[AArch64] lower abs intrinsic to new ABS instruction in SelDag

When feature CSSC is available, the SelectionDag abs intrinsic should map to the
new scalar ABS instruction.

Additionally, the SIMDTwoScalarD tablegen defm includes a pattern match for
scalar i64, which we don't want to use when CSSC is enabled.

spec:
https://developer.arm.com/documentation/ddi0602/2022-09/Base-Instructions/ABS--Absolute-value-

Reviewed By: lenary

Differential Revision: https://reviews.llvm.org/D138812

diff --git a/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp b/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
index a3b82d0..862ed37 100644 (file)
--- a/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
+++ b/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
@@ -545,10 +545,15 @@ AArch64TargetLowering::AArch64TargetLowering(const TargetMachine &TM,
     setOperationAction(ISD::CTPOP, MVT::i32, Legal);
     setOperationAction(ISD::CTPOP, MVT::i64, Legal);
     setOperationAction(ISD::CTPOP, MVT::i128, Expand);
+
     setOperationAction(ISD::PARITY, MVT::i128, Expand);
+
     setOperationAction(ISD::CTTZ, MVT::i32, Legal);
     setOperationAction(ISD::CTTZ, MVT::i64, Legal);
     setOperationAction(ISD::CTTZ, MVT::i128, Expand);
+
+    setOperationAction(ISD::ABS, MVT::i32, Legal);
+    setOperationAction(ISD::ABS, MVT::i64, Legal);
   } else {
     setOperationAction(ISD::CTPOP, MVT::i32, Custom);
     setOperationAction(ISD::CTPOP, MVT::i64, Custom);
@@ -556,10 +561,10 @@ AArch64TargetLowering::AArch64TargetLowering(const TargetMachine &TM,
 
     setOperationAction(ISD::PARITY, MVT::i64, Custom);
     setOperationAction(ISD::PARITY, MVT::i128, Custom);
-  }
 
-  setOperationAction(ISD::ABS, MVT::i32, Custom);
-  setOperationAction(ISD::ABS, MVT::i64, Custom);
+    setOperationAction(ISD::ABS, MVT::i32, Custom);
+    setOperationAction(ISD::ABS, MVT::i64, Custom);
+  }
 
   setOperationAction(ISD::SDIVREM, MVT::i32, Expand);
   setOperationAction(ISD::SDIVREM, MVT::i64, Expand);

diff --git a/llvm/lib/Target/AArch64/AArch64InstrFormats.td b/llvm/lib/Target/AArch64/AArch64InstrFormats.td
index ee3761f..016563c 100644 (file)
--- a/llvm/lib/Target/AArch64/AArch64InstrFormats.td
+++ b/llvm/lib/Target/AArch64/AArch64InstrFormats.td
@@ -7265,12 +7265,15 @@ multiclass SIMDFPCmpTwoScalar<bit U, bit S, bits<5> opc, string asm,
 }
 
 multiclass SIMDTwoScalarD<bit U, bits<5> opc, string asm,
-                          SDPatternOperator OpNode = null_frag> {
+                          SDPatternOperator OpNode = null_frag,
+                          list<Predicate> preds = []> {
   def v1i64       : BaseSIMDTwoScalar<U, 0b11, 0b00, opc, FPR64, FPR64, asm,
     [(set (v1i64 FPR64:$Rd), (OpNode (v1i64 FPR64:$Rn)))]>;
 
+  let Predicates = preds in {
   def : Pat<(i64 (OpNode (i64 FPR64:$Rn))),
             (!cast<Instruction>(NAME # "v1i64") FPR64:$Rn)>;
+  }
 }
 
 let mayRaiseFPException = 1, Uses = [FPCR] in

diff --git a/llvm/lib/Target/AArch64/AArch64InstrInfo.td b/llvm/lib/Target/AArch64/AArch64InstrInfo.td
index 409c9a9..55f88f4 100644 (file)
--- a/llvm/lib/Target/AArch64/AArch64InstrInfo.td
+++ b/llvm/lib/Target/AArch64/AArch64InstrInfo.td
@@ -115,8 +115,9 @@ def HasDotProd       : Predicate<"Subtarget->hasDotProd()">,
                                  AssemblerPredicateWithAll<(all_of FeatureDotProd), "dotprod">;
 def HasCRC           : Predicate<"Subtarget->hasCRC()">,
                                  AssemblerPredicateWithAll<(all_of FeatureCRC), "crc">;
-def HasCSSC           : Predicate<"Subtarget->hasCSSC()">,
+def HasCSSC          : Predicate<"Subtarget->hasCSSC()">,
                                  AssemblerPredicateWithAll<(all_of FeatureCSSC), "cssc">;
+def HasNoCSSC        : Predicate<"!Subtarget->hasCSSC()">;
 def HasLSE           : Predicate<"Subtarget->hasLSE()">,
                                  AssemblerPredicateWithAll<(all_of FeatureLSE), "lse">;
 def HasNoLSE         : Predicate<"!Subtarget->hasLSE()">;
@@ -5092,7 +5093,7 @@ def : Pat<(i64 (int_aarch64_neon_sqsub (i64 FPR64:$Rd),
 // Advanced SIMD two scalar instructions.
 //===----------------------------------------------------------------------===//
 
-defm ABS    : SIMDTwoScalarD<    0, 0b01011, "abs", abs>;
+defm ABS    : SIMDTwoScalarD<    0, 0b01011, "abs", abs, [HasNoCSSC]>;
 defm CMEQ   : SIMDCmpTwoScalarD< 0, 0b01001, "cmeq", AArch64cmeqz>;
 defm CMGE   : SIMDCmpTwoScalarD< 1, 0b01000, "cmge", AArch64cmgez>;
 defm CMGT   : SIMDCmpTwoScalarD< 0, 0b01000, "cmgt", AArch64cmgtz>;
@@ -8554,7 +8555,7 @@ defm RCWSWP  : ReadCheckWriteOperation<0b010, "swp">;
 //===----------------------------------------------------------------------===//
 // General Data-Processing Instructions (FEAT_V94_DP)
 //===----------------------------------------------------------------------===//
-defm ABS : OneOperandData<0b001000, "abs">, Requires<[HasCSSC]>;
+defm ABS : OneOperandData<0b001000, "abs", abs>, Requires<[HasCSSC]>;
 defm CNT : OneOperandData<0b000111, "cnt", ctpop>, Requires<[HasCSSC]>;
 defm CTZ : OneOperandData<0b000110, "ctz", cttz>, Requires<[HasCSSC]>;
 

diff --git a/llvm/test/CodeGen/AArch64/iabs.ll b/llvm/test/CodeGen/AArch64/iabs.ll
index dbcd922..bbb04c7 100644 (file)
--- a/llvm/test/CodeGen/AArch64/iabs.ll
+++ b/llvm/test/CodeGen/AArch64/iabs.ll
@@ -1,12 +1,20 @@
-; RUN: llc < %s -mtriple=arm64-eabi -aarch64-neon-syntax=apple | FileCheck %s
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc < %s -mtriple=arm64-eabi | FileCheck %s
+; RUN: llc < %s -mtriple=arm64-eabi -mattr=+cssc | FileCheck %s -check-prefix=CHECK-CSSC
 
 define i8 @test_i8(i8 %a) nounwind {
 ; CHECK-LABEL: test_i8:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:  sxtb w8, w0
-; CHECK-NEXT:  cmp w8, #0
-; CHECK-NEXT:  cneg w0, w8, mi
-; CHECK-NEXT:  ret
+; CHECK-NEXT:    sxtb w8, w0
+; CHECK-NEXT:    cmp w8, #0
+; CHECK-NEXT:    cneg w0, w8, mi
+; CHECK-NEXT:    ret
+;
+; CHECK-CSSC-LABEL: test_i8:
+; CHECK-CSSC:       // %bb.0:
+; CHECK-CSSC-NEXT:    sxtb w8, w0
+; CHECK-CSSC-NEXT:    abs w0, w8
+; CHECK-CSSC-NEXT:    ret
   %tmp1neg = sub i8 0, %a
   %b = icmp sgt i8 %a, -1
   %abs = select i1 %b, i8 %a, i8 %tmp1neg
@@ -16,10 +24,16 @@ define i8 @test_i8(i8 %a) nounwind {
 define i16 @test_i16(i16 %a) nounwind {
 ; CHECK-LABEL: test_i16:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:  sxth w8, w0
-; CHECK-NEXT:  cmp w8, #0
-; CHECK-NEXT:  cneg w0, w8, mi
-; CHECK-NEXT:  ret
+; CHECK-NEXT:    sxth w8, w0
+; CHECK-NEXT:    cmp w8, #0
+; CHECK-NEXT:    cneg w0, w8, mi
+; CHECK-NEXT:    ret
+;
+; CHECK-CSSC-LABEL: test_i16:
+; CHECK-CSSC:       // %bb.0:
+; CHECK-CSSC-NEXT:    sxth w8, w0
+; CHECK-CSSC-NEXT:    abs w0, w8
+; CHECK-CSSC-NEXT:    ret
   %tmp1neg = sub i16 0, %a
   %b = icmp sgt i16 %a, -1
   %abs = select i1 %b, i16 %a, i16 %tmp1neg
@@ -29,9 +43,14 @@ define i16 @test_i16(i16 %a) nounwind {
 define i32 @test_i32(i32 %a) nounwind {
 ; CHECK-LABEL: test_i32:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:  cmp w0, #0
-; CHECK-NEXT:  cneg w0, w0, mi
-; CHECK-NEXT:  ret
+; CHECK-NEXT:    cmp w0, #0
+; CHECK-NEXT:    cneg w0, w0, mi
+; CHECK-NEXT:    ret
+;
+; CHECK-CSSC-LABEL: test_i32:
+; CHECK-CSSC:       // %bb.0:
+; CHECK-CSSC-NEXT:    abs w0, w0
+; CHECK-CSSC-NEXT:    ret
   %tmp1neg = sub i32 0, %a
   %b = icmp sgt i32 %a, -1
   %abs = select i1 %b, i32 %a, i32 %tmp1neg
@@ -41,12 +60,16 @@ define i32 @test_i32(i32 %a) nounwind {
 define i64 @test_i64(i64 %a) nounwind {
 ; CHECK-LABEL: test_i64:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:  cmp x0, #0
-; CHECK-NEXT:  cneg x0, x0, mi
-; CHECK-NEXT:  ret
+; CHECK-NEXT:    cmp x0, #0
+; CHECK-NEXT:    cneg x0, x0, mi
+; CHECK-NEXT:    ret
+;
+; CHECK-CSSC-LABEL: test_i64:
+; CHECK-CSSC:       // %bb.0:
+; CHECK-CSSC-NEXT:    abs x0, x0
+; CHECK-CSSC-NEXT:    ret
   %tmp1neg = sub i64 0, %a
   %b = icmp sgt i64 %a, -1
   %abs = select i1 %b, i64 %a, i64 %tmp1neg
   ret i64 %abs
 }
-