[AArch64][CodeGen] Restrict bfloat vector operations to what's actually supported

Previously in addTypeForNeon, we would set the operations for bfloat vectors
like other generic types. But as bfloat is a storage-only type a number of
operations shouldn't be set. This patch fixes that.

Reviewed By: dmgreen

Differential Revision: https://reviews.llvm.org/D85101

diff --git a/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp b/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
index b04d245..c904fc7 100644 (file)
--- a/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
+++ b/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
@@ -176,7 +176,8 @@ AArch64TargetLowering::AArch64TargetLowering(const TargetMachine &TM,
     addDRTypeForNEON(MVT::v1i64);
     addDRTypeForNEON(MVT::v1f64);
     addDRTypeForNEON(MVT::v4f16);
-    addDRTypeForNEON(MVT::v4bf16);
+    if (Subtarget->hasBF16())
+      addDRTypeForNEON(MVT::v4bf16);
 
     addQRTypeForNEON(MVT::v4f32);
     addQRTypeForNEON(MVT::v2f64);
@@ -185,7 +186,8 @@ AArch64TargetLowering::AArch64TargetLowering(const TargetMachine &TM,
     addQRTypeForNEON(MVT::v4i32);
     addQRTypeForNEON(MVT::v2i64);
     addQRTypeForNEON(MVT::v8f16);
-    addQRTypeForNEON(MVT::v8bf16);
+    if (Subtarget->hasBF16())
+      addQRTypeForNEON(MVT::v8bf16);
   }
 
   if (Subtarget->hasSVE()) {
@@ -1096,6 +1098,7 @@ void AArch64TargetLowering::addTypeForNEON(MVT VT, MVT PromotedBitwiseVT) {
 
   // F[MIN|MAX][NUM|NAN] are available for all FP NEON types.
   if (VT.isFloatingPoint() &&
+      VT.getVectorElementType() != MVT::bf16 &&
       (VT.getVectorElementType() != MVT::f16 || Subtarget->hasFullFP16()))
     for (unsigned Opcode :
          {ISD::FMINIMUM, ISD::FMAXIMUM, ISD::FMINNUM, ISD::FMAXNUM})

diff --git a/llvm/test/CodeGen/AArch64/bf16-vector-bitcast.ll b/llvm/test/CodeGen/AArch64/bf16-vector-bitcast.ll
index d59f135..2ca14d2 100644 (file)
--- a/llvm/test/CodeGen/AArch64/bf16-vector-bitcast.ll
+++ b/llvm/test/CodeGen/AArch64/bf16-vector-bitcast.ll
@@ -1,4 +1,4 @@
-; RUN: llc < %s -asm-verbose=0 -mtriple=aarch64-none-eabi | FileCheck %s
+; RUN: llc < %s -asm-verbose=0 -mtriple=aarch64-none-eabi -mattr=+bf16 | FileCheck %s
 
 define <4 x i16> @v4bf16_to_v4i16(float, <4 x bfloat> %a) nounwind {
 ; CHECK-LABEL: v4bf16_to_v4i16:

diff --git a/llvm/test/CodeGen/AArch64/bf16-vector-shuffle.ll b/llvm/test/CodeGen/AArch64/bf16-vector-shuffle.ll
index 5aca916..bdc9837 100644 (file)
--- a/llvm/test/CodeGen/AArch64/bf16-vector-shuffle.ll
+++ b/llvm/test/CodeGen/AArch64/bf16-vector-shuffle.ll
@@ -1,5 +1,5 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
-; RUN: llc < %s -asm-verbose=0 -mtriple=aarch64-none-eabi | FileCheck %s
+; RUN: llc < %s -asm-verbose=0 -mtriple=aarch64-none-eabi -mattr=+bf16 | FileCheck %s
 
 ; bfloat16x4_t test_vcreate_bf16(uint64_t a) { return vcreate_bf16(a); }
 define <4 x bfloat> @test_vcreate_bf16(i64 %a) nounwind {

diff --git a/llvm/test/CodeGen/AArch64/bf16.ll b/llvm/test/CodeGen/AArch64/bf16.ll
index f1464ea..f8cedcc 100644 (file)
--- a/llvm/test/CodeGen/AArch64/bf16.ll
+++ b/llvm/test/CodeGen/AArch64/bf16.ll
@@ -1,5 +1,5 @@
-; RUN: llc < %s -asm-verbose=0 -mtriple=arm64-eabi | FileCheck %s
-; RUN: llc < %s -asm-verbose=0 -mtriple=aarch64-eabi | FileCheck %s
+; RUN: llc < %s -asm-verbose=0 -mtriple=arm64-eabi -mattr=+bf16 | FileCheck %s
+; RUN: llc < %s -asm-verbose=0 -mtriple=aarch64-eabi -mattr=+bf16 | FileCheck %s
 
 ; test argument passing and simple load/store