmax_v2f64_vv(<2 x double> %a, <2 x double> %b) { +; CHECK-LABEL: vfmax_v2f64_vv: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetivli a0, 2, e64,m1,ta,mu +; CHECK-NEXT: vfmax.vv v8, v8, v9 +; CHECK-NEXT: ret + %v = call <2 x double> @llvm.maxnum.v2f64(<2 x double> %a, <2 x double> %b) + ret <2 x double> %v +} + +define <2 x double> @vfmax_v2f64_vf(<2 x double> %a, double %b) { +; CHECK-LABEL: vfmax_v2f64_vf: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetivli a0, 2, e64,m1,ta,mu +; CHECK-NEXT: vfmax.vf v8, v8, fa0 +; CHECK-NEXT: ret + %head = insertelement <2 x double> undef, double %b, i32 0 + %splat = shufflevector <2 x double> %head, <2 x double> undef, <2 x i32> zeroinitializer + %v = call <2 x double> @llvm.maxnum.v2f64(<2 x double> %a, <2 x double> %splat) + ret <2 x double> %v +} + +declare <4 x double> @llvm.maxnum.v4f64(<4 x double>, <4 x double>) + +define <4 x double> @vfmax_v4f64_vv(<4 x double> %a, <4 x double> %b) { +; CHECK-LABEL: vfmax_v4f64_vv: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetivli a0, 4, e64,m2,ta,mu +; CHECK-NEXT: vfmax.vv v8, v8, v10 +; CHECK-NEXT: ret + %v = call <4 x double> @llvm.maxnum.v4f64(<4 x double> %a, <4 x double> %b) + ret <4 x double> %v +} + +define <4 x double> @vfmax_v4f64_vf(<4 x double> %a, double %b) { +; CHECK-LABEL: vfmax_v4f64_vf: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetivli a0, 4, e64,m2,ta,mu +; CHECK-NEXT: vfmax.vf v8, v8, fa0 +; CHECK-NEXT: ret + %head = insertelement <4 x double> undef, double %b, i32 0 + %splat = shufflevector <4 x double> %head, <4 x double> undef, <4 x i32> zeroinitializer + %v = call <4 x double> @llvm.maxnum.v4f64(<4 x double> %a, <4 x double> %splat) + ret <4 x double> %v +} + +declare <8 x double> @llvm.maxnum.v8f64(<8 x double>, <8 x double>) + +define <8 x double> @vfmax_v8f64_vv(<8 x double> %a, <8 x double> %b) { +; CHECK-LABEL: vfmax_v8f64_vv: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetivli a0, 8, e64,m4,ta,mu +; CHECK-NEXT: vfmax.vv v8, v8, v12 +; CHECK-NEXT: ret + %v = call <8 x double> @llvm.maxnum.v8f64(<8 x double> %a, <8 x double> %b) + ret <8 x double> %v +} + +define <8 x double> @vfmax_v8f64_vf(<8 x double> %a, double %b) { +; CHECK-LABEL: vfmax_v8f64_vf: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetivli a0, 8, e64,m4,ta,mu +; CHECK-NEXT: vfmax.vf v8, v8, fa0 +; CHECK-NEXT: ret + %head = insertelement <8 x double> undef, double %b, i32 0 + %splat = shufflevector <8 x double> %head, <8 x double> undef, <8 x i32> zeroinitializer + %v = call <8 x double> @llvm.maxnum.v8f64(<8 x double> %a, <8 x double> %splat) + ret <8 x double> %v +} + +declare <16 x double> @llvm.maxnum.v16f64(<16 x double>, <16 x double>) + +define <16 x double> @vfmax_v16f64_vv(<16 x double> %a, <16 x double> %b) { +; CHECK-LABEL: vfmax_v16f64_vv: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetivli a0, 16, e64,m8,ta,mu +; CHECK-NEXT: vfmax.vv v8, v8, v16 +; CHECK-NEXT: ret + %v = call <16 x double> @llvm.maxnum.v16f64(<16 x double> %a, <16 x double> %b) + ret <16 x double> %v +} + +define <16 x double> @vfmax_v16f64_vf(<16 x double> %a, double %b) { +; CHECK-LABEL: vfmax_v16f64_vf: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetivli a0, 16, e64,m8,ta,mu +; CHECK-NEXT: vfmax.vf v8, v8, fa0 +; CHECK-NEXT: ret + %head = insertelement <16 x double> undef, double %b, i32 0 + %splat = shufflevector <16 x double> %head, <16 x double> undef, <16 x i32> zeroinitializer + %v = call <16 x double> @llvm.maxnum.v16f64(<16 x double> %a, <16 x double> %splat) + ret <16 x double> %v +} diff --git a/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-vfmin.ll b/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-vfmin.ll new file mode 100644 --- /dev/null +++ b/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-vfmin.ll @@ -0,0 +1,293 @@ +; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py +; RUN: llc -mtriple=riscv32 -mattr=+d,+experimental-zfh,+experimental-v -target-abi=ilp32d \ +; RUN: -riscv-v-vector-bits-min=128 -verify-machineinstrs < %s | FileCheck %s +; RUN: llc -mtriple=riscv64 -mattr=+d,+experimental-zfh,+experimental-v -target-abi=lp64d \ +; RUN: -riscv-v-vector-bits-min=128 -verify-machineinstrs < %s | FileCheck %s + +declare <2 x half> @llvm.minnum.v2f16(<2 x half>, <2 x half>) + +define <2 x half> @vfmin_v2f16_vv(<2 x half> %a, <2 x half> %b) { +; CHECK-LABEL: vfmin_v2f16_vv: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetivli a0, 2, e16,m1,ta,mu +; CHECK-NEXT: vfmin.vv v8, v8, v9 +; CHECK-NEXT: ret + %v = call <2 x half> @llvm.minnum.v2f16(<2 x half> %a, <2 x half> %b) + ret <2 x half> %v +} + +define <2 x half> @vfmin_v2f16_vf(<2 x half> %a, half %b) { +; CHECK-LABEL: vfmin_v2f16_vf: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetivli a0, 2, e16,m1,ta,mu +; CHECK-NEXT: vfmin.vf v8, v8, fa0 +; CHECK-NEXT: ret + %head = insertelement <2 x half> undef, half %b, i32 0 + %splat = shufflevector <2 x half> %head, <2 x half> undef, <2 x i32> zeroinitializer + %v = call <2 x half> @llvm.minnum.v2f16(<2 x half> %a, <2 x half> %splat) + ret <2 x half> %v +} + +declare <4 x half> @llvm.minnum.v4f16(<4 x half>, <4 x half>) + +define <4 x half> @vfmin_v4f16_vv(<4 x half> %a, <4 x half> %b) { +; CHECK-LABEL: vfmin_v4f16_vv: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetivli a0, 4, e16,m1,ta,mu +; CHECK-NEXT: vfmin.vv v8, v8, v9 +; CHECK-NEXT: ret + %v = call <4 x half> @llvm.minnum.v4f16(<4 x half> %a, <4 x half> %b) + ret <4 x half> %v +} + +define <4 x half> @vfmin_v4f16_vf(<4 x half> %a, half %b) { +; CHECK-LABEL: vfmin_v4f16_vf: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetivli a0, 4, e16,m1,ta,mu +; CHECK-NEXT: vfmin.vf v8, v8, fa0 +; CHECK-NEXT: ret + %head = insertelement <4 x half> undef, half %b, i32 0 + %splat = shufflevector <4 x half> %head, <4 x half> undef, <4 x i32> zeroinitializer + %v = call <4 x half> @llvm.minnum.v4f16(<4 x half> %a, <4 x half> %splat) + ret <4 x half> %v +} + +declare <8 x half> @llvm.minnum.v8f16(<8 x half>, <8 x half>) + +define <8 x half> @vfmin_v8f16_vv(<8 x half> %a, <8 x half> %b) { +; CHECK-LABEL: vfmin_v8f16_vv: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetivli a0, 8, e16,m1,ta,mu +; CHECK-NEXT: vfmin.vv v8, v8, v9 +; CHECK-NEXT: ret + %v = call <8 x half> @llvm.minnum.v8f16(<8 x half> %a, <8 x half> %b) + ret <8 x half> %v +} + +define <8 x half> @vfmin_v8f16_vf(<8 x half> %a, half %b) { +; CHECK-LABEL: vfmin_v8f16_vf: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetivli a0, 8, e16,m1,ta,mu +; CHECK-NEXT: vfmin.vf v8, v8, fa0 +; CHECK-NEXT: ret + %head = insertelement <8 x half> undef, half %b, i32 0 + %splat = shufflevector <8 x half> %head, <8 x half> undef, <8 x i32> zeroinitializer + %v = call <8 x half> @llvm.minnum.v8f16(<8 x half> %a, <8 x half> %splat) + ret <8 x half> %v +} + +declare <16 x half> @llvm.minnum.v16f16(<16 x half>, <16 x half>) + +define <16 x half> @vfmin_v16f16_vv(<16 x half> %a, <16 x half> %b) { +; CHECK-LABEL: vfmin_v16f16_vv: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetivli a0, 16, e16,m2,ta,mu +; CHECK-NEXT: vfmin.vv v8, v8, v10 +; CHECK-NEXT: ret + %v = call <16 x half> @llvm.minnum.v16f16(<16 x half> %a, <16 x half> %b) + ret <16 x half> %v +} + +define <16 x half> @vfmin_v16f16_vf(<16 x half> %a, half %b) { +; CHECK-LABEL: vfmin_v16f16_vf: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetivli a0, 16, e16,m2,ta,mu +; CHECK-NEXT: vfmin.vf v8, v8, fa0 +; CHECK-NEXT: ret + %head = insertelement <16 x half> undef, half %b, i32 0 + %splat = shufflevector <16 x half> %head, <16 x half> undef, <16 x i32> zeroinitializer + %v = call <16 x half> @llvm.minnum.v16f16(<16 x half> %a, <16 x half> %splat) + ret <16 x half> %v +} + +declare <2 x float> @llvm.minnum.v2f32(<2 x float>, <2 x float>) + +define <2 x float> @vfmin_v2f32_vv(<2 x float> %a, <2 x float> %b) { +; CHECK-LABEL: vfmin_v2f32_vv: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetivli a0, 2, e32,m1,ta,mu +; CHECK-NEXT: vfmin.vv v8, v8, v9 +; CHECK-NEXT: ret + %v = call <2 x float> @llvm.minnum.v2f32(<2 x float> %a, <2 x float> %b) + ret <2 x float> %v +} + +define <2 x float> @vfmin_v2f32_vf(<2 x float> %a, float %b) { +; CHECK-LABEL: vfmin_v2f32_vf: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetivli a0, 2, e32,m1,ta,mu +; CHECK-NEXT: vfmin.vf v8, v8, fa0 +; CHECK-NEXT: ret + %head = insertelement <2 x float> undef, float %b, i32 0 + %splat = shufflevector <2 x float> %head, <2 x float> undef, <2 x i32> zeroinitializer + %v = call <2 x float> @llvm.minnum.v2f32(<2 x float> %a, <2 x float> %splat) + ret <2 x float> %v +} + +declare <4 x float> @llvm.minnum.v4f32(<4 x float>, <4 x float>) + +define <4 x float> @vfmin_v4f32_vv(<4 x float> %a, <4 x float> %b) { +; CHECK-LABEL: vfmin_v4f32_vv: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetivli a0, 4, e32,m1,ta,mu +; CHECK-NEXT: vfmin.vv v8, v8, v9 +; CHECK-NEXT: ret + %v = call <4 x float> @llvm.minnum.v4f32(<4 x float> %a, <4 x float> %b) + ret <4 x float> %v +} + +define <4 x float> @vfmin_v4f32_vf(<4 x float> %a, float %b) { +; CHECK-LABEL: vfmin_v4f32_vf: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetivli a0, 4, e32,m1,ta,mu +; CHECK-NEXT: vfmin.vf v8, v8, fa0 +; CHECK-NEXT: ret + %head = insertelement <4 x float> undef, float %b, i32 0 + %splat = shufflevector <4 x float> %head, <4 x float> undef, <4 x i32> zeroinitializer + %v = call <4 x float> @llvm.minnum.v4f32(<4 x float> %a, <4 x float> %splat) + ret <4 x float> %v +} + +declare <8 x float> @llvm.minnum.v8f32(<8 x float>, <8 x float>) + +define <8 x float> @vfmin_v8f32_vv(<8 x float> %a, <8 x float> %b) { +; CHECK-LABEL: vfmin_v8f32_vv: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetivli a0, 8, e32,m2,ta,mu +; CHECK-NEXT: vfmin.vv v8, v8, v10 +; CHECK-NEXT: ret + %v = call <8 x float> @llvm.minnum.v8f32(<8 x float> %a, <8 x float> %b) + ret <8 x float> %v +} + +define <8 x float> @vfmin_v8f32_vf(<8 x float> %a, float %b) { +; CHECK-LABEL: vfmin_v8f32_vf: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetivli a0, 8, e32,m2,ta,mu +; CHECK-NEXT: vfmin.vf v8, v8, fa0 +; CHECK-NEXT: ret + %head = insertelement <8 x float> undef, float %b, i32 0 + %splat = shufflevector <8 x float> %head, <8 x float> undef, <8 x i32> zeroinitializer + %v = call <8 x float> @llvm.minnum.v8f32(<8 x float> %a, <8 x float> %splat) + ret <8 x float> %v +} + +declare <16 x float> @llvm.minnum.v16f32(<16 x float>, <16 x float>) + +define <16 x float> @vfmin_v16f32_vv(<16 x float> %a, <16 x float> %b) { +; CHECK-LABEL: vfmin_v16f32_vv: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetivli a0, 16, e32,m4,ta,mu +; CHECK-NEXT: vfmin.vv v8, v8, v12 +; CHECK-NEXT: ret + %v = call <16 x float> @llvm.minnum.v16f32(<16 x float> %a, <16 x float> %b) + ret <16 x float> %v +} + +define <16 x float> @vfmin_v16f32_vf(<16 x float> %a, float %b) { +; CHECK-LABEL: vfmin_v16f32_vf: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetivli a0, 16, e32,m4,ta,mu +; CHECK-NEXT: vfmin.vf v8, v8, fa0 +; CHECK-NEXT: ret + %head = insertelement <16 x float> undef, float %b, i32 0 + %splat = shufflevector <16 x float> %head, <16 x float> undef, <16 x i32> zeroinitializer + %v = call <16 x float> @llvm.minnum.v16f32(<16 x float> %a, <16 x float> %splat) + ret <16 x float> %v +} + +declare <2 x double> @llvm.minnum.v2f64(<2 x double>, <2 x double>) + +define <2 x double> @vfmin_v2f64_vv(<2 x double> %a, <2 x double> %b) { +; CHECK-LABEL: vfmin_v2f64_vv: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetivli a0, 2, e64,m1,ta,mu +; CHECK-NEXT: vfmin.vv v8, v8, v9 +; CHECK-NEXT: ret + %v = call <2 x double> @llvm.minnum.v2f64(<2 x double> %a, <2 x double> %b) + ret <2 x double> %v +} + +define <2 x double> @vfmin_v2f64_vf(<2 x double> %a, double %b) { +; CHECK-LABEL: vfmin_v2f64_vf: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetivli a0, 2, e64,m1,ta,mu +; CHECK-NEXT: vfmin.vf v8, v8, fa0 +; CHECK-NEXT: ret + %head = insertelement <2 x double> undef, double %b, i32 0 + %splat = shufflevector <2 x double> %head, <2 x double> undef, <2 x i32> zeroinitializer + %v = call <2 x double> @llvm.minnum.v2f64(<2 x double> %a, <2 x double> %splat) + ret <2 x double> %v +} + +declare <4 x double> @llvm.minnum.v4f64(<4 x double>, <4 x double>) + +define <4 x double> @vfmin_v4f64_vv(<4 x double> %a, <4 x double> %b) { +; CHECK-LABEL: vfmin_v4f64_vv: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetivli a0, 4, e64,m2,ta,mu +; CHECK-NEXT: vfmin.vv v8, v8, v10 +; CHECK-NEXT: ret + %v = call <4 x double> @llvm.minnum.v4f64(<4 x double> %a, <4 x double> %b) + ret <4 x double> %v +} + +define <4 x double> @vfmin_v4f64_vf(<4 x double> %a, double %b) { +; CHECK-LABEL: vfmin_v4f64_vf: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetivli a0, 4, e64,m2,ta,mu +; CHECK-NEXT: vfmin.vf v8, v8, fa0 +; CHECK-NEXT: ret + %head = insertelement <4 x double> undef, double %b, i32 0 + %splat = shufflevector <4 x double> %head, <4 x double> undef, <4 x i32> zeroinitializer + %v = call <4 x double> @llvm.minnum.v4f64(<4 x double> %a, <4 x double> %splat) + ret <4 x double> %v +} + +declare <8 x double> @llvm.minnum.v8f64(<8 x double>, <8 x double>) + +define <8 x double> @vfmin_v8f64_vv(<8 x double> %a, <8 x double> %b) { +; CHECK-LABEL: vfmin_v8f64_vv: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetivli a0, 8, e64,m4,ta,mu +; CHECK-NEXT: vfmin.vv v8, v8, v12 +; CHECK-NEXT: ret + %v = call <8 x double> @llvm.minnum.v8f64(<8 x double> %a, <8 x double> %b) + ret <8 x double> %v +} + +define <8 x double> @vfmin_v8f64_vf(<8 x double> %a, double %b) { +; CHECK-LABEL: vfmin_v8f64_vf: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetivli a0, 8, e64,m4,ta,mu +; CHECK-NEXT: vfmin.vf v8, v8, fa0 +; CHECK-NEXT: ret + %head = insertelement <8 x double> undef, double %b, i32 0 + %splat = shufflevector <8 x double> %head, <8 x double> undef, <8 x i32> zeroinitializer + %v = call <8 x double> @llvm.minnum.v8f64(<8 x double> %a, <8 x double> %splat) + ret <8 x double> %v +} + +declare <16 x double> @llvm.minnum.v16f64(<16 x double>, <16 x double>) + +define <16 x double> @vfmin_v16f64_vv(<16 x double> %a, <16 x double> %b) { +; CHECK-LABEL: vfmin_v16f64_vv: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetivli a0, 16, e64,m8,ta,mu +; CHECK-NEXT: vfmin.vv v8, v8, v16 +; CHECK-NEXT: ret + %v = call <16 x double> @llvm.minnum.v16f64(<16 x double> %a, <16 x double> %b) + ret <16 x double> %v +} + +define <16 x double> @vfmin_v16f64_vf(<16 x double> %a, double %b) { +; CHECK-LABEL: vfmin_v16f64_vf: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetivli a0, 16, e64,m8,ta,mu +; CHECK-NEXT: vfmin.vf v8, v8, fa0 +; CHECK-NEXT: ret + %head = insertelement <16 x double> undef, double %b, i32 0 + %splat = shufflevector <16 x double> %head, <16 x double> undef, <16 x i32> zeroinitializer + %v = call <16 x double> @llvm.minnum.v16f64(<16 x double> %a, <16 x double> %splat) + ret <16 x double> %v +} diff --git a/llvm/test/CodeGen/RISCV/rvv/vfmax-sdnode.ll b/llvm/test/CodeGen/RISCV/rvv/vfmax-sdnode.ll new file mode 100644 --- /dev/null +++ b/llvm/test/CodeGen/RISCV/rvv/vfmax-sdnode.ll @@ -0,0 +1,365 @@ +; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py +; RUN: llc -mtriple=riscv32 -mattr=+d,+experimental-zfh,+experimental-v -target-abi=ilp32d \ +; RUN: -verify-machineinstrs < %s | FileCheck %s +; RUN: llc -mtriple=riscv64 -mattr=+d,+experimental-zfh,+experimental-v -target-abi=lp64d \ +; RUN: -verify-machineinstrs < %s | FileCheck %s + +declare @llvm.maxnum.nxv1f16(, ) + +define @vfmax_nxv1f16_vv( %a, %b) { +; CHECK-LABEL: vfmax_nxv1f16_vv: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetvli a0, zero, e16,mf4,ta,mu +; CHECK-NEXT: vfmax.vv v8, v8, v9 +; CHECK-NEXT: ret + %v = call @llvm.maxnum.nxv1f16( %a, %b) + ret %v +} + +define @vfmax_nxv1f16_vf( %a, half %b) { +; CHECK-LABEL: vfmax_nxv1f16_vf: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetvli a0, zero, e16,mf4,ta,mu +; CHECK-NEXT: vfmax.vf v8, v8, fa0 +; CHECK-NEXT: ret + %head = insertelement undef, half %b, i32 0 + %splat = shufflevector %head, undef, zeroinitializer + %v = call @llvm.maxnum.nxv1f16( %a, %splat) + ret %v +} + +declare @llvm.maxnum.nxv2f16(, ) + +define @vfmax_nxv2f16_vv( %a, %b) { +; CHECK-LABEL: vfmax_nxv2f16_vv: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetvli a0, zero, e16,mf2,ta,mu +; CHECK-NEXT: vfmax.vv v8, v8, v9 +; CHECK-NEXT: ret + %v = call @llvm.maxnum.nxv2f16( %a, %b) + ret %v +} + +define @vfmax_nxv2f16_vf( %a, half %b) { +; CHECK-LABEL: vfmax_nxv2f16_vf: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetvli a0, zero, e16,mf2,ta,mu +; CHECK-NEXT: vfmax.vf v8, v8, fa0 +; CHECK-NEXT: ret + %head = insertelement undef, half %b, i32 0 + %splat = shufflevector %head, undef, zeroinitializer + %v = call @llvm.maxnum.nxv2f16( %a, %splat) + ret %v +} + +declare @llvm.maxnum.nxv4f16(, ) + +define @vfmax_nxv4f16_vv( %a, %b) { +; CHECK-LABEL: vfmax_nxv4f16_vv: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetvli a0, zero, e16,m1,ta,mu +; CHECK-NEXT: vfmax.vv v8, v8, v9 +; CHECK-NEXT: ret + %v = call @llvm.maxnum.nxv4f16( %a, %b) + ret %v +} + +define @vfmax_nxv4f16_vf( %a, half %b) { +; CHECK-LABEL: vfmax_nxv4f16_vf: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetvli a0, zero, e16,m1,ta,mu +; CHECK-NEXT: vfmax.vf v8, v8, fa0 +; CHECK-NEXT: ret + %head = insertelement undef, half %b, i32 0 + %splat = shufflevector %head, undef, zeroinitializer + %v = call @llvm.maxnum.nxv4f16( %a, %splat) + ret %v +} + +declare @llvm.maxnum.nxv8f16(, ) + +define @vfmax_nxv8f16_vv( %a, %b) { +; CHECK-LABEL: vfmax_nxv8f16_vv: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetvli a0, zero, e16,m2,ta,mu +; CHECK-NEXT: vfmax.vv v8, v8, v10 +; CHECK-NEXT: ret + %v = call @llvm.maxnum.nxv8f16( %a, %b) + ret %v +} + +define @vfmax_nxv8f16_vf( %a, half %b) { +; CHECK-LABEL: vfmax_nxv8f16_vf: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetvli a0, zero, e16,m2,ta,mu +; CHECK-NEXT: vfmax.vf v8, v8, fa0 +; CHECK-NEXT: ret + %head = insertelement undef, half %b, i32 0 + %splat = shufflevector %head, undef, zeroinitializer + %v = call @llvm.maxnum.nxv8f16( %a, %splat) + ret %v +} + +declare @llvm.maxnum.nxv16f16(, ) + +define @vfmax_nxv16f16_vv( %a, %b) { +; CHECK-LABEL: vfmax_nxv16f16_vv: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetvli a0, zero, e16,m4,ta,mu +; CHECK-NEXT: vfmax.vv v8, v8, v12 +; CHECK-NEXT: ret + %v = call @llvm.maxnum.nxv16f16( %a, %b) + ret %v +} + +define @vfmax_nxv16f16_vf( %a, half %b) { +; CHECK-LABEL: vfmax_nxv16f16_vf: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetvli a0, zero, e16,m4,ta,mu +; CHECK-NEXT: vfmax.vf v8, v8, fa0 +; CHECK-NEXT: ret + %head = insertelement undef, half %b, i32 0 + %splat = shufflevector %head, undef, zeroinitializer + %v = call @llvm.maxnum.nxv16f16( %a, %splat) + ret %v +} + +declare @llvm.maxnum.nxv32f16(, ) + +define @vfmax_nxv32f16_vv( %a, %b) { +; CHECK-LABEL: vfmax_nxv32f16_vv: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetvli a0, zero, e16,m8,ta,mu +; CHECK-NEXT: vfmax.vv v8, v8, v16 +; CHECK-NEXT: ret + %v = call @llvm.maxnum.nxv32f16( %a, %b) + ret %v +} + +define @vfmax_nxv32f16_vf( %a, half %b) { +; CHECK-LABEL: vfmax_nxv32f16_vf: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetvli a0, zero, e16,m8,ta,mu +; CHECK-NEXT: vfmax.vf v8, v8, fa0 +; CHECK-NEXT: ret + %head = insertelement undef, half %b, i32 0 + %splat = shufflevector %head, undef, zeroinitializer + %v = call @llvm.maxnum.nxv32f16( %a, %splat) + ret %v +} + +declare @llvm.maxnum.nxv1f32(, ) + +define @vfmax_nxv1f32_vv( %a, %b) { +; CHECK-LABEL: vfmax_nxv1f32_vv: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetvli a0, zero, e32,mf2,ta,mu +; CHECK-NEXT: vfmax.vv v8, v8, v9 +; CHECK-NEXT: ret + %v = call @llvm.maxnum.nxv1f32( %a, %b) + ret %v +} + +define @vfmax_nxv1f32_vf( %a, float %b) { +; CHECK-LABEL: vfmax_nxv1f32_vf: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetvli a0, zero, e32,mf2,ta,mu +; CHECK-NEXT: vfmax.vf v8, v8, fa0 +; CHECK-NEXT: ret + %head = insertelement undef, float %b, i32 0 + %splat = shufflevector %head, undef, zeroinitializer + %v = call @llvm.maxnum.nxv1f32( %a, %splat) + ret %v +} + +declare @llvm.maxnum.nxv2f32(, ) + +define @vfmax_nxv2f32_vv( %a, %b) { +; CHECK-LABEL: vfmax_nxv2f32_vv: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetvli a0, zero, e32,m1,ta,mu +; CHECK-NEXT: vfmax.vv v8, v8, v9 +; CHECK-NEXT: ret + %v = call @llvm.maxnum.nxv2f32( %a, %b) + ret %v +} + +define @vfmax_nxv2f32_vf( %a, float %b) { +; CHECK-LABEL: vfmax_nxv2f32_vf: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetvli a0, zero, e32,m1,ta,mu +; CHECK-NEXT: vfmax.vf v8, v8, fa0 +; CHECK-NEXT: ret + %head = insertelement undef, float %b, i32 0 + %splat = shufflevector %head, undef, zeroinitializer + %v = call @llvm.maxnum.nxv2f32( %a, %splat) + ret %v +} + +declare @llvm.maxnum.nxv4f32(, ) + +define @vfmax_nxv4f32_vv( %a, %b) { +; CHECK-LABEL: vfmax_nxv4f32_vv: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetvli a0, zero, e32,m2,ta,mu +; CHECK-NEXT: vfmax.vv v8, v8, v10 +; CHECK-NEXT: ret + %v = call @llvm.maxnum.nxv4f32( %a, %b) + ret %v +} + +define @vfmax_nxv4f32_vf( %a, float %b) { +; CHECK-LABEL: vfmax_nxv4f32_vf: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetvli a0, zero, e32,m2,ta,mu +; CHECK-NEXT: vfmax.vf v8, v8, fa0 +; CHECK-NEXT: ret + %head = insertelement undef, float %b, i32 0 + %splat = shufflevector %head, undef, zeroinitializer + %v = call @llvm.maxnum.nxv4f32( %a, %splat) + ret %v +} + +declare @llvm.maxnum.nxv8f32(, ) + +define @vfmax_nxv8f32_vv( %a, %b) { +; CHECK-LABEL: vfmax_nxv8f32_vv: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetvli a0, zero, e32,m4,ta,mu +; CHECK-NEXT: vfmax.vv v8, v8, v12 +; CHECK-NEXT: ret + %v = call @llvm.maxnum.nxv8f32( %a, %b) + ret %v +} + +define @vfmax_nxv8f32_vf( %a, float %b) { +; CHECK-LABEL: vfmax_nxv8f32_vf: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetvli a0, zero, e32,m4,ta,mu +; CHECK-NEXT: vfmax.vf v8, v8, fa0 +; CHECK-NEXT: ret + %head = insertelement undef, float %b, i32 0 + %splat = shufflevector %head, undef, zeroinitializer + %v = call @llvm.maxnum.nxv8f32( %a, %splat) + ret %v +} + +declare @llvm.maxnum.nxv16f32(, ) + +define @vfmax_nxv16f32_vv( %a, %b) { +; CHECK-LABEL: vfmax_nxv16f32_vv: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetvli a0, zero, e32,m8,ta,mu +; CHECK-NEXT: vfmax.vv v8, v8, v16 +; CHECK-NEXT: ret + %v = call @llvm.maxnum.nxv16f32( %a, %b) + ret %v +} + +define @vfmax_nxv16f32_vf( %a, float %b) { +; CHECK-LABEL: vfmax_nxv16f32_vf: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetvli a0, zero, e32,m8,ta,mu +; CHECK-NEXT: vfmax.vf v8, v8, fa0 +; CHECK-NEXT: ret + %head = insertelement undef, float %b, i32 0 + %splat = shufflevector %head, undef, zeroinitializer + %v = call @llvm.maxnum.nxv16f32( %a, %splat) + ret %v +} + +declare @llvm.maxnum.nxv1f64(, ) + +define @vfmax_nxv1f64_vv( %a, %b) { +; CHECK-LABEL: vfmax_nxv1f64_vv: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetvli a0, zero, e64,m1,ta,mu +; CHECK-NEXT: vfmax.vv v8, v8, v9 +; CHECK-NEXT: ret + %v = call @llvm.maxnum.nxv1f64( %a, %b) + ret %v +} + +define @vfmax_nxv1f64_vf( %a, double %b) { +; CHECK-LABEL: vfmax_nxv1f64_vf: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetvli a0, zero, e64,m1,ta,mu +; CHECK-NEXT: vfmax.vf v8, v8, fa0 +; CHECK-NEXT: ret + %head = insertelement undef, double %b, i32 0 + %splat = shufflevector %head, undef, zeroinitializer + %v = call @llvm.maxnum.nxv1f64( %a, %splat) + ret %v +} + +declare @llvm.maxnum.nxv2f64(, ) + +define @vfmax_nxv2f64_vv( %a, %b) { +; CHECK-LABEL: vfmax_nxv2f64_vv: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetvli a0, zero, e64,m2,ta,mu +; CHECK-NEXT: vfmax.vv v8, v8, v10 +; CHECK-NEXT: ret + %v = call @llvm.maxnum.nxv2f64( %a, %b) + ret %v +} + +define @vfmax_nxv2f64_vf( %a, double %b) { +; CHECK-LABEL: vfmax_nxv2f64_vf: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetvli a0, zero, e64,m2,ta,mu +; CHECK-NEXT: vfmax.vf v8, v8, fa0 +; CHECK-NEXT: ret + %head = insertelement undef, double %b, i32 0 + %splat = shufflevector %head, undef, zeroinitializer + %v = call @llvm.maxnum.nxv2f64( %a, %splat) + ret %v +} + +declare @llvm.maxnum.nxv4f64(, ) + +define @vfmax_nxv4f64_vv( %a, %b) { +; CHECK-LABEL: vfmax_nxv4f64_vv: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetvli a0, zero, e64,m4,ta,mu +; CHECK-NEXT: vfmax.vv v8, v8, v12 +; CHECK-NEXT: ret + %v = call @llvm.maxnum.nxv4f64( %a, %b) + ret %v +} + +define @vfmax_nxv4f64_vf( %a, double %b) { +; CHECK-LABEL: vfmax_nxv4f64_vf: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetvli a0, zero, e64,m4,ta,mu +; CHECK-NEXT: vfmax.vf v8, v8, fa0 +; CHECK-NEXT: ret + %head = insertelement undef, double %b, i32 0 + %splat = shufflevector %head, undef, zeroinitializer + %v = call @llvm.maxnum.nxv4f64( %a, %splat) + ret %v +} + +declare @llvm.maxnum.nxv8f64(, ) + +define @vfmax_nxv8f64_vv( %a, %b) { +; CHECK-LABEL: vfmax_nxv8f64_vv: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetvli a0, zero, e64,m8,ta,mu +; CHECK-NEXT: vfmax.vv v8, v8, v16 +; CHECK-NEXT: ret + %v = call @llvm.maxnum.nxv8f64( %a, %b) + ret %v +} + +define @vfmax_nxv8f64_vf( %a, double %b) { +; CHECK-LABEL: vfmax_nxv8f64_vf: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetvli a0, zero, e64,m8,ta,mu +; CHECK-NEXT: vfmax.vf v8, v8, fa0 +; CHECK-NEXT: ret + %head = insertelement undef, double %b, i32 0 + %splat = shufflevector %head, undef, zeroinitializer + %v = call @llvm.maxnum.nxv8f64( %a, %splat) + ret %v +} diff --git a/llvm/test/CodeGen/RISCV/rvv/vfmin-sdnode.ll b/llvm/test/CodeGen/RISCV/rvv/vfmin-sdnode.ll new file mode 100644 --- /dev/null +++ b/llvm/test/CodeGen/RISCV/rvv/vfmin-sdnode.ll @@ -0,0 +1,365 @@ +; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py +; RUN: llc -mtriple=riscv32 -mattr=+d,+experimental-zfh,+experimental-v -target-abi=ilp32d \ +; RUN: -verify-machineinstrs < %s | FileCheck %s +; RUN: llc -mtriple=riscv64 -mattr=+d,+experimental-zfh,+experimental-v -target-abi=lp64d \ +; RUN: -verify-machineinstrs < %s | FileCheck %s + +declare @llvm.minnum.nxv1f16(, ) + +define @vfmin_nxv1f16_vv( %a, %b) { +; CHECK-LABEL: vfmin_nxv1f16_vv: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetvli a0, zero, e16,mf4,ta,mu +; CHECK-NEXT: vfmin.vv v8, v8, v9 +; CHECK-NEXT: ret + %v = call @llvm.minnum.nxv1f16( %a, %b) + ret %v +} + +define @vfmin_nxv1f16_vf( %a, half %b) { +; CHECK-LABEL: vfmin_nxv1f16_vf: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetvli a0, zero, e16,mf4,ta,mu +; CHECK-NEXT: vfmin.vf v8, v8, fa0 +; CHECK-NEXT: ret + %head = insertelement undef, half %b, i32 0 + %splat = shufflevector %head, undef, zeroinitializer + %v = call @llvm.minnum.nxv1f16( %a, %splat) + ret %v +} + +declare @llvm.minnum.nxv2f16(, ) + +define @vfmin_nxv2f16_vv( %a, %b) { +; CHECK-LABEL: vfmin_nxv2f16_vv: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetvli a0, zero, e16,mf2,ta,mu +; CHECK-NEXT: vfmin.vv v8, v8, v9 +; CHECK-NEXT: ret + %v = call @llvm.minnum.nxv2f16( %a, %b) + ret %v +} + +define @vfmin_nxv2f16_vf( %a, half %b) { +; CHECK-LABEL: vfmin_nxv2f16_vf: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetvli a0, zero, e16,mf2,ta,mu +; CHECK-NEXT: vfmin.vf v8, v8, fa0 +; CHECK-NEXT: ret + %head = insertelement undef, half %b, i32 0 + %splat = shufflevector %head, undef, zeroinitializer + %v = call @llvm.minnum.nxv2f16( %a, %splat) + ret %v +} + +declare @llvm.minnum.nxv4f16(, ) + +define @vfmin_nxv4f16_vv( %a, %b) { +; CHECK-LABEL: vfmin_nxv4f16_vv: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetvli a0, zero, e16,m1,ta,mu +; CHECK-NEXT: vfmin.vv v8, v8, v9 +; CHECK-NEXT: ret + %v = call @llvm.minnum.nxv4f16( %a, %b) + ret %v +} + +define @vfmin_nxv4f16_vf( %a, half %b) { +; CHECK-LABEL: vfmin_nxv4f16_vf: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetvli a0, zero, e16,m1,ta,mu +; CHECK-NEXT: vfmin.vf v8, v8, fa0 +; CHECK-NEXT: ret + %head = insertelement undef, half %b, i32 0 + %splat = shufflevector %head, undef, zeroinitializer + %v = call @llvm.minnum.nxv4f16( %a, %splat) + ret %v +} + +declare @llvm.minnum.nxv8f16(, ) + +define @vfmin_nxv8f16_vv( %a, %b) { +; CHECK-LABEL: vfmin_nxv8f16_vv: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetvli a0, zero, e16,m2,ta,mu +; CHECK-NEXT: vfmin.vv v8, v8, v10 +; CHECK-NEXT: ret + %v = call @llvm.minnum.nxv8f16( %a, %b) + ret %v +} + +define @vfmin_nxv8f16_vf( %a, half %b) { +; CHECK-LABEL: vfmin_nxv8f16_vf: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetvli a0, zero, e16,m2,ta,mu +; CHECK-NEXT: vfmin.vf v8, v8, fa0 +; CHECK-NEXT: ret + %head = insertelement undef, half %b, i32 0 + %splat = shufflevector %head, undef, zeroinitializer + %v = call @llvm.minnum.nxv8f16( %a, %splat) + ret %v +} + +declare @llvm.minnum.nxv16f16(, ) + +define @vfmin_nxv16f16_vv( %a, %b) { +; CHECK-LABEL: vfmin_nxv16f16_vv: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetvli a0, zero, e16,m4,ta,mu +; CHECK-NEXT: vfmin.vv v8, v8, v12 +; CHECK-NEXT: ret + %v = call @llvm.minnum.nxv16f16( %a, %b) + ret %v +} + +define @vfmin_nxv16f16_vf( %a, half %b) { +; CHECK-LABEL: vfmin_nxv16f16_vf: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetvli a0, zero, e16,m4,ta,mu +; CHECK-NEXT: vfmin.vf v8, v8, fa0 +; CHECK-NEXT: ret + %head = insertelement undef, half %b, i32 0 + %splat = shufflevector %head, undef, zeroinitializer + %v = call @llvm.minnum.nxv16f16( %a, %splat) + ret %v +} + +declare @llvm.minnum.nxv32f16(, ) + +define @vfmin_nxv32f16_vv( %a, %b) { +; CHECK-LABEL: vfmin_nxv32f16_vv: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetvli a0, zero, e16,m8,ta,mu +; CHECK-NEXT: vfmin.vv v8, v8, v16 +; CHECK-NEXT: ret + %v = call @llvm.minnum.nxv32f16( %a, %b) + ret %v +} + +define @vfmin_nxv32f16_vf( %a, half %b) { +; CHECK-LABEL: vfmin_nxv32f16_vf: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetvli a0, zero, e16,m8,ta,mu +; CHECK-NEXT: vfmin.vf v8, v8, fa0 +; CHECK-NEXT: ret + %head = insertelement undef, half %b, i32 0 + %splat = shufflevector %head, undef, zeroinitializer + %v = call @llvm.minnum.nxv32f16( %a, %splat) + ret %v +} + +declare @llvm.minnum.nxv1f32(, ) + +define @vfmin_nxv1f32_vv( %a, %b) { +; CHECK-LABEL: vfmin_nxv1f32_vv: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetvli a0, zero, e32,mf2,ta,mu +; CHECK-NEXT: vfmin.vv v8, v8, v9 +; CHECK-NEXT: ret + %v = call @llvm.minnum.nxv1f32( %a, %b) + ret %v +} + +define @vfmin_nxv1f32_vf( %a, float %b) { +; CHECK-LABEL: vfmin_nxv1f32_vf: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetvli a0, zero, e32,mf2,ta,mu +; CHECK-NEXT: vfmin.vf v8, v8, fa0 +; CHECK-NEXT: ret + %head = insertelement undef, float %b, i32 0 + %splat = shufflevector %head, undef, zeroinitializer + %v = call @llvm.minnum.nxv1f32( %a, %splat) + ret %v +} + +declare @llvm.minnum.nxv2f32(, ) + +define @vfmin_nxv2f32_vv( %a, %b) { +; CHECK-LABEL: vfmin_nxv2f32_vv: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetvli a0, zero, e32,m1,ta,mu +; CHECK-NEXT: vfmin.vv v8, v8, v9 +; CHECK-NEXT: ret + %v = call @llvm.minnum.nxv2f32( %a, %b) + ret %v +} + +define @vfmin_nxv2f32_vf( %a, float %b) { +; CHECK-LABEL: vfmin_nxv2f32_vf: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetvli a0, zero, e32,m1,ta,mu +; CHECK-NEXT: vfmin.vf v8, v8, fa0 +; CHECK-NEXT: ret + %head = insertelement undef, float %b, i32 0 + %splat = shufflevector %head, undef, zeroinitializer + %v = call @llvm.minnum.nxv2f32( %a, %splat) + ret %v +} + +declare @llvm.minnum.nxv4f32(, ) + +define @vfmin_nxv4f32_vv( %a, %b) { +; CHECK-LABEL: vfmin_nxv4f32_vv: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetvli a0, zero, e32,m2,ta,mu +; CHECK-NEXT: vfmin.vv v8, v8, v10 +; CHECK-NEXT: ret + %v = call @llvm.minnum.nxv4f32( %a, %b) + ret %v +} + +define @vfmin_nxv4f32_vf( %a, float %b) { +; CHECK-LABEL: vfmin_nxv4f32_vf: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetvli a0, zero, e32,m2,ta,mu +; CHECK-NEXT: vfmin.vf v8, v8, fa0 +; CHECK-NEXT: ret + %head = insertelement undef, float %b, i32 0 + %splat = shufflevector %head, undef, zeroinitializer + %v = call @llvm.minnum.nxv4f32( %a, %splat) + ret %v +} + +declare @llvm.minnum.nxv8f32(, ) + +define @vfmin_nxv8f32_vv( %a, %b) { +; CHECK-LABEL: vfmin_nxv8f32_vv: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetvli a0, zero, e32,m4,ta,mu +; CHECK-NEXT: vfmin.vv v8, v8, v12 +; CHECK-NEXT: ret + %v = call @llvm.minnum.nxv8f32( %a, %b) + ret %v +} + +define @vfmin_nxv8f32_vf( %a, float %b) { +; CHECK-LABEL: vfmin_nxv8f32_vf: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetvli a0, zero, e32,m4,ta,mu +; CHECK-NEXT: vfmin.vf v8, v8, fa0 +; CHECK-NEXT: ret + %head = insertelement undef, float %b, i32 0 + %splat = shufflevector %head, undef, zeroinitializer + %v = call @llvm.minnum.nxv8f32( %a, %splat) + ret %v +} + +declare @llvm.minnum.nxv16f32(, ) + +define @vfmin_nxv16f32_vv( %a, %b) { +; CHECK-LABEL: vfmin_nxv16f32_vv: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetvli a0, zero, e32,m8,ta,mu +; CHECK-NEXT: vfmin.vv v8, v8, v16 +; CHECK-NEXT: ret + %v = call @llvm.minnum.nxv16f32( %a, %b) + ret %v +} + +define @vfmin_nxv16f32_vf( %a, float %b) { +; CHECK-LABEL: vfmin_nxv16f32_vf: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetvli a0, zero, e32,m8,ta,mu +; CHECK-NEXT: vfmin.vf v8, v8, fa0 +; CHECK-NEXT: ret + %head = insertelement undef, float %b, i32 0 + %splat = shufflevector %head, undef, zeroinitializer + %v = call @llvm.minnum.nxv16f32( %a, %splat) + ret %v +} + +declare @llvm.minnum.nxv1f64(, ) + +define @vfmin_nxv1f64_vv( %a, %b) { +; CHECK-LABEL: vfmin_nxv1f64_vv: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetvli a0, zero, e64,m1,ta,mu +; CHECK-NEXT: vfmin.vv v8, v8, v9 +; CHECK-NEXT: ret + %v = call @llvm.minnum.nxv1f64( %a, %b) + ret %v +} + +define @vfmin_nxv1f64_vf( %a, double %b) { +; CHECK-LABEL: vfmin_nxv1f64_vf: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetvli a0, zero, e64,m1,ta,mu +; CHECK-NEXT: vfmin.vf v8, v8, fa0 +; CHECK-NEXT: ret + %head = insertelement undef, double %b, i32 0 + %splat = shufflevector %head, undef, zeroinitializer + %v = call @llvm.minnum.nxv1f64( %a, %splat) + ret %v +} + +declare @llvm.minnum.nxv2f64(, ) + +define @vfmin_nxv2f64_vv( %a, %b) { +; CHECK-LABEL: vfmin_nxv2f64_vv: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetvli a0, zero, e64,m2,ta,mu +; CHECK-NEXT: vfmin.vv v8, v8, v10 +; CHECK-NEXT: ret + %v = call @llvm.minnum.nxv2f64( %a, %b) + ret %v +} + +define @vfmin_nxv2f64_vf( %a, double %b) { +; CHECK-LABEL: vfmin_nxv2f64_vf: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetvli a0, zero, e64,m2,ta,mu +; CHECK-NEXT: vfmin.vf v8, v8, fa0 +; CHECK-NEXT: ret + %head = insertelement undef, double %b, i32 0 + %splat = shufflevector %head, undef, zeroinitializer + %v = call @llvm.minnum.nxv2f64( %a, %splat) + ret %v +} + +declare @llvm.minnum.nxv4f64(, ) + +define @vfmin_nxv4f64_vv( %a, %b) { +; CHECK-LABEL: vfmin_nxv4f64_vv: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetvli a0, zero, e64,m4,ta,mu +; CHECK-NEXT: vfmin.vv v8, v8, v12 +; CHECK-NEXT: ret + %v = call @llvm.minnum.nxv4f64( %a, %b) + ret %v +} + +define @vfmin_nxv4f64_vf( %a, double %b) { +; CHECK-LABEL: vfmin_nxv4f64_vf: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetvli a0, zero, e64,m4,ta,mu +; CHECK-NEXT: vfmin.vf v8, v8, fa0 +; CHECK-NEXT: ret + %head = insertelement undef, double %b, i32 0 + %splat = shufflevector %head, undef, zeroinitializer + %v = call @llvm.minnum.nxv4f64( %a, %splat) + ret %v +} + +declare @llvm.minnum.nxv8f64(, ) + +define @vfmin_nxv8f64_vv( %a, %b) { +; CHECK-LABEL: vfmin_nxv8f64_vv: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetvli a0, zero, e64,m8,ta,mu +; CHECK-NEXT: vfmin.vv v8, v8, v16 +; CHECK-NEXT: ret + %v = call @llvm.minnum.nxv8f64( %a, %b) + ret %v +} + +define @vfmin_nxv8f64_vf( %a, double %b) { +; CHECK-LABEL: vfmin_nxv8f64_vf: +; CHECK: # %bb.0: +; CHECK-NEXT: vsetvli a0, zero, e64,m8,ta,mu +; CHECK-NEXT: vfmin.vf v8, v8, fa0 +; CHECK-NEXT: ret + %head = insertelement undef, double %b, i32 0 + %splat = shufflevector %head, undef, zeroinitializer + %v = call @llvm.minnum.nxv8f64( %a, %splat) + ret %v +}