================
@@ -0,0 +1,63 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
+; RUN: opt -codegenprepare -mtriple=x86_64-unknown-linux-gnu -mattr=+avx2 -S <
%s | FileCheck %s
+
+define i8 @hoist_bitcast_i256_to_v32i8(ptr %a0) {
+; CHECK-LABEL: @hoist_bitcast_i256_to_v32i8(
+; CHECK-NEXT: entry:
+; CHECK-NEXT: [[SRC256:%.*]] = load i256, ptr [[A0:%.*]], align 1
+; CHECK-NEXT: [[SRC256_BITCAST:%.*]] = bitcast i256 [[SRC256]] to <32 x i8>
+; CHECK-NEXT: [[ISZERO:%.*]] = icmp eq i256 [[SRC256]], 0
+; CHECK-NEXT: br i1 [[ISZERO]], label [[EXIT:%.*]], label [[REDUCTION:%.*]]
+; CHECK: reduction:
+; CHECK-NEXT: [[RED:%.*]] = call i8 @llvm.vector.reduce.umax.v32i8(<32 x
i8> [[SRC256_BITCAST]])
+; CHECK-NEXT: br label [[EXIT]]
+; CHECK: exit:
+; CHECK-NEXT: [[RESULT:%.*]] = phi i8 [ 0, [[ENTRY:%.*]] ], [ [[RED]],
[[REDUCTION]] ]
+; CHECK-NEXT: ret i8 [[RESULT]]
+;
+entry:
+ %src256 = load i256, ptr %a0, align 1
+ %iszero = icmp eq i256 %src256, 0
+ br i1 %iszero, label %exit, label %reduction
+
+reduction:
+ %src256.bitcast = bitcast i256 %src256 to <32 x i8>
+ %red = call i8 @llvm.vector.reduce.umax.v32i8(<32 x i8> %src256.bitcast)
+ br label %exit
+
+exit:
+ %result = phi i8 [ 0, %entry ], [ %red, %reduction ]
+ ret i8 %result
+}
+
+define i8 @test_legal_i64(ptr %a0) {
+; CHECK-LABEL: @test_legal_i64(
+; CHECK-NEXT: entry:
+; CHECK-NEXT: [[SRC64:%.*]] = load i64, ptr [[A0:%.*]], align 8
+; CHECK-NEXT: [[ISZERO:%.*]] = icmp eq i64 [[SRC64]], 0
+; CHECK-NEXT: br i1 [[ISZERO]], label [[EXIT:%.*]], label [[REDUCTION:%.*]]
+; CHECK: reduction:
+; CHECK-NEXT: [[SRC64_BITCAST:%.*]] = bitcast i64 [[SRC64]] to <8 x i8>
+; CHECK-NEXT: [[RED:%.*]] = call i8 @llvm.vector.reduce.umax.v8i8(<8 x i8>
[[SRC64_BITCAST]])
+; CHECK-NEXT: br label [[EXIT]]
+; CHECK: exit:
+; CHECK-NEXT: [[RESULT:%.*]] = phi i8 [ 0, [[ENTRY:%.*]] ], [ [[RED]],
[[REDUCTION]] ]
+; CHECK-NEXT: ret i8 [[RESULT]]
+;
+entry:
+ %src64 = load i64, ptr %a0, align 8
+ %iszero = icmp eq i64 %src64, 0
+ br i1 %iszero, label %exit, label %reduction
+
+reduction:
+ %src64.bitcast = bitcast i64 %src64 to <8 x i8>
+ %red = call i8 @llvm.vector.reduce.umax.v8i8(<8 x i8> %src64.bitcast)
+ br label %exit
+
+exit:
+ %result = phi i8 [ 0, %entry ], [ %red, %reduction ]
+ ret i8 %result
+}
+
+declare i8 @llvm.vector.reduce.umax.v8i8(<8 x i8>)
+declare i8 @llvm.vector.reduce.umax.v32i8(<32 x i8>)
----------------
RKSimon wrote:
you shouldn't need to declare these
https://github.com/llvm/llvm-project/pull/207568
_______________________________________________
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
