[llvm] [LoopVectorizer] Add support for partial reductions (PR #92418)

[Sam Tebbs via llvm-commits]

================
@@ -0,0 +1,283 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 4
+; RUN: opt -passes=loop-vectorize -force-vector-interleave=1 -S < %s | FileCheck %s
+
+target datalayout = "e-m:e-i8:8:32-i16:16:32-i64:64-i128:128-n32:64-S128"
+target triple = "aarch64-none-unknown-elf"
+
+define void @dotp(ptr %a, ptr %b) #0 {
+; CHECK-LABEL: define void @dotp(
+; CHECK-SAME: ptr [[A:%.*]], ptr [[B:%.*]]) #[[ATTR0:[0-9]+]] {
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[TMP0:%.*]] = call i64 @llvm.vscale.i64()
+; CHECK-NEXT:    [[TMP1:%.*]] = mul i64 [[TMP0]], 16
+; CHECK-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 0, [[TMP1]]
+; CHECK-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
+; CHECK:       vector.ph:
+; CHECK-NEXT:    [[TMP2:%.*]] = call i64 @llvm.vscale.i64()
+; CHECK-NEXT:    [[TMP3:%.*]] = mul i64 [[TMP2]], 16
+; CHECK-NEXT:    [[N_MOD_VF:%.*]] = urem i64 0, [[TMP3]]
+; CHECK-NEXT:    [[N_VEC:%.*]] = sub i64 0, [[N_MOD_VF]]
+; CHECK-NEXT:    [[TMP4:%.*]] = call i64 @llvm.vscale.i64()
+; CHECK-NEXT:    [[TMP5:%.*]] = mul i64 [[TMP4]], 16
+; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
+; CHECK:       vector.body:
+; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
+; CHECK-NEXT:    [[VEC_PHI:%.*]] = phi <vscale x 4 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE:%.*]], [[VECTOR_BODY]] ]
+; CHECK-NEXT:    [[TMP6:%.*]] = add i64 [[INDEX]], 0
+; CHECK-NEXT:    [[TMP7:%.*]] = getelementptr i8, ptr [[A]], i64 [[TMP6]]
+; CHECK-NEXT:    [[TMP8:%.*]] = getelementptr i8, ptr [[TMP7]], i32 0
+; CHECK-NEXT:    [[WIDE_LOAD:%.*]] = load <vscale x 16 x i8>, ptr [[TMP8]], align 1
+; CHECK-NEXT:    [[TMP9:%.*]] = zext <vscale x 16 x i8> [[WIDE_LOAD]] to <vscale x 16 x i32>
+; CHECK-NEXT:    [[TMP10:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP6]]
+; CHECK-NEXT:    [[TMP11:%.*]] = getelementptr i8, ptr [[TMP10]], i32 0
+; CHECK-NEXT:    [[WIDE_LOAD1:%.*]] = load <vscale x 16 x i8>, ptr [[TMP11]], align 1
+; CHECK-NEXT:    [[TMP15:%.*]] = zext <vscale x 16 x i8> [[WIDE_LOAD1]] to <vscale x 16 x i32>
+; CHECK-NEXT:    [[TMP16:%.*]] = mul <vscale x 16 x i32> [[TMP15]], [[TMP9]]
+; CHECK-NEXT:    [[PARTIAL_REDUCE]] = call <vscale x 4 x i32> @llvm.experimental.vector.partial.reduce.add.nxv4i32.nxv16i32(<vscale x 4 x i32> [[VEC_PHI]], <vscale x 16 x i32> [[TMP16]])
+; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP5]]
+; CHECK-NEXT:    [[TMP14:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
+; CHECK-NEXT:    br i1 [[TMP14]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
+; CHECK:       middle.block:
+; CHECK-NEXT:    [[TMP17:%.*]] = call i32 @llvm.vector.reduce.add.nxv4i32(<vscale x 4 x i32> [[PARTIAL_REDUCE]])
+; CHECK-NEXT:    [[CMP_N:%.*]] = icmp eq i64 0, [[N_VEC]]
+; CHECK-NEXT:    br i1 [[CMP_N]], label [[FOR_COND_CLEANUP_LOOPEXIT:%.*]], label [[SCALAR_PH]]
+;
+entry:
+  br label %for.body
+
+for.cond.cleanup.loopexit:                        ; preds = %for.body
+  %0 = lshr i32 %add, 0
----------------
SamTebbs33 wrote:

Done.

https://github.com/llvm/llvm-project/pull/92418