From 878141adab9e1e23e458e88f66f336c32c60148a Mon Sep 17 00:00:00 2001
From: Paul Walker <paul.walker@arm.com>
Date: Fri, 27 Sep 2024 15:32:57 +0100
Subject: [PATCH] [LLVM][InstCombine][SVE] fcvtnt(a,all_active,b) !=
 fcvtnt(undef,all_active,b)

The "narrowing top" convert instructions leave the bottom half of
active elements untouched and thus the first paramater of their
associated intrinsic remains live even when there are no inactive
lanes.
---
 llvm/lib/Target/AArch64/AArch64TargetTransformInfo.cpp   | 9 +++++----
 .../AArch64/sve-intrinsic-comb-all-active-lanes-cvt.ll   | 8 ++++----
 2 files changed, 9 insertions(+), 8 deletions(-)

diff --git a/llvm/lib/Target/AArch64/AArch64TargetTransformInfo.cpp b/llvm/lib/Target/AArch64/AArch64TargetTransformInfo.cpp
index da0798ebf7957..4be236a7827f9 100644
--- a/llvm/lib/Target/AArch64/AArch64TargetTransformInfo.cpp
+++ b/llvm/lib/Target/AArch64/AArch64TargetTransformInfo.cpp
@@ -2156,11 +2156,7 @@ AArch64TTIImpl::instCombineIntrinsic(InstCombiner &IC,
   case Intrinsic::aarch64_sve_fcvt_f64f32:
   case Intrinsic::aarch64_sve_fcvtlt_f32f16:
   case Intrinsic::aarch64_sve_fcvtlt_f64f32:
-  case Intrinsic::aarch64_sve_fcvtnt_bf16f32:
-  case Intrinsic::aarch64_sve_fcvtnt_f16f32:
-  case Intrinsic::aarch64_sve_fcvtnt_f32f64:
   case Intrinsic::aarch64_sve_fcvtx_f32f64:
-  case Intrinsic::aarch64_sve_fcvtxnt_f32f64:
   case Intrinsic::aarch64_sve_fcvtzs:
   case Intrinsic::aarch64_sve_fcvtzs_i32f16:
   case Intrinsic::aarch64_sve_fcvtzs_i32f64:
@@ -2182,6 +2178,11 @@ AArch64TTIImpl::instCombineIntrinsic(InstCombiner &IC,
   case Intrinsic::aarch64_sve_ucvtf_f32i64:
   case Intrinsic::aarch64_sve_ucvtf_f64i32:
     return instCombineSVEAllOrNoActiveUnary(IC, II);
+  case Intrinsic::aarch64_sve_fcvtnt_bf16f32:
+  case Intrinsic::aarch64_sve_fcvtnt_f16f32:
+  case Intrinsic::aarch64_sve_fcvtnt_f32f64:
+  case Intrinsic::aarch64_sve_fcvtxnt_f32f64:
+    return instCombineSVENoActiveReplace(IC, II, true);
   case Intrinsic::aarch64_sve_st1_scatter:
   case Intrinsic::aarch64_sve_st1_scatter_scalar_offset:
   case Intrinsic::aarch64_sve_st1_scatter_sxtw:
diff --git a/llvm/test/Transforms/InstCombine/AArch64/sve-intrinsic-comb-all-active-lanes-cvt.ll b/llvm/test/Transforms/InstCombine/AArch64/sve-intrinsic-comb-all-active-lanes-cvt.ll
index 374a985191768..04550156be30b 100644
--- a/llvm/test/Transforms/InstCombine/AArch64/sve-intrinsic-comb-all-active-lanes-cvt.ll
+++ b/llvm/test/Transforms/InstCombine/AArch64/sve-intrinsic-comb-all-active-lanes-cvt.ll
@@ -138,7 +138,7 @@ define <vscale x 8 x bfloat> @test_fcvtnt_bf16_f32(<vscale x 8 x bfloat> %a, <vs
 ; CHECK-LABEL: define <vscale x 8 x bfloat> @test_fcvtnt_bf16_f32(
 ; CHECK-SAME: <vscale x 8 x bfloat> [[A:%.*]], <vscale x 4 x float> [[B:%.*]]) {
 ; CHECK-NEXT:    [[PG:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.ptrue.nxv8i1(i32 31)
-; CHECK-NEXT:    [[OUT:%.*]] = call <vscale x 8 x bfloat> @llvm.aarch64.sve.fcvtnt.bf16f32(<vscale x 8 x bfloat> undef, <vscale x 8 x i1> [[PG]], <vscale x 4 x float> [[B]])
+; CHECK-NEXT:    [[OUT:%.*]] = call <vscale x 8 x bfloat> @llvm.aarch64.sve.fcvtnt.bf16f32(<vscale x 8 x bfloat> [[A]], <vscale x 8 x i1> [[PG]], <vscale x 4 x float> [[B]])
 ; CHECK-NEXT:    ret <vscale x 8 x bfloat> [[OUT]]
 ;
   %pg = tail call <vscale x 8 x i1> @llvm.aarch64.sve.ptrue.nxv8i1(i32 31)
@@ -150,7 +150,7 @@ define <vscale x 8 x half> @test_fcvtnt_f16_f32(<vscale x 8 x half> %a, <vscale
 ; CHECK-LABEL: define <vscale x 8 x half> @test_fcvtnt_f16_f32(
 ; CHECK-SAME: <vscale x 8 x half> [[A:%.*]], <vscale x 4 x float> [[B:%.*]]) {
 ; CHECK-NEXT:    [[PG:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.ptrue.nxv4i1(i32 31)
-; CHECK-NEXT:    [[OUT:%.*]] = call <vscale x 8 x half> @llvm.aarch64.sve.fcvtnt.f16f32(<vscale x 8 x half> undef, <vscale x 4 x i1> [[PG]], <vscale x 4 x float> [[B]])
+; CHECK-NEXT:    [[OUT:%.*]] = call <vscale x 8 x half> @llvm.aarch64.sve.fcvtnt.f16f32(<vscale x 8 x half> [[A]], <vscale x 4 x i1> [[PG]], <vscale x 4 x float> [[B]])
 ; CHECK-NEXT:    ret <vscale x 8 x half> [[OUT]]
 ;
   %pg = tail call <vscale x 4 x i1> @llvm.aarch64.sve.ptrue.nxv4i1(i32 31)
@@ -162,7 +162,7 @@ define <vscale x 4 x float> @test_fcvtnt_f32_f64(<vscale x 4 x float> %a, <vscal
 ; CHECK-LABEL: define <vscale x 4 x float> @test_fcvtnt_f32_f64(
 ; CHECK-SAME: <vscale x 4 x float> [[A:%.*]], <vscale x 2 x double> [[B:%.*]]) {
 ; CHECK-NEXT:    [[PG:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.ptrue.nxv2i1(i32 31)
-; CHECK-NEXT:    [[OUT:%.*]] = call <vscale x 4 x float> @llvm.aarch64.sve.fcvtnt.f32f64(<vscale x 4 x float> undef, <vscale x 2 x i1> [[PG]], <vscale x 2 x double> [[B]])
+; CHECK-NEXT:    [[OUT:%.*]] = call <vscale x 4 x float> @llvm.aarch64.sve.fcvtnt.f32f64(<vscale x 4 x float> [[A]], <vscale x 2 x i1> [[PG]], <vscale x 2 x double> [[B]])
 ; CHECK-NEXT:    ret <vscale x 4 x float> [[OUT]]
 ;
   %pg = tail call <vscale x 2 x i1> @llvm.aarch64.sve.ptrue.nxv2i1(i32 31)
@@ -186,7 +186,7 @@ define <vscale x 4 x float> @test_fcvtxnt_f32_f64(<vscale x 4 x float> %a, <vsca
 ; CHECK-LABEL: define <vscale x 4 x float> @test_fcvtxnt_f32_f64(
 ; CHECK-SAME: <vscale x 4 x float> [[A:%.*]], <vscale x 2 x double> [[B:%.*]]) {
 ; CHECK-NEXT:    [[PG:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.ptrue.nxv2i1(i32 31)
-; CHECK-NEXT:    [[OUT:%.*]] = call <vscale x 4 x float> @llvm.aarch64.sve.fcvtxnt.f32f64(<vscale x 4 x float> undef, <vscale x 2 x i1> [[PG]], <vscale x 2 x double> [[B]])
+; CHECK-NEXT:    [[OUT:%.*]] = call <vscale x 4 x float> @llvm.aarch64.sve.fcvtxnt.f32f64(<vscale x 4 x float> [[A]], <vscale x 2 x i1> [[PG]], <vscale x 2 x double> [[B]])
 ; CHECK-NEXT:    ret <vscale x 4 x float> [[OUT]]
 ;
   %pg = tail call <vscale x 2 x i1> @llvm.aarch64.sve.ptrue.nxv2i1(i32 31)