RPCS3/llvm
1
0
Fork 0
mirror of https://github.com/RPCS3/llvm.git synced 2024-12-20 11:08:27 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

ARM64: shuffle patterns around for fmin/fmax & add tests.

Browse Source 
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@205205 91177308-0d34-0410-b5e6-96231b3b80d8
This commit is contained in:
Tim Northover 2014-03-31 15:46:30 +00:00
parent 576c3f709f
commit 4417e07e39
2 changed files with 109 additions and 8 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

16
lib/Target/ARM64/ARM64InstrInfo.td
View File

@ -2857,12 +2857,20 @@ def : Pat<(f32 (int_arm64_neon_faddv (v4f32 V128:$Rn))),
(FADDPv2i32p (EXTRACT_SUBREG (FADDPv4f32 V128:$Rn, V128:$Rn), dsub))>;
def : Pat<(f64 (int_arm64_neon_faddv (v2f64 V128:$Rn))),
(FADDPv2i64p V128:$Rn)>;
def : Pat<(f32 (int_arm64_neon_fmaxnmv (v2f32 V64:$Rn))),
(FMAXNMPv2i32p V64:$Rn)>;
def : Pat<(f64 (int_arm64_neon_fmaxnmv (v2f64 V128:$Rn))),
(FMAXNMPv2i64p V128:$Rn)>;
def : Pat<(f32 (int_arm64_neon_fmaxv (v2f32 V64:$Rn))),
(FMAXPv2i32p V64:$Rn)>;
def : Pat<(f64 (int_arm64_neon_fmaxv (v2f64 V128:$Rn))),
(FMAXPv2i64p V128:$Rn)>;
def : Pat<(f32 (int_arm64_neon_fminnmv (v2f32 V64:$Rn))),
(FMINNMPv2i32p V64:$Rn)>;
def : Pat<(f64 (int_arm64_neon_fminnmv (v2f64 V128:$Rn))),
(FMINNMPv2i64p V128:$Rn)>;
def : Pat<(f32 (int_arm64_neon_fminv (v2f32 V64:$Rn))),
(FMINPv2i32p V64:$Rn)>;
def : Pat<(f64 (int_arm64_neon_fminv (v2f64 V128:$Rn))),
(FMINPv2i64p V128:$Rn)>;
@ -3072,17 +3080,9 @@ defm UMINV : SIMDAcrossLanesBHS<1, 0b11010, "uminv">;
defm SADDLV : SIMDAcrossLanesHSD<0, 0b00011, "saddlv">;
defm UADDLV : SIMDAcrossLanesHSD<1, 0b00011, "uaddlv">;
defm FMAXNMV : SIMDAcrossLanesS<0b01100, 0, "fmaxnmv", int_arm64_neon_fmaxnmv>;
def : Pat<(f32 (int_arm64_neon_fmaxnmv (v2f32 V64:$Rn))),
(EXTRACT_SUBREG (FMAXNMPv2f32 V64:$Rn, V64:$Rn), ssub)>;
defm FMAXV : SIMDAcrossLanesS<0b01111, 0, "fmaxv", int_arm64_neon_fmaxv>;
def : Pat<(f32 (int_arm64_neon_fmaxv (v2f32 V64:$Rn))),
(EXTRACT_SUBREG (FMAXPv2f32 V64:$Rn, V64:$Rn), ssub)>;
defm FMINNMV : SIMDAcrossLanesS<0b01100, 1, "fminnmv", int_arm64_neon_fminnmv>;
def : Pat<(f32 (int_arm64_neon_fminnmv (v2f32 V64:$Rn))),
(EXTRACT_SUBREG (FMINNMPv2f32 V64:$Rn, V64:$Rn), ssub)>;
defm FMINV : SIMDAcrossLanesS<0b01111, 1, "fminv", int_arm64_neon_fminv>;
def : Pat<(f32 (int_arm64_neon_fminv (v2f32 V64:$Rn))),
(EXTRACT_SUBREG (FMINPv2f32 V64:$Rn, V64:$Rn), ssub)>;
multiclass SIMDAcrossLanesSignedIntrinsic<string baseOpc, Intrinsic intOp> {
// If there is a sign extension after this intrinsic, consume it as smov already

101
test/CodeGen/ARM64/fminv.ll Normal file
View File

@ -0,0 +1,101 @@
; RUN: llc -mtriple=arm64-linux-gnu -o - %s | FileCheck %s
define float @test_fminv_v2f32(<2 x float> %in) {
; CHECK: test_fminv_v2f32:
; CHECK: fminp s0, v0.2s
%min = call float @llvm.arm64.neon.fminv.f32.v2f32(<2 x float> %in)
ret float %min
}
define float @test_fminv_v4f32(<4 x float> %in) {
; CHECK: test_fminv_v4f32:
; CHECK: fminv s0, v0.4s
%min = call float @llvm.arm64.neon.fminv.f32.v4f32(<4 x float> %in)
ret float %min
}
define double @test_fminv_v2f64(<2 x double> %in) {
; CHECK: test_fminv_v2f64:
; CHECK: fminp d0, v0.2d
%min = call double @llvm.arm64.neon.fminv.f64.v2f64(<2 x double> %in)
ret double %min
}
declare float @llvm.arm64.neon.fminv.f32.v2f32(<2 x float>)
declare float @llvm.arm64.neon.fminv.f32.v4f32(<4 x float>)
declare double @llvm.arm64.neon.fminv.f64.v2f64(<2 x double>)
define float @test_fmaxv_v2f32(<2 x float> %in) {
; CHECK: test_fmaxv_v2f32:
; CHECK: fmaxp s0, v0.2s
%max = call float @llvm.arm64.neon.fmaxv.f32.v2f32(<2 x float> %in)
ret float %max
}
define float @test_fmaxv_v4f32(<4 x float> %in) {
; CHECK: test_fmaxv_v4f32:
; CHECK: fmaxv s0, v0.4s
%max = call float @llvm.arm64.neon.fmaxv.f32.v4f32(<4 x float> %in)
ret float %max
}
define double @test_fmaxv_v2f64(<2 x double> %in) {
; CHECK: test_fmaxv_v2f64:
; CHECK: fmaxp d0, v0.2d
%max = call double @llvm.arm64.neon.fmaxv.f64.v2f64(<2 x double> %in)
ret double %max
}
declare float @llvm.arm64.neon.fmaxv.f32.v2f32(<2 x float>)
declare float @llvm.arm64.neon.fmaxv.f32.v4f32(<4 x float>)
declare double @llvm.arm64.neon.fmaxv.f64.v2f64(<2 x double>)
define float @test_fminnmv_v2f32(<2 x float> %in) {
; CHECK: test_fminnmv_v2f32:
; CHECK: fminnmp s0, v0.2s
%minnm = call float @llvm.arm64.neon.fminnmv.f32.v2f32(<2 x float> %in)
ret float %minnm
}
define float @test_fminnmv_v4f32(<4 x float> %in) {
; CHECK: test_fminnmv_v4f32:
; CHECK: fminnmv s0, v0.4s
%minnm = call float @llvm.arm64.neon.fminnmv.f32.v4f32(<4 x float> %in)
ret float %minnm
}
define double @test_fminnmv_v2f64(<2 x double> %in) {
; CHECK: test_fminnmv_v2f64:
; CHECK: fminnmp d0, v0.2d
%minnm = call double @llvm.arm64.neon.fminnmv.f64.v2f64(<2 x double> %in)
ret double %minnm
}
declare float @llvm.arm64.neon.fminnmv.f32.v2f32(<2 x float>)
declare float @llvm.arm64.neon.fminnmv.f32.v4f32(<4 x float>)
declare double @llvm.arm64.neon.fminnmv.f64.v2f64(<2 x double>)
define float @test_fmaxnmv_v2f32(<2 x float> %in) {
; CHECK: test_fmaxnmv_v2f32:
; CHECK: fmaxnmp s0, v0.2s
%maxnm = call float @llvm.arm64.neon.fmaxnmv.f32.v2f32(<2 x float> %in)
ret float %maxnm
}
define float @test_fmaxnmv_v4f32(<4 x float> %in) {
; CHECK: test_fmaxnmv_v4f32:
; CHECK: fmaxnmv s0, v0.4s
%maxnm = call float @llvm.arm64.neon.fmaxnmv.f32.v4f32(<4 x float> %in)
ret float %maxnm
}
define double @test_fmaxnmv_v2f64(<2 x double> %in) {
; CHECK: test_fmaxnmv_v2f64:
; CHECK: fmaxnmp d0, v0.2d
%maxnm = call double @llvm.arm64.neon.fmaxnmv.f64.v2f64(<2 x double> %in)
ret double %maxnm
}
declare float @llvm.arm64.neon.fmaxnmv.f32.v2f32(<2 x float>)
declare float @llvm.arm64.neon.fmaxnmv.f32.v4f32(<4 x float>)
declare double @llvm.arm64.neon.fmaxnmv.f64.v2f64(<2 x double>)