From 60f57b36587c99175cb380406e2502a592a0c400 Mon Sep 17 00:00:00 2001
From: David Green <david.green@arm.com>
Date: Tue, 29 Mar 2022 10:12:44 +0100
Subject: [PATCH] [AArch64] Ensure fixed point fptoi_sat has correct saturation
 width

D113200 introduced an error where it was converting FP_TO_SI_SAT with
multiply to a fixed point floating point convert. The saturation
bitwidth needs to be equal to the floating point width, or else the
routine would truncate the result as opposed to saturating it.

Fixes #54601
---
 .../Target/AArch64/AArch64ISelLowering.cpp    |  2 +-
 llvm/test/CodeGen/AArch64/fcvt_combine.ll     | 31 ++++++++++++++++---
 2 files changed, 28 insertions(+), 5 deletions(-)

diff --git a/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp b/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
index 30b9b6096620..38079a191f72 100644
--- a/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
+++ b/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
@@ -13973,7 +13973,7 @@ static SDValue performFpToIntCombine(SDNode *N, SelectionDAG &DAG,
   if (N->getOpcode() == ISD::FP_TO_SINT_SAT ||
       N->getOpcode() == ISD::FP_TO_UINT_SAT) {
     EVT SatVT = cast<VTSDNode>(N->getOperand(1))->getVT();
-    if (SatVT.getScalarSizeInBits() != IntBits)
+    if (SatVT.getScalarSizeInBits() != IntBits || IntBits != FloatBits)
       return SDValue();
   }
 
diff --git a/llvm/test/CodeGen/AArch64/fcvt_combine.ll b/llvm/test/CodeGen/AArch64/fcvt_combine.ll
index 24713c444024..67af07e05ab0 100644
--- a/llvm/test/CodeGen/AArch64/fcvt_combine.ll
+++ b/llvm/test/CodeGen/AArch64/fcvt_combine.ll
@@ -285,6 +285,7 @@ declare <2 x i32> @llvm.fptosi.sat.v2i32.v2f64(<2 x double>)
 declare <2 x i64> @llvm.fptosi.sat.v2i64.v2f64(<2 x double>)
 declare <4 x i32> @llvm.fptosi.sat.v4i32.v4f32(<4 x float>)
 declare <2 x i16> @llvm.fptosi.sat.v2i16.v2f32(<2 x float>)
+declare <4 x i16> @llvm.fptosi.sat.v4i16.v4f32(<4 x float>)
 declare <2 x i64> @llvm.fptosi.sat.v2i64.v2f32(<2 x float>)
 declare <3 x i32> @llvm.fptosi.sat.v3i32.v3f32(<3 x float>)
 declare <8 x i16> @llvm.fptosi.sat.v8i16.v8f16(<8 x half>)
@@ -326,8 +327,14 @@ define <2 x i64> @test3_sat(<2 x double> %d) {
 define <2 x i32> @test4_sat(<2 x double> %d) {
 ; CHECK-LABEL: test4_sat:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    fcvtzs v0.2d, v0.2d, #4
-; CHECK-NEXT:    xtn v0.2s, v0.2d
+; CHECK-NEXT:    fmov v1.2d, #16.00000000
+; CHECK-NEXT:    fmul v0.2d, v0.2d, v1.2d
+; CHECK-NEXT:    mov d1, v0.d[1]
+; CHECK-NEXT:    fcvtzs w8, d0
+; CHECK-NEXT:    fmov s0, w8
+; CHECK-NEXT:    fcvtzs w8, d1
+; CHECK-NEXT:    mov v0.s[1], w8
+; CHECK-NEXT:    // kill: def $d0 killed $d0 killed $q0
 ; CHECK-NEXT:    ret
   %mul.i = fmul <2 x double> %d, <double 16.000000e+00, double 16.000000e+00>
   %vcvt.i = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f64(<2 x double> %mul.i)
@@ -338,13 +345,29 @@ define <2 x i32> @test4_sat(<2 x double> %d) {
 define <2 x i16> @test5_sat(<2 x float> %f) {
 ; CHECK-LABEL: test5_sat:
 ; CHECK:       // %bb.0:
+; CHECK-NEXT:    movi v1.2s, #127, msl #8
 ; CHECK-NEXT:    fcvtzs v0.2s, v0.2s, #4
+; CHECK-NEXT:    smin v0.2s, v0.2s, v1.2s
+; CHECK-NEXT:    mvni v1.2s, #127, msl #8
+; CHECK-NEXT:    smax v0.2s, v0.2s, v1.2s
 ; CHECK-NEXT:    ret
   %mul.i = fmul <2 x float> %f, <float 16.000000e+00, float 16.000000e+00>
   %vcvt.i = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f32(<2 x float> %mul.i)
   ret <2 x i16> %vcvt.i
 }
 
+; Truncate float to i16
+define <4 x i16> @test5l_sat(<4 x float> %f) {
+; CHECK-LABEL: test5l_sat:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    fcvtzs v0.4s, v0.4s, #4
+; CHECK-NEXT:    sqxtn v0.4h, v0.4s
+; CHECK-NEXT:    ret
+  %mul.i = fmul <4 x float> %f, <float 16.000000e+00, float 16.000000e+00, float 16.000000e+00, float 16.000000e+00>
+  %vcvt.i = call <4 x i16> @llvm.fptosi.sat.v4i16.v4f32(<4 x float> %mul.i)
+  ret <4 x i16> %vcvt.i
+}
+
 ; Don't convert float to i64
 define <2 x i64> @test6_sat(<2 x float> %f) {
 ; CHECK-LABEL: test6_sat:
@@ -389,8 +412,8 @@ define <2 x i32> @test8_sat(<2 x float> %f) {
 define <2 x i32> @test9_sat(<2 x float> %f) {
 ; CHECK-LABEL: test9_sat:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    adrp x8, .LCPI26_0
-; CHECK-NEXT:    ldr d1, [x8, :lo12:.LCPI26_0]
+; CHECK-NEXT:    adrp x8, .LCPI27_0
+; CHECK-NEXT:    ldr d1, [x8, :lo12:.LCPI27_0]
 ; CHECK-NEXT:    fmul v0.2s, v0.2s, v1.2s
 ; CHECK-NEXT:    fcvtzu v0.2s, v0.2s
 ; CHECK-NEXT:    ret
-- 
2.35.1.1021.g381101b075-goog