[llvm] [LV] Handle partial sub-reductions with sub in middle block. (PR #178919)

[Sander de Smalen via llvm-commits]

================
@@ -0,0 +1,76 @@
+; REQUIRES: asserts
+; RUN: opt -passes=loop-vectorize                                          \
+; RUN:     -enable-epilogue-vectorization=false -debug-only=loop-vectorize \
+; RUN:     -disable-output < %s 2>&1 | FileCheck %s
+
+; CHECK: LV: Checking a loop in 'sub_reduction'
+; CHECK: Cost of 1 for VF 16: EXPRESSION vp<{{.*}}> = ir<%acc> + partial.reduce.add (mul (ir<%load1> sext to i32), (ir<%load2> sext to i32))
+
+; CHECK: LV: Checking a loop in 'add_sub_chained_reduction'
+; CHECK: Cost of 1 for VF 16: EXPRESSION vp<{{.*}}> = ir<%acc> + partial.reduce.add (mul (ir<%load1> sext to i32), (ir<%load2> sext to i32))
+; CHECK: Cost of 9 for VF 16: EXPRESSION vp<{{.*}}> = vp<%9> + partial.reduce.add (sub (0, mul (ir<%load2> sext to i32), (ir<%load3> sext to i32)))
+
+target triple = "aarch64"
+
+; Test the cost of a SUB reduction, where the SUB is implemented outside the loop
+; and therefore not part of the partial reduction.
+define i32 @sub_reduction(ptr %arr1, ptr %arr2, i32 %init, i32 %n) #0 {
----------------
sdesmalen-arm wrote:

Good idea, done!

https://github.com/llvm/llvm-project/pull/178919