[AArch64] Add lowering pattern for scalar fp16 facge and facgt

Summary: The fp16 scalar version of facge and facgt requires a custom patter matching, as the result type is not the same width of the operands.

Reviewers: olista01, javed.absar, pbarrio

Reviewed By: javed.absar

Subscribers: kristof.beyls, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D60212

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@358083 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/AArch64/AArch64InstrInfo.td b/lib/Target/AArch64/AArch64InstrInfo.td
index c54970331abb66b91e36b566d6a235a24ec4e768..06d25bb3e37561396adea2e7cbc50c4cd1b4405a 100644 (file)
--- a/lib/Target/AArch64/AArch64InstrInfo.td
+++ b/lib/Target/AArch64/AArch64InstrInfo.td
@@ -5355,6 +5355,16 @@ def : Pat<(i64 (int_aarch64_neon_vcvtfp2fxu (f16 FPR16:$Rn), vecshiftR64:$imm)),
             (i64 (IMPLICIT_DEF)),
             (FCVTZUh FPR16:$Rn, vecshiftR64:$imm),
             hsub))>;
+def : Pat<(i32 (int_aarch64_neon_facge (f16 FPR16:$Rn), (f16 FPR16:$Rm))),
+          (i32 (INSERT_SUBREG
+            (i32 (IMPLICIT_DEF)),
+            (FACGE16 FPR16:$Rn, FPR16:$Rm),
+            hsub))>;
+def : Pat<(i32 (int_aarch64_neon_facgt (f16 FPR16:$Rn), (f16 FPR16:$Rm))),
+          (i32 (INSERT_SUBREG
+            (i32 (IMPLICIT_DEF)),
+            (FACGT16 FPR16:$Rn, FPR16:$Rm),
+            hsub))>;
 
 defm SHL      : SIMDScalarLShiftD<   0, 0b01010, "shl", AArch64vshl>;
 defm SLI      : SIMDScalarLShiftDTied<1, 0b01010, "sli">;

diff --git a/test/CodeGen/AArch64/fp16_intrinsic_scalar_2op.ll b/test/CodeGen/AArch64/fp16_intrinsic_scalar_2op.ll
index 13a18b10e9f92a43b5e18ed89e85de71abbdb182..19365a6f2f768be27abf9eba19ab9e248ae93366 100644 (file)
--- a/test/CodeGen/AArch64/fp16_intrinsic_scalar_2op.ll
+++ b/test/CodeGen/AArch64/fp16_intrinsic_scalar_2op.ll
@@ -7,6 +7,8 @@ declare half @llvm.aarch64.neon.frsqrts.f16(half, half)
 declare half @llvm.aarch64.neon.frecps.f16(half, half)
 declare half @llvm.aarch64.neon.fmulx.f16(half, half)
 declare half @llvm.fabs.f16(half)
+declare i32 @llvm.aarch64.neon.facge.i32.f16(half, half)
+declare i32 @llvm.aarch64.neon.facgt.i32.f16(half, half)
 
 define dso_local half @t_vabdh_f16(half %a, half %b) {
 ; CHECK-LABEL: t_vabdh_f16:
@@ -318,3 +320,25 @@ entry:
   %vcvth_n_u32_f16 = tail call i32 @llvm.aarch64.neon.vcvtfp2fxu.i32.f16(half %a, i32 16)
   ret i32 %vcvth_n_u32_f16
 }
+
+define dso_local i16 @vcageh_f16_test(half %a, half %b) {
+; CHECK-LABEL: vcageh_f16_test:
+; CHECK:        facge   h0, h0, h1
+; CHECK-NEXT:   fmov    w0, s0
+; CHECK-NEXT:   ret
+entry:
+  %facg = tail call i32 @llvm.aarch64.neon.facge.i32.f16(half %a, half %b)
+  %0 = trunc i32 %facg to i16
+  ret i16 %0
+}
+
+define dso_local i16 @vcagth_f16_test(half %a, half %b) {
+; CHECK-LABEL: vcagth_f16_test:
+; CHECK:        facgt   h0, h0, h1
+; CHECK-NEXT:   fmov    w0, s0
+; CHECK-NEXT:   ret
+entry:
+  %facg = tail call i32 @llvm.aarch64.neon.facgt.i32.f16(half %a, half %b)
+  %0 = trunc i32 %facg to i16
+  ret i16 %0
+}