[PATCH] D57779: [SLP] Add support for throttling.

[Alexey Bataev via Phabricator via llvm-commits]

ABataev added inline comments.


================
Comment at: test/Transforms/SLPVectorizer/X86/ctlz.ll:76
 ; CHECK-LABEL: @ctlz_4i32(
-; CHECK-NEXT:    [[LD0:%.*]] = load i32, i32* getelementptr inbounds ([8 x i32], [8 x i32]* @src32, i32 0, i64 0), align 4
-; CHECK-NEXT:    [[LD1:%.*]] = load i32, i32* getelementptr inbounds ([8 x i32], [8 x i32]* @src32, i32 0, i64 1), align 4
-; CHECK-NEXT:    [[LD2:%.*]] = load i32, i32* getelementptr inbounds ([8 x i32], [8 x i32]* @src32, i32 0, i64 2), align 4
-; CHECK-NEXT:    [[LD3:%.*]] = load i32, i32* getelementptr inbounds ([8 x i32], [8 x i32]* @src32, i32 0, i64 3), align 4
-; CHECK-NEXT:    [[CTLZ0:%.*]] = call i32 @llvm.ctlz.i32(i32 [[LD0]], i1 false)
-; CHECK-NEXT:    [[CTLZ1:%.*]] = call i32 @llvm.ctlz.i32(i32 [[LD1]], i1 false)
-; CHECK-NEXT:    [[CTLZ2:%.*]] = call i32 @llvm.ctlz.i32(i32 [[LD2]], i1 false)
-; CHECK-NEXT:    [[CTLZ3:%.*]] = call i32 @llvm.ctlz.i32(i32 [[LD3]], i1 false)
-; CHECK-NEXT:    store i32 [[CTLZ0]], i32* getelementptr inbounds ([8 x i32], [8 x i32]* @dst32, i32 0, i64 0), align 4
-; CHECK-NEXT:    store i32 [[CTLZ1]], i32* getelementptr inbounds ([8 x i32], [8 x i32]* @dst32, i32 0, i64 1), align 4
-; CHECK-NEXT:    store i32 [[CTLZ2]], i32* getelementptr inbounds ([8 x i32], [8 x i32]* @dst32, i32 0, i64 2), align 4
-; CHECK-NEXT:    store i32 [[CTLZ3]], i32* getelementptr inbounds ([8 x i32], [8 x i32]* @dst32, i32 0, i64 3), align 4
+; CHECK-NEXT:    [[TMP1:%.*]] = load <4 x i32>, <4 x i32>* bitcast ([8 x i32]* @src32 to <4 x i32>*), align 4
+; CHECK-NEXT:    [[TMP2:%.*]] = call <4 x i32> @llvm.ctlz.v4i32(<4 x i32> [[TMP1]], i1 false)
----------------
There is definitely something wrong with this patch. For example, in this case. There is no partial tree. Why is it vectorized?


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D57779/new/

https://reviews.llvm.org/D57779