[llvm] dc9a771 - [AArch64] Add a basic pattern for FACGE and FACGT
David Green via llvm-commits
llvm-commits at lists.llvm.org
Mon Apr 17 08:01:03 PDT 2023
Author: David Green
Date: 2023-04-17T16:00:57+01:00
New Revision: dc9a7719facacd23be3acf5f5d437797560d89c3
URL: https://github.com/llvm/llvm-project/commit/dc9a7719facacd23be3acf5f5d437797560d89c3
DIFF: https://github.com/llvm/llvm-project/commit/dc9a7719facacd23be3acf5f5d437797560d89c3.diff
LOG: [AArch64] Add a basic pattern for FACGE and FACGT
These instructions take the absolute of two values and perform a FCMGE/FCMGT
between them. This patch adds some basic patterns for them.
Differential Revision: https://reviews.llvm.org/D147945
Added:
Modified:
llvm/lib/Target/AArch64/AArch64InstrInfo.td
llvm/test/CodeGen/AArch64/vacg.ll
Removed:
################################################################################
diff --git a/llvm/lib/Target/AArch64/AArch64InstrInfo.td b/llvm/lib/Target/AArch64/AArch64InstrInfo.td
index ba7d2f65d209a..45c8edc2593a0 100644
--- a/llvm/lib/Target/AArch64/AArch64InstrInfo.td
+++ b/llvm/lib/Target/AArch64/AArch64InstrInfo.td
@@ -774,6 +774,12 @@ def AArch64faddp : PatFrags<(ops node:$Rn, node:$Rm),
[(AArch64addp_n node:$Rn, node:$Rm),
(int_aarch64_neon_faddp node:$Rn, node:$Rm)]>;
def AArch64roundingvlshr : ComplexPattern<vAny, 2, "SelectRoundingVLShr", [AArch64vlshr]>;
+def AArch64facge : PatFrags<(ops node:$Rn, node:$Rm),
+ [(AArch64fcmge (fabs node:$Rn), (fabs node:$Rm)),
+ (int_aarch64_neon_facge node:$Rn, node:$Rm)]>;
+def AArch64facgt : PatFrags<(ops node:$Rn, node:$Rm),
+ [(AArch64fcmgt (fabs node:$Rn), (fabs node:$Rm)),
+ (int_aarch64_neon_facgt node:$Rn, node:$Rm)]>;
def SDT_AArch64SETTAG : SDTypeProfile<0, 2, [SDTCisPtrTy<0>, SDTCisPtrTy<1>]>;
def AArch64stg : SDNode<"AArch64ISD::STG", SDT_AArch64SETTAG, [SDNPHasChain, SDNPMayStore, SDNPMemOperand]>;
@@ -4874,8 +4880,8 @@ let Predicates = [HasNEON, HasFullFP16] in {
foreach VT = [ v4f16, v8f16 ] in
def : Pat<(fabs (fsub VT:$Rn, VT:$Rm)), (!cast<Instruction>("FABD"#VT) VT:$Rn, VT:$Rm)>;
}
-defm FACGE : SIMDThreeSameVectorFPCmp<1,0,0b101,"facge",int_aarch64_neon_facge>;
-defm FACGT : SIMDThreeSameVectorFPCmp<1,1,0b101,"facgt",int_aarch64_neon_facgt>;
+defm FACGE : SIMDThreeSameVectorFPCmp<1,0,0b101,"facge",AArch64facge>;
+defm FACGT : SIMDThreeSameVectorFPCmp<1,1,0b101,"facgt",AArch64facgt>;
defm FADDP : SIMDThreeSameVectorFP<1,0,0b010,"faddp", AArch64faddp>;
defm FADD : SIMDThreeSameVectorFP<0,0,0b010,"fadd", any_fadd>;
defm FCMEQ : SIMDThreeSameVectorFPCmp<0, 0, 0b100, "fcmeq", AArch64fcmeq>;
diff --git a/llvm/test/CodeGen/AArch64/vacg.ll b/llvm/test/CodeGen/AArch64/vacg.ll
index 346defa875a49..402d260806675 100644
--- a/llvm/test/CodeGen/AArch64/vacg.ll
+++ b/llvm/test/CodeGen/AArch64/vacg.ll
@@ -5,9 +5,7 @@
define <4 x i32> @gt_v4f32(<4 x float> %a, <4 x float> %b) {
; CHECK-LABEL: gt_v4f32:
; CHECK: // %bb.0: // %entry
-; CHECK-NEXT: fabs v0.4s, v0.4s
-; CHECK-NEXT: fabs v1.4s, v1.4s
-; CHECK-NEXT: fcmgt v0.4s, v0.4s, v1.4s
+; CHECK-NEXT: facgt v0.4s, v0.4s, v1.4s
; CHECK-NEXT: ret
entry:
%vabs1.i = tail call <4 x float> @llvm.fabs.v4f32(<4 x float> %a)
@@ -20,9 +18,7 @@ entry:
define <4 x i32> @ge_v4f32(<4 x float> %a, <4 x float> %b) {
; CHECK-LABEL: ge_v4f32:
; CHECK: // %bb.0: // %entry
-; CHECK-NEXT: fabs v0.4s, v0.4s
-; CHECK-NEXT: fabs v1.4s, v1.4s
-; CHECK-NEXT: fcmge v0.4s, v0.4s, v1.4s
+; CHECK-NEXT: facge v0.4s, v0.4s, v1.4s
; CHECK-NEXT: ret
entry:
%vabs1.i = tail call <4 x float> @llvm.fabs.v4f32(<4 x float> %a)
@@ -35,9 +31,7 @@ entry:
define <4 x i32> @lt_v4f32(<4 x float> %a, <4 x float> %b) {
; CHECK-LABEL: lt_v4f32:
; CHECK: // %bb.0: // %entry
-; CHECK-NEXT: fabs v0.4s, v0.4s
-; CHECK-NEXT: fabs v1.4s, v1.4s
-; CHECK-NEXT: fcmgt v0.4s, v1.4s, v0.4s
+; CHECK-NEXT: facgt v0.4s, v1.4s, v0.4s
; CHECK-NEXT: ret
entry:
%vabs1.i = tail call <4 x float> @llvm.fabs.v4f32(<4 x float> %a)
@@ -50,9 +44,7 @@ entry:
define <4 x i32> @le_v4f32(<4 x float> %a, <4 x float> %b) {
; CHECK-LABEL: le_v4f32:
; CHECK: // %bb.0: // %entry
-; CHECK-NEXT: fabs v0.4s, v0.4s
-; CHECK-NEXT: fabs v1.4s, v1.4s
-; CHECK-NEXT: fcmge v0.4s, v1.4s, v0.4s
+; CHECK-NEXT: facge v0.4s, v1.4s, v0.4s
; CHECK-NEXT: ret
entry:
%vabs1.i = tail call <4 x float> @llvm.fabs.v4f32(<4 x float> %a)
@@ -65,9 +57,7 @@ entry:
define <2 x i32> @gt_v2f32(<2 x float> %a, <2 x float> %b) {
; CHECK-LABEL: gt_v2f32:
; CHECK: // %bb.0: // %entry
-; CHECK-NEXT: fabs v0.2s, v0.2s
-; CHECK-NEXT: fabs v1.2s, v1.2s
-; CHECK-NEXT: fcmgt v0.2s, v0.2s, v1.2s
+; CHECK-NEXT: facgt v0.2s, v0.2s, v1.2s
; CHECK-NEXT: ret
entry:
%vabs1.i = tail call <2 x float> @llvm.fabs.v2f32(<2 x float> %a)
@@ -80,9 +70,7 @@ entry:
define <2 x i32> @ge_v2f32(<2 x float> %a, <2 x float> %b) {
; CHECK-LABEL: ge_v2f32:
; CHECK: // %bb.0: // %entry
-; CHECK-NEXT: fabs v0.2s, v0.2s
-; CHECK-NEXT: fabs v1.2s, v1.2s
-; CHECK-NEXT: fcmge v0.2s, v0.2s, v1.2s
+; CHECK-NEXT: facge v0.2s, v0.2s, v1.2s
; CHECK-NEXT: ret
entry:
%vabs1.i = tail call <2 x float> @llvm.fabs.v2f32(<2 x float> %a)
@@ -95,9 +83,7 @@ entry:
define <4 x i16> @gt_v4f16(<4 x half> %a, <4 x half> %b) {
; CHECK-LABEL: gt_v4f16:
; CHECK: // %bb.0: // %entry
-; CHECK-NEXT: fabs v0.4h, v0.4h
-; CHECK-NEXT: fabs v1.4h, v1.4h
-; CHECK-NEXT: fcmgt v0.4h, v0.4h, v1.4h
+; CHECK-NEXT: facgt v0.4h, v0.4h, v1.4h
; CHECK-NEXT: ret
entry:
%vabs1.i = tail call <4 x half> @llvm.fabs.v4f16(<4 x half> %a)
@@ -110,9 +96,7 @@ entry:
define <4 x i16> @ge_v4f16(<4 x half> %a, <4 x half> %b) {
; CHECK-LABEL: ge_v4f16:
; CHECK: // %bb.0: // %entry
-; CHECK-NEXT: fabs v0.4h, v0.4h
-; CHECK-NEXT: fabs v1.4h, v1.4h
-; CHECK-NEXT: fcmge v0.4h, v0.4h, v1.4h
+; CHECK-NEXT: facge v0.4h, v0.4h, v1.4h
; CHECK-NEXT: ret
entry:
%vabs1.i = tail call <4 x half> @llvm.fabs.v4f16(<4 x half> %a)
@@ -125,9 +109,7 @@ entry:
define <8 x i16> @gt_v8f16(<8 x half> %a, <8 x half> %b) {
; CHECK-LABEL: gt_v8f16:
; CHECK: // %bb.0: // %entry
-; CHECK-NEXT: fabs v0.8h, v0.8h
-; CHECK-NEXT: fabs v1.8h, v1.8h
-; CHECK-NEXT: fcmgt v0.8h, v0.8h, v1.8h
+; CHECK-NEXT: facgt v0.8h, v0.8h, v1.8h
; CHECK-NEXT: ret
entry:
%vabs1.i = tail call <8 x half> @llvm.fabs.v8f16(<8 x half> %a)
@@ -140,9 +122,7 @@ entry:
define <8 x i16> @ge_v8f16(<8 x half> %a, <8 x half> %b) {
; CHECK-LABEL: ge_v8f16:
; CHECK: // %bb.0: // %entry
-; CHECK-NEXT: fabs v0.8h, v0.8h
-; CHECK-NEXT: fabs v1.8h, v1.8h
-; CHECK-NEXT: fcmge v0.8h, v0.8h, v1.8h
+; CHECK-NEXT: facge v0.8h, v0.8h, v1.8h
; CHECK-NEXT: ret
entry:
%vabs1.i = tail call <8 x half> @llvm.fabs.v8f16(<8 x half> %a)
@@ -155,9 +135,7 @@ entry:
define <2 x i64> @gt_v2f64(<2 x double> %a, <2 x double> %b) {
; CHECK-LABEL: gt_v2f64:
; CHECK: // %bb.0: // %entry
-; CHECK-NEXT: fabs v0.2d, v0.2d
-; CHECK-NEXT: fabs v1.2d, v1.2d
-; CHECK-NEXT: fcmgt v0.2d, v0.2d, v1.2d
+; CHECK-NEXT: facgt v0.2d, v0.2d, v1.2d
; CHECK-NEXT: ret
entry:
%vabs1.i = tail call <2 x double> @llvm.fabs.v2f64(<2 x double> %a)
@@ -170,9 +148,7 @@ entry:
define <2 x i64> @ge_v2f64(<2 x double> %a, <2 x double> %b) {
; CHECK-LABEL: ge_v2f64:
; CHECK: // %bb.0: // %entry
-; CHECK-NEXT: fabs v0.2d, v0.2d
-; CHECK-NEXT: fabs v1.2d, v1.2d
-; CHECK-NEXT: fcmge v0.2d, v0.2d, v1.2d
+; CHECK-NEXT: facge v0.2d, v0.2d, v1.2d
; CHECK-NEXT: ret
entry:
%vabs1.i = tail call <2 x double> @llvm.fabs.v2f64(<2 x double> %a)
More information about the llvm-commits
mailing list