[llvm] 55fc5eb - [LV] Add additional cost model tests with inductions and truncates.
Florian Hahn via llvm-commits
llvm-commits at lists.llvm.org
Tue Apr 23 01:25:03 PDT 2024
Author: Florian Hahn
Date: 2024-04-23T09:24:01+01:00
New Revision: 55fc5eb95fbd26ac0539089cfc4e287ad7a233c3
URL: https://github.com/llvm/llvm-project/commit/55fc5eb95fbd26ac0539089cfc4e287ad7a233c3
DIFF: https://github.com/llvm/llvm-project/commit/55fc5eb95fbd26ac0539089cfc4e287ad7a233c3.diff
LOG: [LV] Add additional cost model tests with inductions and truncates.
Add test coverage for additional cases not covered by current tests with
multiple inductions and truncates.
Added:
llvm/test/Transforms/LoopVectorize/X86/masked-store-cost.ll
Modified:
llvm/test/Transforms/LoopVectorize/X86/induction-costs.ll
Removed:
################################################################################
diff --git a/llvm/test/Transforms/LoopVectorize/X86/induction-costs.ll b/llvm/test/Transforms/LoopVectorize/X86/induction-costs.ll
index 8ce87d0ef17100..9ae36ae4d58755 100644
--- a/llvm/test/Transforms/LoopVectorize/X86/induction-costs.ll
+++ b/llvm/test/Transforms/LoopVectorize/X86/induction-costs.ll
@@ -213,6 +213,79 @@ exit:
ret void
}
+define void @multiple_pointer_ivs_with_scalar_uses_only(ptr %A, ptr %B) #0 {
+entry:
+ br label %loop
+
+loop:
+ %iv.1 = phi i32 [ 100, %entry ], [ %dec, %loop ]
+ %iv.2 = phi i32 [ 2048, %entry ], [ %add38, %loop ]
+ %ptr.iv.1 = phi ptr [ %A, %entry ], [ %outptr.0, %loop ]
+ %ptr.iv.2 = phi ptr [ %B, %entry ], [ %incdec.ptr36, %loop ]
+ %ptr.iv.3 = phi ptr [ %B, %entry ], [ %incdec.ptr33, %loop ]
+ %incdec.ptr33 = getelementptr i8, ptr %ptr.iv.3, i64 1
+ %0 = load i8, ptr %ptr.iv.3, align 1
+ %conv34 = zext i8 %0 to i32
+ %incdec.ptr36 = getelementptr i8, ptr %ptr.iv.2, i64 1
+ %1 = load i8, ptr %ptr.iv.2, align 1
+ %conv37 = zext i8 %1 to i32
+ %add38 = add i32 %conv34, %conv37
+ %shr42 = lshr i32 %iv.2, 1
+ %conv43 = trunc i32 %shr42 to i8
+ store i8 %conv43, ptr %ptr.iv.1, align 1
+ %dec = add i32 %iv.1, 1
+ %outptr.0 = getelementptr i8, ptr %ptr.iv.1, i64 2
+ %cmp30.not = icmp eq i32 %dec, 0
+ br i1 %cmp30.not, label %exit, label %loop
+
+exit:
+ ret void
+}
+
+define i16 @iv_and_step_trunc() {
+entry:
+ br label %loop
+
+loop:
+ %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]
+ %rec = phi i16 [ 0, %entry ], [ %rec.next, %loop ]
+ %iv.next = add i64 %iv, 1
+ %0 = trunc i64 %iv to i16
+ %1 = trunc i64 %iv.next to i16
+ %rec.next = mul i16 %0, %1
+ %ec = icmp eq i64 %iv, 1
+ br i1 %ec, label %exit, label %loop
+
+exit:
+ ret i16 %rec
+}
+
+define i32 @test_scalar_predicated_cost(i64 %x, i64 %y, ptr %A) #0 {
+entry:
+ br label %loop.header
+
+loop.header:
+ %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop.latch ]
+ %cmp9.not = icmp ule i64 %iv, %y
+ br i1 %cmp9.not, label %loop.latch, label %if.then
+
+if.then:
+ %or = or i64 %x, %iv
+ %gep = getelementptr i32, ptr %A, i64 %iv
+ %t = trunc i64 %or to i32
+ store i32 %t, ptr %gep, align 4
+ br label %loop.latch
+
+loop.latch:
+ %iv.next = add i64 %iv, 1
+ %ec = icmp eq i64 %iv, 100
+ br i1 %ec, label %exit, label %loop.header
+
+exit:
+ ret i32 0
+}
+
+
attributes #0 = { "min-legal-vector-width"="0" "target-cpu"="skylake-avx512" }
;.
; CHECK: [[LOOP0]] = distinct !{[[LOOP0]], [[META1:![0-9]+]], [[META2:![0-9]+]]}
diff --git a/llvm/test/Transforms/LoopVectorize/X86/masked-store-cost.ll b/llvm/test/Transforms/LoopVectorize/X86/masked-store-cost.ll
new file mode 100644
index 00000000000000..2c8841caf87050
--- /dev/null
+++ b/llvm/test/Transforms/LoopVectorize/X86/masked-store-cost.ll
@@ -0,0 +1,111 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 4
+; RUN: opt -p loop-vectorize -mtriple=x86_64-apple-macosx -S %s | FileCheck %s
+
+target datalayout = "e-m:o-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128"
+
+define i32 @test_scalar_predicated_cost(i64 %x, i64 %y, ptr %A) #0 {
+; CHECK-LABEL: define i32 @test_scalar_predicated_cost(
+; CHECK-SAME: i64 [[X:%.*]], i64 [[Y:%.*]], ptr [[A:%.*]]) #[[ATTR0:[0-9]+]] {
+; CHECK-NEXT: entry:
+; CHECK-NEXT: br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
+; CHECK: vector.ph:
+; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <8 x i64> poison, i64 [[Y]], i64 0
+; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <8 x i64> [[BROADCAST_SPLATINSERT]], <8 x i64> poison, <8 x i32> zeroinitializer
+; CHECK-NEXT: [[BROADCAST_SPLATINSERT4:%.*]] = insertelement <8 x i64> poison, i64 [[X]], i64 0
+; CHECK-NEXT: [[BROADCAST_SPLAT5:%.*]] = shufflevector <8 x i64> [[BROADCAST_SPLATINSERT4]], <8 x i64> poison, <8 x i32> zeroinitializer
+; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]
+; CHECK: vector.body:
+; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
+; CHECK-NEXT: [[VEC_IND:%.*]] = phi <8 x i64> [ <i64 0, i64 1, i64 2, i64 3, i64 4, i64 5, i64 6, i64 7>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
+; CHECK-NEXT: [[STEP_ADD:%.*]] = add <8 x i64> [[VEC_IND]], <i64 8, i64 8, i64 8, i64 8, i64 8, i64 8, i64 8, i64 8>
+; CHECK-NEXT: [[STEP_ADD1:%.*]] = add <8 x i64> [[STEP_ADD]], <i64 8, i64 8, i64 8, i64 8, i64 8, i64 8, i64 8, i64 8>
+; CHECK-NEXT: [[STEP_ADD2:%.*]] = add <8 x i64> [[STEP_ADD1]], <i64 8, i64 8, i64 8, i64 8, i64 8, i64 8, i64 8, i64 8>
+; CHECK-NEXT: [[TMP0:%.*]] = add i64 [[INDEX]], 0
+; CHECK-NEXT: [[TMP1:%.*]] = add i64 [[INDEX]], 8
+; CHECK-NEXT: [[TMP2:%.*]] = add i64 [[INDEX]], 16
+; CHECK-NEXT: [[TMP3:%.*]] = add i64 [[INDEX]], 24
+; CHECK-NEXT: [[TMP4:%.*]] = icmp ule <8 x i64> [[VEC_IND]], [[BROADCAST_SPLAT]]
+; CHECK-NEXT: [[TMP5:%.*]] = icmp ule <8 x i64> [[STEP_ADD]], [[BROADCAST_SPLAT]]
+; CHECK-NEXT: [[TMP6:%.*]] = icmp ule <8 x i64> [[STEP_ADD1]], [[BROADCAST_SPLAT]]
+; CHECK-NEXT: [[TMP7:%.*]] = icmp ule <8 x i64> [[STEP_ADD2]], [[BROADCAST_SPLAT]]
+; CHECK-NEXT: [[TMP8:%.*]] = xor <8 x i1> [[TMP4]], <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>
+; CHECK-NEXT: [[TMP9:%.*]] = xor <8 x i1> [[TMP5]], <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>
+; CHECK-NEXT: [[TMP10:%.*]] = xor <8 x i1> [[TMP6]], <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>
+; CHECK-NEXT: [[TMP11:%.*]] = xor <8 x i1> [[TMP7]], <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>
+; CHECK-NEXT: [[TMP12:%.*]] = or <8 x i64> [[BROADCAST_SPLAT5]], [[VEC_IND]]
+; CHECK-NEXT: [[TMP13:%.*]] = or <8 x i64> [[BROADCAST_SPLAT5]], [[STEP_ADD]]
+; CHECK-NEXT: [[TMP14:%.*]] = or <8 x i64> [[BROADCAST_SPLAT5]], [[STEP_ADD1]]
+; CHECK-NEXT: [[TMP15:%.*]] = or <8 x i64> [[BROADCAST_SPLAT5]], [[STEP_ADD2]]
+; CHECK-NEXT: [[TMP16:%.*]] = getelementptr i32, ptr [[A]], i64 [[TMP0]]
+; CHECK-NEXT: [[TMP17:%.*]] = getelementptr i32, ptr [[A]], i64 [[TMP1]]
+; CHECK-NEXT: [[TMP18:%.*]] = getelementptr i32, ptr [[A]], i64 [[TMP2]]
+; CHECK-NEXT: [[TMP19:%.*]] = getelementptr i32, ptr [[A]], i64 [[TMP3]]
+; CHECK-NEXT: [[TMP20:%.*]] = trunc <8 x i64> [[TMP12]] to <8 x i32>
+; CHECK-NEXT: [[TMP21:%.*]] = trunc <8 x i64> [[TMP13]] to <8 x i32>
+; CHECK-NEXT: [[TMP22:%.*]] = trunc <8 x i64> [[TMP14]] to <8 x i32>
+; CHECK-NEXT: [[TMP23:%.*]] = trunc <8 x i64> [[TMP15]] to <8 x i32>
+; CHECK-NEXT: [[TMP24:%.*]] = getelementptr i32, ptr [[TMP16]], i32 0
+; CHECK-NEXT: [[TMP25:%.*]] = getelementptr i32, ptr [[TMP16]], i32 8
+; CHECK-NEXT: [[TMP26:%.*]] = getelementptr i32, ptr [[TMP16]], i32 16
+; CHECK-NEXT: [[TMP27:%.*]] = getelementptr i32, ptr [[TMP16]], i32 24
+; CHECK-NEXT: call void @llvm.masked.store.v8i32.p0(<8 x i32> [[TMP20]], ptr [[TMP24]], i32 4, <8 x i1> [[TMP8]])
+; CHECK-NEXT: call void @llvm.masked.store.v8i32.p0(<8 x i32> [[TMP21]], ptr [[TMP25]], i32 4, <8 x i1> [[TMP9]])
+; CHECK-NEXT: call void @llvm.masked.store.v8i32.p0(<8 x i32> [[TMP22]], ptr [[TMP26]], i32 4, <8 x i1> [[TMP10]])
+; CHECK-NEXT: call void @llvm.masked.store.v8i32.p0(<8 x i32> [[TMP23]], ptr [[TMP27]], i32 4, <8 x i1> [[TMP11]])
+; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 32
+; CHECK-NEXT: [[VEC_IND_NEXT]] = add <8 x i64> [[STEP_ADD2]], <i64 8, i64 8, i64 8, i64 8, i64 8, i64 8, i64 8, i64 8>
+; CHECK-NEXT: [[TMP28:%.*]] = icmp eq i64 [[INDEX_NEXT]], 96
+; CHECK-NEXT: br i1 [[TMP28]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
+; CHECK: middle.block:
+; CHECK-NEXT: br i1 false, label [[EXIT:%.*]], label [[SCALAR_PH]]
+; CHECK: scalar.ph:
+; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ 96, [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]
+; CHECK-NEXT: br label [[LOOP_HEADER:%.*]]
+; CHECK: loop.header:
+; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[LOOP_LATCH:%.*]] ]
+; CHECK-NEXT: [[CMP9_NOT:%.*]] = icmp ule i64 [[IV]], [[Y]]
+; CHECK-NEXT: br i1 [[CMP9_NOT]], label [[LOOP_LATCH]], label [[IF_THEN:%.*]]
+; CHECK: if.then:
+; CHECK-NEXT: [[OR:%.*]] = or i64 [[X]], [[IV]]
+; CHECK-NEXT: [[GEP:%.*]] = getelementptr i32, ptr [[A]], i64 [[IV]]
+; CHECK-NEXT: [[T:%.*]] = trunc i64 [[OR]] to i32
+; CHECK-NEXT: store i32 [[T]], ptr [[GEP]], align 4
+; CHECK-NEXT: br label [[LOOP_LATCH]]
+; CHECK: loop.latch:
+; CHECK-NEXT: [[IV_NEXT]] = add i64 [[IV]], 1
+; CHECK-NEXT: [[EC:%.*]] = icmp eq i64 [[IV]], 100
+; CHECK-NEXT: br i1 [[EC]], label [[EXIT]], label [[LOOP_HEADER]], !llvm.loop [[LOOP3:![0-9]+]]
+; CHECK: exit:
+; CHECK-NEXT: ret i32 0
+;
+entry:
+ br label %loop.header
+
+loop.header:
+ %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop.latch ]
+ %cmp9.not = icmp ule i64 %iv, %y
+ br i1 %cmp9.not, label %loop.latch, label %if.then
+
+if.then:
+ %or = or i64 %x, %iv
+ %gep = getelementptr i32, ptr %A, i64 %iv
+ %t = trunc i64 %or to i32
+ store i32 %t, ptr %gep, align 4
+ br label %loop.latch
+
+loop.latch:
+ %iv.next = add i64 %iv, 1
+ %ec = icmp eq i64 %iv, 100
+ br i1 %ec, label %exit, label %loop.header
+
+exit:
+ ret i32 0
+}
+
+attributes #0 = { "min-legal-vector-width"="0" "target-cpu"="skylake-avx512" }
+;.
+; CHECK: [[LOOP0]] = distinct !{[[LOOP0]], [[META1:![0-9]+]], [[META2:![0-9]+]]}
+; CHECK: [[META1]] = !{!"llvm.loop.isvectorized", i32 1}
+; CHECK: [[META2]] = !{!"llvm.loop.unroll.runtime.disable"}
+; CHECK: [[LOOP3]] = distinct !{[[LOOP3]], [[META2]], [[META1]]}
+;.
More information about the llvm-commits
mailing list