From 8843b7105c41353de8945d0c02fe735a8a57c232 Mon Sep 17 00:00:00 2001
From: Ahmed Bougacha <ahmed.bougacha@gmail.com>
Date: Fri, 6 Nov 2015 23:16:53 +0000
Subject: [PATCH] [AArch64][FastISel] Don't even try to select vector icmps.

We used to try to constant-fold them to i32 immediates.
Given that fast-isel doesn't otherwise support vNi1, when selecting
the result users, we'd fallback to SDAG anyway.
However, if the users were in another block, we'd insert broken
cross-class copies (GPR32 to FPR64).

Give up, let SDAG agree with itself on a vNi1 legalization strategy.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@252364 91177308-0d34-0410-b5e6-96231b3b80d8
---
 lib/Target/AArch64/AArch64FastISel.cpp    |   4 +
 test/CodeGen/AArch64/fast-isel-cmp-vec.ll | 100 ++++++++++++++++++++++
 2 files changed, 104 insertions(+)
 create mode 100644 test/CodeGen/AArch64/fast-isel-cmp-vec.ll

diff --git a/lib/Target/AArch64/AArch64FastISel.cpp b/lib/Target/AArch64/AArch64FastISel.cpp
index 2f50480efbe..284f5263f90 100644
--- a/lib/Target/AArch64/AArch64FastISel.cpp
+++ b/lib/Target/AArch64/AArch64FastISel.cpp
@@ -2451,6 +2451,10 @@ bool AArch64FastISel::selectIndirectBr(const Instruction *I) {
 bool AArch64FastISel::selectCmp(const Instruction *I) {
   const CmpInst *CI = cast<CmpInst>(I);
 
+  // Vectors of i1 are weird: bail out.
+  if (CI->getType()->isVectorTy())
+    return false;
+
   // Try to optimize or fold the cmp.
   CmpInst::Predicate Predicate = optimizeCmpPredicate(CI);
   unsigned ResultReg = 0;
diff --git a/test/CodeGen/AArch64/fast-isel-cmp-vec.ll b/test/CodeGen/AArch64/fast-isel-cmp-vec.ll
new file mode 100644
index 00000000000..2855419a1ca
--- /dev/null
+++ b/test/CodeGen/AArch64/fast-isel-cmp-vec.ll
@@ -0,0 +1,100 @@
+; RUN: llc -mtriple=aarch64-apple-darwin -fast-isel -verify-machineinstrs \
+; RUN:   -aarch64-atomic-cfg-tidy=0 -disable-cgp -disable-branch-fold \
+; RUN:   < %s | FileCheck %s
+
+;
+; Verify that we don't mess up vector comparisons in fast-isel.
+;
+
+define <2 x i32> @icmp_v2i32(<2 x i32> %a) {
+; CHECK-LABEL: icmp_v2i32:
+; CHECK:      ; BB#0:
+; CHECK-NEXT:  cmeq.2s [[CMP:v[0-9]+]], v0, #0
+; CHECK-NEXT: ; BB#1:
+; CHECK-NEXT:  movi.2s [[MASK:v[0-9]+]], #0x1
+; CHECK-NEXT:  and.8b v0, [[CMP]], [[MASK]]
+; CHECK-NEXT:  ret
+  %c = icmp eq <2 x i32> %a, zeroinitializer
+  br label %bb2
+bb2:
+  %z = zext <2 x i1> %c to <2 x i32>
+  ret <2 x i32> %z
+}
+
+define <2 x i32> @icmp_constfold_v2i32(<2 x i32> %a) {
+; CHECK-LABEL: icmp_constfold_v2i32:
+; CHECK:      ; BB#0:
+; CHECK-NEXT:  movi d[[CMP:[0-9]+]], #0xffffffffffffffff
+; CHECK-NEXT: ; BB#1:
+; CHECK-NEXT:  movi.2s [[MASK:v[0-9]+]], #0x1
+; CHECK-NEXT:  and.8b v0, v[[CMP]], [[MASK]]
+; CHECK-NEXT:  ret
+  %1 = icmp eq <2 x i32> %a, %a
+  br label %bb2
+bb2:
+  %2 = zext <2 x i1> %1 to <2 x i32>
+  ret <2 x i32> %2
+}
+
+define <4 x i32> @icmp_v4i32(<4 x i32> %a) {
+; CHECK-LABEL: icmp_v4i32:
+; CHECK:      ; BB#0:
+; CHECK-NEXT:  cmeq.4s [[CMP:v[0-9]+]], v0, #0
+; CHECK-NEXT:  xtn.4h [[CMPV4I16:v[0-9]+]], [[CMP]]
+; CHECK-NEXT: ; BB#1:
+; CHECK-NEXT:  movi.4h [[MASK:v[0-9]+]], #0x1
+; CHECK-NEXT:  and.8b [[ZEXT:v[0-9]+]], [[CMPV4I16]], [[MASK]]
+; CHECK-NEXT:  ushll.4s v0, [[ZEXT]], #0
+; CHECK-NEXT:  ret
+  %c = icmp eq <4 x i32> %a, zeroinitializer
+  br label %bb2
+bb2:
+  %z = zext <4 x i1> %c to <4 x i32>
+  ret <4 x i32> %z
+}
+
+define <4 x i32> @icmp_constfold_v4i32(<4 x i32> %a) {
+; CHECK-LABEL: icmp_constfold_v4i32:
+; CHECK:      ; BB#0:
+; CHECK-NEXT:  movi d[[CMP:[0-9]+]], #0xffffffffffffffff
+; CHECK-NEXT: ; BB#1:
+; CHECK-NEXT:  movi.4h [[MASK:v[0-9]+]], #0x1
+; CHECK-NEXT:  and.8b [[ZEXT:v[0-9]+]], v[[CMP]], [[MASK]]
+; CHECK-NEXT:  ushll.4s v0, [[ZEXT]], #0
+; CHECK-NEXT:  ret
+  %1 = icmp eq <4 x i32> %a, %a
+  br label %bb2
+bb2:
+  %2 = zext <4 x i1> %1 to <4 x i32>
+  ret <4 x i32> %2
+}
+
+define <16 x i8> @icmp_v16i8(<16 x i8> %a) {
+; CHECK-LABEL: icmp_v16i8:
+; CHECK:      ; BB#0:
+; CHECK-NEXT:  cmeq.16b [[CMP:v[0-9]+]], v0, #0
+; CHECK-NEXT: ; BB#1:
+; CHECK-NEXT:  movi.16b [[MASK:v[0-9]+]], #0x1
+; CHECK-NEXT:  and.16b v0, [[CMP]], [[MASK]]
+; CHECK-NEXT:  ret
+  %c = icmp eq <16 x i8> %a, zeroinitializer
+  br label %bb2
+bb2:
+  %z = zext <16 x i1> %c to <16 x i8>
+  ret <16 x i8> %z
+}
+
+define <16 x i8> @icmp_constfold_v16i8(<16 x i8> %a) {
+; CHECK-LABEL: icmp_constfold_v16i8:
+; CHECK:      ; BB#0:
+; CHECK-NEXT:  movi.2d [[CMP:v[0-9]+]], #0xffffffffffffffff
+; CHECK-NEXT: ; BB#1:
+; CHECK-NEXT:  movi.16b [[MASK:v[0-9]+]], #0x1
+; CHECK-NEXT:  and.16b v0, [[CMP]], [[MASK]]
+; CHECK-NEXT:  ret
+  %1 = icmp eq <16 x i8> %a, %a
+  br label %bb2
+bb2:
+  %2 = zext <16 x i1> %1 to <16 x i8>
+  ret <16 x i8> %2
+}