AArch64: add missing ACLE intrinsics mapping to general arithmetic operation from VFP instructions.

E.g. float64x1_t vadd_f64(float64x1_t a, float64x1_t b) -> FADD Dd, Dn, Dm.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@196208 91177308-0d34-0410-b5e6-96231b3b80d8
This commit is contained in:
Hao Liu 2013-12-03 05:58:30 +00:00
parent 6379a82d7c
commit 5025a48f68
2 changed files with 148 additions and 0 deletions

View File

@ -5651,6 +5651,37 @@ defm : Neon_ScalarXIndexedElem_MLAL_Patterns<int_arm_neon_vqsubs,
int_arm_neon_vqdmull, SQDMLSLdsv_4S, v1i64, FPR64, FPR32, v1i32, v4i32,
i32, VPR128Lo, neon_uimm2_bare>;
// Scalar general arithmetic operation
class Neon_Scalar_GeneralMath2D_pattern<SDPatternOperator opnode,
Instruction INST>
: Pat<(v1f64 (opnode (v1f64 FPR64:$Rn))), (INST FPR64:$Rn)>;
class Neon_Scalar_GeneralMath3D_pattern<SDPatternOperator opnode,
Instruction INST>
: Pat<(v1f64 (opnode (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
(INST FPR64:$Rn, FPR64:$Rm)>;
class Neon_Scalar_GeneralMath4D_pattern<SDPatternOperator opnode,
Instruction INST>
: Pat<(v1f64 (opnode (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm),
(v1f64 FPR64:$Ra))),
(INST FPR64:$Rn, FPR64:$Rm, FPR64:$Ra)>;
def : Neon_Scalar_GeneralMath3D_pattern<fadd, FADDddd>;
def : Neon_Scalar_GeneralMath3D_pattern<fmul, FMULddd>;
def : Neon_Scalar_GeneralMath3D_pattern<fsub, FSUBddd>;
def : Neon_Scalar_GeneralMath3D_pattern<fdiv, FDIVddd>;
def : Neon_Scalar_GeneralMath3D_pattern<int_arm_neon_vabds, FABDddd>;
def : Neon_Scalar_GeneralMath3D_pattern<int_arm_neon_vmaxs, FMAXddd>;
def : Neon_Scalar_GeneralMath3D_pattern<int_arm_neon_vmins, FMINddd>;
def : Neon_Scalar_GeneralMath3D_pattern<int_aarch64_neon_vmaxnm, FMAXNMddd>;
def : Neon_Scalar_GeneralMath3D_pattern<int_aarch64_neon_vminnm, FMINNMddd>;
def : Neon_Scalar_GeneralMath2D_pattern<fabs, FABSdd>;
def : Neon_Scalar_GeneralMath2D_pattern<fneg, FNEGdd>;
def : Neon_Scalar_GeneralMath4D_pattern<fma, FMADDdddd>;
def : Neon_Scalar_GeneralMath4D_pattern<fmsub, FMSUBdddd>;
// Scalar Signed saturating doubling multiply returning
// high half (scalar, by element)

View File

@ -118,3 +118,120 @@ define <2 x double> @sub2xdouble(<2 x double> %A, <2 x double> %B) {
ret <2 x double> %tmp3
}
define <1 x double> @test_vadd_f64(<1 x double> %a, <1 x double> %b) {
; CHECK-LABEL: test_vadd_f64
; CHECK: fadd d{{[0-9]+}}, d{{[0-9]+}}, d{{[0-9]+}}
%1 = fadd <1 x double> %a, %b
ret <1 x double> %1
}
define <1 x double> @test_vmul_f64(<1 x double> %a, <1 x double> %b) {
; CHECK-LABEL: test_vmul_f64
; CHECK: fmul d{{[0-9]+}}, d{{[0-9]+}}, d{{[0-9]+}}
%1 = fmul <1 x double> %a, %b
ret <1 x double> %1
}
define <1 x double> @test_vdiv_f64(<1 x double> %a, <1 x double> %b) {
; CHECK-LABEL: test_vdiv_f64
; CHECK: fdiv d{{[0-9]+}}, d{{[0-9]+}}, d{{[0-9]+}}
%1 = fdiv <1 x double> %a, %b
ret <1 x double> %1
}
define <1 x double> @test_vmla_f64(<1 x double> %a, <1 x double> %b, <1 x double> %c) {
; CHECK-LABEL: test_vmla_f64
; CHECK: fmul d{{[0-9]+}}, d{{[0-9]+}}, d{{[0-9]+}}
; CHECK: fadd d{{[0-9]+}}, d{{[0-9]+}}, d{{[0-9]+}}
%1 = fmul <1 x double> %b, %c
%2 = fadd <1 x double> %1, %a
ret <1 x double> %2
}
define <1 x double> @test_vmls_f64(<1 x double> %a, <1 x double> %b, <1 x double> %c) {
; CHECK-LABEL: test_vmls_f64
; CHECK: fmul d{{[0-9]+}}, d{{[0-9]+}}, d{{[0-9]+}}
; CHECK: fsub d{{[0-9]+}}, d{{[0-9]+}}, d{{[0-9]+}}
%1 = fmul <1 x double> %b, %c
%2 = fsub <1 x double> %a, %1
ret <1 x double> %2
}
define <1 x double> @test_vfms_f64(<1 x double> %a, <1 x double> %b, <1 x double> %c) {
; CHECK-LABEL: test_vfms_f64
; CHECK: fmsub d{{[0-9]+}}, d{{[0-9]+}}, d{{[0-9]+}}, d{{[0-9]+}}
%1 = fsub <1 x double> <double -0.000000e+00>, %b
%2 = tail call <1 x double> @llvm.fma.v1f64(<1 x double> %1, <1 x double> %c, <1 x double> %a)
ret <1 x double> %2
}
define <1 x double> @test_vfma_f64(<1 x double> %a, <1 x double> %b, <1 x double> %c) {
; CHECK-LABEL: test_vfma_f64
; CHECK: fmadd d{{[0-9]+}}, d{{[0-9]+}}, d{{[0-9]+}}, d{{[0-9]+}}
%1 = tail call <1 x double> @llvm.fma.v1f64(<1 x double> %b, <1 x double> %c, <1 x double> %a)
ret <1 x double> %1
}
define <1 x double> @test_vsub_f64(<1 x double> %a, <1 x double> %b) {
; CHECK-LABEL: test_vsub_f64
; CHECK: fsub d{{[0-9]+}}, d{{[0-9]+}}, d{{[0-9]+}}
%1 = fsub <1 x double> %a, %b
ret <1 x double> %1
}
define <1 x double> @test_vabd_f64(<1 x double> %a, <1 x double> %b) {
; CHECK-LABEL: test_vabd_f64
; CHECK: fabd d{{[0-9]+}}, d{{[0-9]+}}, d{{[0-9]+}}
%1 = tail call <1 x double> @llvm.arm.neon.vabds.v1f64(<1 x double> %a, <1 x double> %b)
ret <1 x double> %1
}
define <1 x double> @test_vmax_f64(<1 x double> %a, <1 x double> %b) {
; CHECK-LABEL: test_vmax_f64
; CHECK: fmax d{{[0-9]+}}, d{{[0-9]+}}, d{{[0-9]+}}
%1 = tail call <1 x double> @llvm.arm.neon.vmaxs.v1f64(<1 x double> %a, <1 x double> %b)
ret <1 x double> %1
}
define <1 x double> @test_vmin_f64(<1 x double> %a, <1 x double> %b) {
; CHECK-LABEL: test_vmin_f64
; CHECK: fmin d{{[0-9]+}}, d{{[0-9]+}}, d{{[0-9]+}}
%1 = tail call <1 x double> @llvm.arm.neon.vmins.v1f64(<1 x double> %a, <1 x double> %b)
ret <1 x double> %1
}
define <1 x double> @test_vmaxnm_f64(<1 x double> %a, <1 x double> %b) {
; CHECK-LABEL: test_vmaxnm_f64
; CHECK: fmaxnm d{{[0-9]+}}, d{{[0-9]+}}, d{{[0-9]+}}
%1 = tail call <1 x double> @llvm.aarch64.neon.vmaxnm.v1f64(<1 x double> %a, <1 x double> %b)
ret <1 x double> %1
}
define <1 x double> @test_vminnm_f64(<1 x double> %a, <1 x double> %b) {
; CHECK-LABEL: test_vminnm_f64
; CHECK: fminnm d{{[0-9]+}}, d{{[0-9]+}}, d{{[0-9]+}}
%1 = tail call <1 x double> @llvm.aarch64.neon.vminnm.v1f64(<1 x double> %a, <1 x double> %b)
ret <1 x double> %1
}
define <1 x double> @test_vabs_f64(<1 x double> %a) {
; CHECK-LABEL: test_vabs_f64
; CHECK: fabs d{{[0-9]+}}, d{{[0-9]+}}
%1 = tail call <1 x double> @llvm.fabs.v1f64(<1 x double> %a)
ret <1 x double> %1
}
define <1 x double> @test_vneg_f64(<1 x double> %a) {
; CHECK-LABEL: test_vneg_f64
; CHECK: fneg d{{[0-9]+}}, d{{[0-9]+}}
%1 = fsub <1 x double> <double -0.000000e+00>, %a
ret <1 x double> %1
}
declare <1 x double> @llvm.fabs.v1f64(<1 x double>)
declare <1 x double> @llvm.aarch64.neon.vminnm.v1f64(<1 x double>, <1 x double>)
declare <1 x double> @llvm.aarch64.neon.vmaxnm.v1f64(<1 x double>, <1 x double>)
declare <1 x double> @llvm.arm.neon.vmins.v1f64(<1 x double>, <1 x double>)
declare <1 x double> @llvm.arm.neon.vmaxs.v1f64(<1 x double>, <1 x double>)
declare <1 x double> @llvm.arm.neon.vabds.v1f64(<1 x double>, <1 x double>)
declare <1 x double> @llvm.fma.v1f64(<1 x double>, <1 x double>, <1 x double>)