[llvm] [VPlan] Add transformation to narrow interleave groups. (PR #106441)

via llvm-commits llvm-commits at lists.llvm.org
Wed Jan 8 17:01:28 PST 2025


================
@@ -0,0 +1,746 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
+; RUN: opt -p loop-vectorize -mcpu=skylake -S %s | FileCheck %s
+; https://github.com/llvm/llvm-project/issues/82936
+
+target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128"
+target triple = "x86_64-unknown-linux"
+
+define void @test_4xi64(ptr noalias %data, ptr noalias %factor, i64 noundef %n) {
+; CHECK-LABEL: define void @test_4xi64(
+; CHECK-SAME: ptr noalias [[DATA:%.*]], ptr noalias [[FACTOR:%.*]], i64 noundef [[N:%.*]]) #[[ATTR0:[0-9]+]] {
+; CHECK-NEXT:  [[ENTRY:.*]]:
+; CHECK-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N]], 4
+; CHECK-NEXT:    br i1 [[MIN_ITERS_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]]
+; CHECK:       [[VECTOR_PH]]:
+; CHECK-NEXT:    [[N_MOD_VF:%.*]] = urem i64 [[N]], 4
+; CHECK-NEXT:    [[N_VEC:%.*]] = sub i64 [[N]], [[N_MOD_VF]]
+; CHECK-NEXT:    br label %[[VECTOR_BODY:.*]]
+; CHECK:       [[VECTOR_BODY]]:
+; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
+; CHECK-NEXT:    [[IV:%.*]] = add i64 [[INDEX]], 0
+; CHECK-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds i64, ptr [[FACTOR]], i64 [[IV]]
+; CHECK-NEXT:    [[TMP2:%.*]] = getelementptr inbounds i64, ptr [[ARRAYIDX]], i32 0
+; CHECK-NEXT:    [[TMP3:%.*]] = getelementptr inbounds { i64, i64, i64, i64 }, ptr [[DATA]], i64 [[IV]], i32 0
+; CHECK-NEXT:    [[WIDE_LOAD:%.*]] = load <4 x i64>, ptr [[TMP3]], align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load i64, ptr [[TMP2]], align 8
+; CHECK-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <4 x i64> poison, i64 [[TMP4]], i64 0
+; CHECK-NEXT:    [[STRIDED_VEC2:%.*]] = shufflevector <4 x i64> [[BROADCAST_SPLATINSERT]], <4 x i64> poison, <4 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP7:%.*]] = mul <4 x i64> [[STRIDED_VEC2]], [[WIDE_LOAD]]
+; CHECK-NEXT:    store <4 x i64> [[TMP7]], ptr [[TMP3]], align 8
+; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 1
+; 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:    [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]]
+; CHECK-NEXT:    br i1 [[CMP_N]], label %[[EXIT:.*]], label %[[SCALAR_PH]]
+; CHECK:       [[SCALAR_PH]]:
+; CHECK-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], %[[MIDDLE_BLOCK]] ], [ 0, %[[ENTRY]] ]
+; CHECK-NEXT:    br label %[[LOOP:.*]]
+; CHECK:       [[LOOP]]:
+; CHECK-NEXT:    [[IV1:%.*]] = phi i64 [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], %[[LOOP]] ]
+; CHECK-NEXT:    [[DATA_2:%.*]] = getelementptr inbounds i64, ptr [[FACTOR]], i64 [[IV1]]
+; CHECK-NEXT:    [[L_2:%.*]] = load i64, ptr [[DATA_2]], align 8
+; CHECK-NEXT:    [[DATA_0:%.*]] = getelementptr inbounds { i64, i64, i64, i64 }, ptr [[DATA]], i64 [[IV1]], i32 0
+; CHECK-NEXT:    [[L_0:%.*]] = load i64, ptr [[DATA_0]], align 8
+; CHECK-NEXT:    [[MUL_0:%.*]] = mul i64 [[L_2]], [[L_0]]
+; CHECK-NEXT:    store i64 [[MUL_0]], ptr [[DATA_0]], align 8
+; CHECK-NEXT:    [[DATA_1:%.*]] = getelementptr inbounds { i64, i64, i64, i64 }, ptr [[DATA]], i64 [[IV1]], i32 1
+; CHECK-NEXT:    [[L_1:%.*]] = load i64, ptr [[DATA_1]], align 8
+; CHECK-NEXT:    [[MUL_1:%.*]] = mul i64 [[L_2]], [[L_1]]
+; CHECK-NEXT:    store i64 [[MUL_1]], ptr [[DATA_1]], align 8
+; CHECK-NEXT:    [[DATA_4:%.*]] = getelementptr inbounds { i64, i64, i64, i64 }, ptr [[DATA]], i64 [[IV1]], i32 2
+; CHECK-NEXT:    [[L_4:%.*]] = load i64, ptr [[DATA_4]], align 8
+; CHECK-NEXT:    [[MUL_2:%.*]] = mul i64 [[L_2]], [[L_4]]
+; CHECK-NEXT:    store i64 [[MUL_2]], ptr [[DATA_4]], align 8
+; CHECK-NEXT:    [[DATA_3:%.*]] = getelementptr inbounds { i64, i64, i64, i64 }, ptr [[DATA]], i64 [[IV1]], i32 3
+; CHECK-NEXT:    [[L_3:%.*]] = load i64, ptr [[DATA_3]], align 8
+; CHECK-NEXT:    [[MUL_3:%.*]] = mul i64 [[L_2]], [[L_3]]
+; CHECK-NEXT:    store i64 [[MUL_3]], ptr [[DATA_3]], align 8
+; CHECK-NEXT:    [[IV_NEXT]] = add nuw nsw i64 [[IV1]], 1
+; CHECK-NEXT:    [[EC:%.*]] = icmp eq i64 [[IV_NEXT]], [[N]]
+; CHECK-NEXT:    br i1 [[EC]], label %[[EXIT]], label %[[LOOP]], !llvm.loop [[LOOP3:![0-9]+]]
+; CHECK:       [[EXIT]]:
+; CHECK-NEXT:    ret void
+;
+entry:
+  br label %loop
+
+loop:
+  %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]
+  %arrayidx = getelementptr inbounds i64, ptr %factor, i64 %iv
+  %l.factor = load i64, ptr %arrayidx, align 8
+  %data.0 = getelementptr inbounds { i64 , i64, i64, i64 }, ptr %data, i64 %iv, i32 0
+  %l.0 = load i64, ptr %data.0, align 8
+  %mul.0 = mul i64 %l.factor, %l.0
+  store i64 %mul.0, ptr %data.0, align 8
+  %data.1 = getelementptr inbounds { i64 , i64, i64, i64 }, ptr %data, i64 %iv, i32 1
+  %l.1 = load i64, ptr %data.1, align 8
+  %mul.1 = mul i64 %l.factor, %l.1
+  store i64 %mul.1, ptr %data.1, align 8
+  %data.2 = getelementptr inbounds { i64 , i64, i64, i64 }, ptr %data, i64 %iv, i32 2
+  %l.2 = load i64, ptr %data.2, align 8
+  %mul.2 = mul i64 %l.factor, %l.2
+  store i64 %mul.2, ptr %data.2, align 8
+  %data.3 = getelementptr inbounds { i64 , i64, i64, i64 }, ptr %data, i64 %iv, i32 3
+  %l.3 = load i64, ptr %data.3, align 8
+  %mul.3 = mul i64 %l.factor, %l.3
+  store i64 %mul.3, ptr %data.3, align 8
+  %iv.next = add nuw nsw i64 %iv, 1
+  %ec = icmp eq i64 %iv.next, %n
+  br i1 %ec, label %exit, label %loop
+
+exit:
+  ret void
+}
+
+
+define void @test_2xi64(ptr noalias %data, ptr noalias %factor, i64 noundef %n) {
+; CHECK-LABEL: define void @test_2xi64(
+; CHECK-SAME: ptr noalias [[DATA:%.*]], ptr noalias [[FACTOR:%.*]], i64 noundef [[N:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:  [[ENTRY:.*]]:
+; CHECK-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N]], 4
+; CHECK-NEXT:    br i1 [[MIN_ITERS_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]]
+; CHECK:       [[VECTOR_PH]]:
+; CHECK-NEXT:    [[N_MOD_VF:%.*]] = urem i64 [[N]], 4
+; CHECK-NEXT:    [[N_VEC:%.*]] = sub i64 [[N]], [[N_MOD_VF]]
+; CHECK-NEXT:    br label %[[VECTOR_BODY:.*]]
+; CHECK:       [[VECTOR_BODY]]:
+; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
+; CHECK-NEXT:    [[IV:%.*]] = add i64 [[INDEX]], 0
+; CHECK-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds i64, ptr [[FACTOR]], i64 [[IV]]
+; CHECK-NEXT:    [[TMP2:%.*]] = getelementptr inbounds i64, ptr [[ARRAYIDX]], i32 0
+; CHECK-NEXT:    [[WIDE_LOAD:%.*]] = load <4 x i64>, ptr [[TMP2]], align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = shl nsw i64 [[IV]], 1
+; CHECK-NEXT:    [[TMP4:%.*]] = getelementptr inbounds i64, ptr [[DATA]], i64 [[TMP3]]
+; CHECK-NEXT:    [[WIDE_VEC:%.*]] = load <8 x i64>, ptr [[TMP4]], align 8
+; CHECK-NEXT:    [[STRIDED_VEC:%.*]] = shufflevector <8 x i64> [[WIDE_VEC]], <8 x i64> poison, <4 x i32> <i32 0, i32 2, i32 4, i32 6>
+; CHECK-NEXT:    [[STRIDED_VEC1:%.*]] = shufflevector <8 x i64> [[WIDE_VEC]], <8 x i64> poison, <4 x i32> <i32 1, i32 3, i32 5, i32 7>
+; CHECK-NEXT:    [[TMP6:%.*]] = mul <4 x i64> [[WIDE_LOAD]], [[STRIDED_VEC]]
+; CHECK-NEXT:    [[TMP9:%.*]] = mul <4 x i64> [[WIDE_LOAD]], [[STRIDED_VEC1]]
+; CHECK-NEXT:    [[TMP11:%.*]] = shufflevector <4 x i64> [[TMP6]], <4 x i64> [[TMP9]], <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
+; CHECK-NEXT:    [[INTERLEAVED_VEC:%.*]] = shufflevector <8 x i64> [[TMP11]], <8 x i64> poison, <8 x i32> <i32 0, i32 4, i32 1, i32 5, i32 2, i32 6, i32 3, i32 7>
+; CHECK-NEXT:    store <8 x i64> [[INTERLEAVED_VEC]], ptr [[TMP4]], align 8
+; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
----------------
ayalz wrote:

This case didn't seems to get narrowed, because VF=4 and IG's have 2 members?

https://github.com/llvm/llvm-project/pull/106441


More information about the llvm-commits mailing list