[llvm] [X86][SimplifyCFG] Support hoisting load/store with conditional faulting (PR #96878)

[Yingwei Zheng via llvm-commits]

================
@@ -0,0 +1,460 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
+; RUN: opt < %s -mtriple=x86_64 -mattr=+cf -passes=simplifycfg -simplifycfg-require-and-preserve-domtree=1 -S -simplifycfg-hoist-loads-stores-with-cond-faulting=true | FileCheck %s
+
+;; The redundant bitcast/insertelement will be opimized out in instcombine pass.
+define void @basic(i32 %a, ptr %b, ptr %p, ptr %q) {
+; CHECK-LABEL: @basic(
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[COND:%.*]] = icmp eq i32 [[A:%.*]], 0
+; CHECK-NEXT:    [[TMP0:%.*]] = bitcast i1 [[COND]] to <1 x i1>
+; CHECK-NEXT:    [[TMP1:%.*]] = call <1 x i32> @llvm.masked.load.v1i32.p0(ptr [[B:%.*]], i32 4, <1 x i1> [[TMP0]], <1 x i32> poison), !dbg [[DBG8:![0-9]+]]
+; CHECK-NEXT:    [[TMP2:%.*]] = bitcast <1 x i32> [[TMP1]] to i32
+; CHECK-NEXT:    [[TMP3:%.*]] = bitcast i32 [[TMP2]] to <1 x i32>
+; CHECK-NEXT:    call void @llvm.masked.store.v1i32.p0(<1 x i32> [[TMP3]], ptr [[P:%.*]], i32 4, <1 x i1> [[TMP0]])
+; CHECK-NEXT:    [[TMP4:%.*]] = xor i1 [[COND]], true
+; CHECK-NEXT:    [[TMP5:%.*]] = bitcast i1 [[TMP4]] to <1 x i1>
+; CHECK-NEXT:    call void @llvm.masked.store.v1i64.p0(<1 x i64> <i64 1>, ptr [[P]], i32 8, <1 x i1> [[TMP5]]), !dbg [[DBG12:![0-9]+]]
+; CHECK-NEXT:    call void @llvm.masked.store.v1i16.p0(<1 x i16> <i16 2>, ptr [[Q:%.*]], i32 8, <1 x i1> [[TMP5]]), !dbg [[DBG12]]
+; CHECK-NEXT:    ret void
+;
+entry:
+  %cond = icmp eq i32 %a, 0
+  br i1 %cond, label %if.true, label %if.false
+
+if.false:
+  store i64 1, ptr %p, align 8, !dbg !8
+  store i16 2, ptr %q, align 8, !dbg !8
+  br label %if.end
+
+if.true:
+  %0 = load i32, ptr %b, align 4,  !dbg !9
+  store i32 %0, ptr %p, align 4
+  br label %if.end
+
+if.end:
+  ret void
+}
+
----------------
dtcxzyw wrote:

It doesn't work for this pattern:

A
 | \\
 |   \\
 |  B (load/store)
 |  /
 | /
 C

```
define i32 @cload1(ptr noundef %x) {
entry:
  %tobool.not = icmp eq ptr %x, null
  br i1 %tobool.not, label %cond.end, label %cond.true

cond.true:                                        ; preds = %entry
  %0 = load i32, ptr %x, align 4
  br label %cond.end

cond.end:                                         ; preds = %entry, %cond.true
  %cond = phi i32 [ %0, %cond.true ], [ 0, %entry ]
  ret i32 %cond
}

```

https://github.com/llvm/llvm-project/pull/96878