[AArch64] Add missing bitcast patterns for bf16<->f16 converts.

[davemgreen]

This fills in the missing patterns for bicasting v4f16 to/from v4bf16, and v8f16 to/from v8f16. Clean up some formatting whilst here.

Fixes #159772

[llvmbot]

@llvm/pr-subscribers-backend-aarch64

Author: David Green (davemgreen)

Changes

This fills in the missing patterns for bicasting v4f16 to/from v4bf16, and v8f16 to/from v8f16. Clean up some formatting whilst here.

Fixes #159772

---

Full diff: https://github.com/llvm/llvm-project/pull/159816.diff

2 Files Affected:

• (modified) llvm/lib/Target/AArch64/AArch64InstrInfo.td (+16-4)
• (modified) llvm/test/CodeGen/AArch64/bf16-vector-bitcast.ll (+40)

diff --git a/llvm/lib/Target/AArch64/AArch64InstrInfo.td b/llvm/lib/Target/AArch64/AArch64InstrInfo.td
index 3fcafc6d35090..05e3888a28dde 100644
--- a/llvm/lib/Target/AArch64/AArch64InstrInfo.td
+++ b/llvm/lib/Target/AArch64/AArch64InstrInfo.td
@@ -9857,8 +9857,14 @@ def : Pat<(v4bf16 (bitconvert (v2f32 FPR64:$src))),
 def : Pat<(v4bf16 (bitconvert (v1f64 FPR64:$src))),
                              (v4bf16 (REV64v4i16 FPR64:$src))>;
 }
-def : Pat<(v4f16 (bitconvert (v4i16 FPR64:$src))), (v4f16 FPR64:$src)>;
-def : Pat<(v4bf16 (bitconvert (v4i16 FPR64:$src))), (v4bf16 FPR64:$src)>;
+def : Pat<(v4f16 (bitconvert (v4i16 FPR64:$src))),
+          (v4f16 FPR64:$src)>;
+def : Pat<(v4f16 (bitconvert (v4bf16 FPR64:$src))),
+          (v4f16 FPR64:$src)>;
+def : Pat<(v4bf16 (bitconvert (v4i16 FPR64:$src))),
+          (v4bf16 FPR64:$src)>;
+def : Pat<(v4bf16 (bitconvert (v4f16 FPR64:$src))),
+          (v4bf16 FPR64:$src)>;
 
 let Predicates = [IsLE] in {
 def : Pat<(v8i8  (bitconvert (v1i64 FPR64:$src))), (v8i8  FPR64:$src)>;
@@ -10187,8 +10193,14 @@ def : Pat<(v8bf16 (bitconvert (v2f64 FPR128:$src))),
 def : Pat<(v8bf16 (bitconvert (v4f32 FPR128:$src))),
                              (v8bf16 (REV32v8i16 FPR128:$src))>;
 }
-def : Pat<(v8f16 (bitconvert (v8i16 FPR128:$src))), (v8f16 FPR128:$src)>;
-def : Pat<(v8bf16 (bitconvert (v8i16 FPR128:$src))), (v8bf16 FPR128:$src)>;
+def : Pat<(v8f16 (bitconvert (v8i16 FPR128:$src))),
+          (v8f16 FPR128:$src)>;
+def : Pat<(v8bf16 (bitconvert (v8i16 FPR128:$src))),
+          (v8bf16 FPR128:$src)>;
+def : Pat<(v8f16 (bitconvert (v8bf16 FPR128:$src))),
+          (v8f16 FPR128:$src)>;
+def : Pat<(v8bf16 (bitconvert (v8f16 FPR128:$src))),
+          (v8bf16 FPR128:$src)>;
 
 let Predicates = [IsLE] in {
 def : Pat<(v16i8 (bitconvert (f128  FPR128:$src))), (v16i8 FPR128:$src)>;
diff --git a/llvm/test/CodeGen/AArch64/bf16-vector-bitcast.ll b/llvm/test/CodeGen/AArch64/bf16-vector-bitcast.ll
index 1c216e7357215..e371748a43b29 100644
--- a/llvm/test/CodeGen/AArch64/bf16-vector-bitcast.ll
+++ b/llvm/test/CodeGen/AArch64/bf16-vector-bitcast.ll
@@ -11,6 +11,16 @@ entry:
   ret <4 x i16> %1
 }
 
+define <4 x half> @v4bf16_to_v4f16(float, <4 x bfloat> %a) nounwind {
+; CHECK-LABEL: v4bf16_to_v4f16:
+; CHECK:       // %bb.0: // %entry
+; CHECK-NEXT:    fmov d0, d1
+; CHECK-NEXT:    ret
+entry:
+  %1 = bitcast <4 x bfloat> %a to <4 x half>
+  ret <4 x half> %1
+}
+
 define <2 x i32> @v4bf16_to_v2i32(float, <4 x bfloat> %a) nounwind {
 ; CHECK-LABEL: v4bf16_to_v2i32:
 ; CHECK:       // %bb.0: // %entry
@@ -82,6 +92,16 @@ entry:
   ret <4 x bfloat> %1
 }
 
+define <4 x bfloat> @v4f16_to_v4bf16(float, <4 x half> %a) nounwind {
+; CHECK-LABEL: v4f16_to_v4bf16:
+; CHECK:       // %bb.0: // %entry
+; CHECK-NEXT:    fmov d0, d1
+; CHECK-NEXT:    ret
+entry:
+  %1 = bitcast <4 x half> %a to <4 x bfloat>
+  ret <4 x bfloat> %1
+}
+
 define <4 x bfloat> @v2i32_to_v4bf16(float, <2 x i32> %a) nounwind {
 ; CHECK-LABEL: v2i32_to_v4bf16:
 ; CHECK:       // %bb.0: // %entry
@@ -152,6 +172,16 @@ entry:
   ret <8 x i16> %1
 }
 
+define <8 x half> @v8bf16_to_v8f16(float, <8 x bfloat> %a) nounwind {
+; CHECK-LABEL: v8bf16_to_v8f16:
+; CHECK:       // %bb.0: // %entry
+; CHECK-NEXT:    mov v0.16b, v1.16b
+; CHECK-NEXT:    ret
+entry:
+  %1 = bitcast <8 x bfloat> %a to <8 x half>
+  ret <8 x half> %1
+}
+
 define <4 x i32> @v8bf16_to_v4i32(float, <8 x bfloat> %a) nounwind {
 ; CHECK-LABEL: v8bf16_to_v4i32:
 ; CHECK:       // %bb.0: // %entry
@@ -202,6 +232,16 @@ entry:
   ret <8 x bfloat> %1
 }
 
+define <8 x bfloat> @v8f16_to_v8bf16(float, <8 x half> %a) nounwind {
+; CHECK-LABEL: v8f16_to_v8bf16:
+; CHECK:       // %bb.0: // %entry
+; CHECK-NEXT:    mov v0.16b, v1.16b
+; CHECK-NEXT:    ret
+entry:
+  %1 = bitcast <8 x half> %a to <8 x bfloat>
+  ret <8 x bfloat> %1
+}
+
 define <8 x bfloat> @v4i32_to_v8bf16(float, <4 x i32> %a) nounwind {
 ; CHECK-LABEL: v4i32_to_v8bf16:
 ; CHECK:       // %bb.0: // %entry

[nasherm]

Thanks, LGTM