[LoopVectorize] Add tests for dereferenceable loads in more loops #118470
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
* Adds tests for strided accesses. * Adds tests for reverse loops. As part of this I've moved one of the negative tests from load-deref-pred-align.ll into a new file (load-deref-pred-neg-off.ll) because the pointer type had a size of 16 bits and I realised it's probably not sensible for allocas that are >16 bits in size!
|
@llvm/pr-subscribers-llvm-transforms Author: David Sherwood (david-arm) Changes
As part of this I've moved one of the negative tests from Patch is 36.14 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/118470.diff 2 Files Affected:
diff --git a/llvm/test/Transforms/LoopVectorize/load-deref-pred-align.ll b/llvm/test/Transforms/LoopVectorize/load-deref-pred-align.ll
index 1ef01e3b793d5b..bf22d63850835d 100644
--- a/llvm/test/Transforms/LoopVectorize/load-deref-pred-align.ll
+++ b/llvm/test/Transforms/LoopVectorize/load-deref-pred-align.ll
@@ -1,7 +1,7 @@
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
; RUN: opt -passes=loop-vectorize -force-vector-width=2 -force-vector-interleave=1 -S %s | FileCheck %s
-target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128-ni:1-p:16:16:16:16"
+target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128-ni:1"
declare void @init(ptr nocapture nofree)
@@ -200,104 +200,6 @@ loop_exit:
}
-; Test where offset relative to alloca is negative and we shouldn't
-; treat predicated loads as being always dereferenceable.
-define i8 @test_negative_off(i16 %len, ptr %test_base) {
-; CHECK-LABEL: @test_negative_off(
-; CHECK-NEXT: entry:
-; CHECK-NEXT: [[ALLOCA:%.*]] = alloca [64638 x i8], align 1
-; CHECK-NEXT: call void @init(ptr [[ALLOCA]])
-; CHECK-NEXT: br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
-; CHECK: vector.ph:
-; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]
-; CHECK: vector.body:
-; CHECK-NEXT: [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[PRED_LOAD_CONTINUE2:%.*]] ]
-; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <2 x i8> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP18:%.*]], [[PRED_LOAD_CONTINUE2]] ]
-; CHECK-NEXT: [[DOTCAST:%.*]] = trunc i32 [[INDEX]] to i16
-; CHECK-NEXT: [[OFFSET_IDX:%.*]] = add i16 -1000, [[DOTCAST]]
-; CHECK-NEXT: [[TMP0:%.*]] = add i16 [[OFFSET_IDX]], 0
-; CHECK-NEXT: [[TMP1:%.*]] = add i16 [[OFFSET_IDX]], 1
-; CHECK-NEXT: [[TMP2:%.*]] = getelementptr inbounds i1, ptr [[TEST_BASE:%.*]], i16 [[TMP0]]
-; CHECK-NEXT: [[TMP3:%.*]] = getelementptr inbounds i1, ptr [[TEST_BASE]], i16 [[TMP1]]
-; CHECK-NEXT: [[TMP4:%.*]] = load i1, ptr [[TMP2]], align 1
-; CHECK-NEXT: [[TMP5:%.*]] = load i1, ptr [[TMP3]], align 1
-; CHECK-NEXT: [[TMP6:%.*]] = insertelement <2 x i1> poison, i1 [[TMP4]], i32 0
-; CHECK-NEXT: [[TMP7:%.*]] = insertelement <2 x i1> [[TMP6]], i1 [[TMP5]], i32 1
-; CHECK-NEXT: [[TMP8:%.*]] = extractelement <2 x i1> [[TMP7]], i32 0
-; CHECK-NEXT: br i1 [[TMP8]], label [[PRED_LOAD_IF:%.*]], label [[PRED_LOAD_CONTINUE:%.*]]
-; CHECK: pred.load.if:
-; CHECK-NEXT: [[TMP9:%.*]] = getelementptr i8, ptr [[ALLOCA]], i16 [[TMP0]]
-; CHECK-NEXT: [[TMP10:%.*]] = load i8, ptr [[TMP9]], align 1
-; CHECK-NEXT: [[TMP11:%.*]] = insertelement <2 x i8> poison, i8 [[TMP10]], i32 0
-; CHECK-NEXT: br label [[PRED_LOAD_CONTINUE]]
-; CHECK: pred.load.continue:
-; CHECK-NEXT: [[TMP12:%.*]] = phi <2 x i8> [ poison, [[VECTOR_BODY]] ], [ [[TMP11]], [[PRED_LOAD_IF]] ]
-; CHECK-NEXT: [[TMP13:%.*]] = extractelement <2 x i1> [[TMP7]], i32 1
-; CHECK-NEXT: br i1 [[TMP13]], label [[PRED_LOAD_IF1:%.*]], label [[PRED_LOAD_CONTINUE2]]
-; CHECK: pred.load.if1:
-; CHECK-NEXT: [[TMP14:%.*]] = getelementptr i8, ptr [[ALLOCA]], i16 [[TMP1]]
-; CHECK-NEXT: [[TMP15:%.*]] = load i8, ptr [[TMP14]], align 1
-; CHECK-NEXT: [[TMP16:%.*]] = insertelement <2 x i8> [[TMP12]], i8 [[TMP15]], i32 1
-; CHECK-NEXT: br label [[PRED_LOAD_CONTINUE2]]
-; CHECK: pred.load.continue2:
-; CHECK-NEXT: [[TMP17:%.*]] = phi <2 x i8> [ [[TMP12]], [[PRED_LOAD_CONTINUE]] ], [ [[TMP16]], [[PRED_LOAD_IF1]] ]
-; CHECK-NEXT: [[PREDPHI:%.*]] = select <2 x i1> [[TMP7]], <2 x i8> [[TMP17]], <2 x i8> zeroinitializer
-; CHECK-NEXT: [[TMP18]] = add <2 x i8> [[VEC_PHI]], [[PREDPHI]]
-; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 2
-; CHECK-NEXT: [[TMP19:%.*]] = icmp eq i32 [[INDEX_NEXT]], 12
-; CHECK-NEXT: br i1 [[TMP19]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP6:![0-9]+]]
-; CHECK: middle.block:
-; CHECK-NEXT: [[TMP20:%.*]] = call i8 @llvm.vector.reduce.add.v2i8(<2 x i8> [[TMP18]])
-; CHECK-NEXT: br i1 true, label [[LOOP_EXIT:%.*]], label [[SCALAR_PH]]
-; CHECK: scalar.ph:
-; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i16 [ -988, [[MIDDLE_BLOCK]] ], [ -1000, [[ENTRY:%.*]] ]
-; CHECK-NEXT: [[BC_MERGE_RDX:%.*]] = phi i8 [ [[TMP20]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY]] ]
-; CHECK-NEXT: br label [[LOOP:%.*]]
-; CHECK: loop:
-; CHECK-NEXT: [[IV:%.*]] = phi i16 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[LATCH:%.*]] ]
-; CHECK-NEXT: [[ACCUM:%.*]] = phi i8 [ [[BC_MERGE_RDX]], [[SCALAR_PH]] ], [ [[ACCUM_NEXT:%.*]], [[LATCH]] ]
-; CHECK-NEXT: [[IV_NEXT]] = add i16 [[IV]], 1
-; CHECK-NEXT: [[TEST_ADDR:%.*]] = getelementptr inbounds i1, ptr [[TEST_BASE]], i16 [[IV]]
-; CHECK-NEXT: [[EARLYCND:%.*]] = load i1, ptr [[TEST_ADDR]], align 1
-; CHECK-NEXT: br i1 [[EARLYCND]], label [[PRED:%.*]], label [[LATCH]]
-; CHECK: pred:
-; CHECK-NEXT: [[ADDR:%.*]] = getelementptr i8, ptr [[ALLOCA]], i16 [[IV]]
-; CHECK-NEXT: [[VAL:%.*]] = load i8, ptr [[ADDR]], align 1
-; CHECK-NEXT: br label [[LATCH]]
-; CHECK: latch:
-; CHECK-NEXT: [[VAL_PHI:%.*]] = phi i8 [ 0, [[LOOP]] ], [ [[VAL]], [[PRED]] ]
-; CHECK-NEXT: [[ACCUM_NEXT]] = add i8 [[ACCUM]], [[VAL_PHI]]
-; CHECK-NEXT: [[EXIT:%.*]] = icmp ugt i16 [[IV]], -990
-; CHECK-NEXT: br i1 [[EXIT]], label [[LOOP_EXIT]], label [[LOOP]], !llvm.loop [[LOOP7:![0-9]+]]
-; CHECK: loop_exit:
-; CHECK-NEXT: [[ACCUM_NEXT_LCSSA:%.*]] = phi i8 [ [[ACCUM_NEXT]], [[LATCH]] ], [ [[TMP20]], [[MIDDLE_BLOCK]] ]
-; CHECK-NEXT: ret i8 [[ACCUM_NEXT_LCSSA]]
-;
-entry:
- %alloca = alloca [64638 x i8]
- call void @init(ptr %alloca)
- br label %loop
-loop:
- %iv = phi i16 [ -1000, %entry ], [ %iv.next, %latch ]
- %accum = phi i8 [ 0, %entry ], [ %accum.next, %latch ]
- %iv.next = add i16 %iv, 1
- %test_addr = getelementptr inbounds i1, ptr %test_base, i16 %iv
- %earlycnd = load i1, ptr %test_addr
- br i1 %earlycnd, label %pred, label %latch
-pred:
- %addr = getelementptr i8, ptr %alloca, i16 %iv
- %val = load i8, ptr %addr
- br label %latch
-latch:
- %val.phi = phi i8 [ 0, %loop ], [ %val, %pred ]
- %accum.next = add i8 %accum, %val.phi
- %exit = icmp ugt i16 %iv, -990
- br i1 %exit, label %loop_exit, label %loop
-loop_exit:
- ret i8 %accum.next
-}
-
-
define i32 @loop_requires_scev_predicate(ptr %dest, i32 %end) {
; CHECK-LABEL: @loop_requires_scev_predicate(
; CHECK-NEXT: entry:
@@ -423,3 +325,430 @@ for.inc:
exit:
ret i32 0
}
+
+
+; Test reverse loops where we should be able to prove loads in predicated blocks
+; are safe to load unconditionally.
+define void @test_rev_loops_deref_loads(ptr nocapture noundef writeonly %dest) {
+; CHECK-LABEL: @test_rev_loops_deref_loads(
+; CHECK-NEXT: entry:
+; CHECK-NEXT: [[LOCAL_DEST:%.*]] = alloca [1024 x i32], align 4
+; CHECK-NEXT: [[LOCAL_SRC:%.*]] = alloca [1024 x i32], align 4
+; CHECK-NEXT: [[LOCAL_CMP:%.*]] = alloca [1024 x i32], align 4
+; CHECK-NEXT: call void @init(ptr [[LOCAL_SRC]])
+; CHECK-NEXT: call void @init(ptr [[LOCAL_CMP]])
+; CHECK-NEXT: br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
+; CHECK: vector.ph:
+; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]
+; CHECK: vector.body:
+; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[PRED_STORE_CONTINUE4:%.*]] ]
+; CHECK-NEXT: [[OFFSET_IDX:%.*]] = sub i64 1023, [[INDEX]]
+; CHECK-NEXT: [[TMP0:%.*]] = add i64 [[OFFSET_IDX]], 0
+; CHECK-NEXT: [[TMP1:%.*]] = getelementptr inbounds [1024 x i32], ptr [[LOCAL_CMP]], i64 0, i64 [[TMP0]]
+; CHECK-NEXT: [[TMP2:%.*]] = getelementptr inbounds i32, ptr [[TMP1]], i32 0
+; CHECK-NEXT: [[TMP3:%.*]] = getelementptr inbounds i32, ptr [[TMP2]], i32 -1
+; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <2 x i32>, ptr [[TMP3]], align 4
+; CHECK-NEXT: [[REVERSE:%.*]] = shufflevector <2 x i32> [[WIDE_LOAD]], <2 x i32> poison, <2 x i32> <i32 1, i32 0>
+; CHECK-NEXT: [[TMP4:%.*]] = icmp eq <2 x i32> [[REVERSE]], splat (i32 3)
+; CHECK-NEXT: [[TMP5:%.*]] = xor <2 x i1> [[TMP4]], splat (i1 true)
+; CHECK-NEXT: [[TMP9:%.*]] = extractelement <2 x i1> [[TMP5]], i32 0
+; CHECK-NEXT: br i1 [[TMP9]], label [[PRED_STORE_IF:%.*]], label [[PRED_STORE_CONTINUE:%.*]]
+; CHECK: pred.store.if:
+; CHECK-NEXT: [[TMP7:%.*]] = getelementptr inbounds [1024 x i32], ptr [[LOCAL_SRC]], i64 0, i64 [[TMP0]]
+; CHECK-NEXT: [[TMP11:%.*]] = load i32, ptr [[TMP7]], align 4
+; CHECK-NEXT: [[TMP10:%.*]] = getelementptr inbounds [1024 x i32], ptr [[LOCAL_DEST]], i64 0, i64 [[TMP0]]
+; CHECK-NEXT: [[TMP12:%.*]] = shl nsw i32 [[TMP11]], 2
+; CHECK-NEXT: store i32 [[TMP12]], ptr [[TMP10]], align 4
+; CHECK-NEXT: br label [[PRED_STORE_CONTINUE]]
+; CHECK: pred.store.continue:
+; CHECK-NEXT: [[TMP13:%.*]] = extractelement <2 x i1> [[TMP5]], i32 1
+; CHECK-NEXT: br i1 [[TMP13]], label [[PRED_STORE_IF3:%.*]], label [[PRED_STORE_CONTINUE4]]
+; CHECK: pred.store.if1:
+; CHECK-NEXT: [[TMP14:%.*]] = add i64 [[OFFSET_IDX]], -1
+; CHECK-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1024 x i32], ptr [[LOCAL_SRC]], i64 0, i64 [[TMP14]]
+; CHECK-NEXT: [[TMP16:%.*]] = load i32, ptr [[TMP21]], align 4
+; CHECK-NEXT: [[TMP15:%.*]] = getelementptr inbounds [1024 x i32], ptr [[LOCAL_DEST]], i64 0, i64 [[TMP14]]
+; CHECK-NEXT: [[TMP17:%.*]] = shl nsw i32 [[TMP16]], 2
+; CHECK-NEXT: store i32 [[TMP17]], ptr [[TMP15]], align 4
+; CHECK-NEXT: br label [[PRED_STORE_CONTINUE4]]
+; CHECK: pred.store.continue2:
+; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2
+; CHECK-NEXT: [[TMP18:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1024
+; CHECK-NEXT: br i1 [[TMP18]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP10:![0-9]+]]
+; CHECK: middle.block:
+; CHECK-NEXT: br i1 true, label [[EXIT:%.*]], label [[SCALAR_PH]]
+; CHECK: scalar.ph:
+; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ -1, [[MIDDLE_BLOCK]] ], [ 1023, [[ENTRY:%.*]] ]
+; CHECK-NEXT: br label [[FOR_BODY:%.*]]
+; CHECK: for.body:
+; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[FOR_INC:%.*]] ]
+; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [1024 x i32], ptr [[LOCAL_CMP]], i64 0, i64 [[IV]]
+; CHECK-NEXT: [[TMP19:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
+; CHECK-NEXT: [[CMP3_NOT:%.*]] = icmp eq i32 [[TMP19]], 3
+; CHECK-NEXT: br i1 [[CMP3_NOT]], label [[FOR_INC]], label [[IF_THEN:%.*]]
+; CHECK: if.then:
+; CHECK-NEXT: [[ARRAYIDX5:%.*]] = getelementptr inbounds [1024 x i32], ptr [[LOCAL_SRC]], i64 0, i64 [[IV]]
+; CHECK-NEXT: [[TMP20:%.*]] = load i32, ptr [[ARRAYIDX5]], align 4
+; CHECK-NEXT: [[MUL:%.*]] = shl nsw i32 [[TMP20]], 2
+; CHECK-NEXT: [[ARRAYIDX7:%.*]] = getelementptr inbounds [1024 x i32], ptr [[LOCAL_DEST]], i64 0, i64 [[IV]]
+; CHECK-NEXT: store i32 [[MUL]], ptr [[ARRAYIDX7]], align 4
+; CHECK-NEXT: br label [[FOR_INC]]
+; CHECK: for.inc:
+; CHECK-NEXT: [[IV_NEXT]] = add nsw i64 [[IV]], -1
+; CHECK-NEXT: [[CMP2_NOT:%.*]] = icmp eq i64 [[IV]], 0
+; CHECK-NEXT: br i1 [[CMP2_NOT]], label [[EXIT]], label [[FOR_BODY]], !llvm.loop [[LOOP11:![0-9]+]]
+; CHECK: exit:
+; CHECK-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr [[DEST:%.*]], ptr [[LOCAL_DEST]], i64 1024, i1 false)
+; CHECK-NEXT: ret void
+;
+entry:
+ %local_dest = alloca [1024 x i32], align 4
+ %local_src = alloca [1024 x i32], align 4
+ %local_cmp = alloca [1024 x i32], align 4
+ call void @init(ptr %local_src)
+ call void @init(ptr %local_cmp)
+ br label %for.body
+
+for.body:
+ %iv = phi i64 [ 1023, %entry ], [ %iv.next, %for.inc ]
+ %arrayidx = getelementptr inbounds [1024 x i32], ptr %local_cmp, i64 0, i64 %iv
+ %0 = load i32, ptr %arrayidx, align 4
+ %cmp3.not = icmp eq i32 %0, 3
+ br i1 %cmp3.not, label %for.inc, label %if.then
+
+if.then:
+ %arrayidx5 = getelementptr inbounds [1024 x i32], ptr %local_src, i64 0, i64 %iv
+ %1 = load i32, ptr %arrayidx5, align 4
+ %mul = shl nsw i32 %1, 2
+ %arrayidx7 = getelementptr inbounds [1024 x i32], ptr %local_dest, i64 0, i64 %iv
+ store i32 %mul, ptr %arrayidx7, align 4
+ br label %for.inc
+
+for.inc:
+ %iv.next = add nsw i64 %iv, -1
+ %cmp2.not = icmp eq i64 %iv, 0
+ br i1 %cmp2.not, label %exit, label %for.body
+
+exit:
+ call void @llvm.memcpy.p0.p0.i64(ptr %dest, ptr %local_dest, i64 1024, i1 false)
+ ret void
+}
+
+
+; Test reverse loops where we *cannot* prove loads in predicated blocks are safe
+; to load unconditionally.
+define void @test_rev_loops_non_deref_loads(ptr nocapture noundef writeonly %dest) {
+; CHECK-LABEL: @test_rev_loops_non_deref_loads(
+; CHECK-NEXT: entry:
+; CHECK-NEXT: [[LOCAL_DEST:%.*]] = alloca [1024 x i32], align 4
+; CHECK-NEXT: [[LOCAL_SRC:%.*]] = alloca [1024 x i32], align 4
+; CHECK-NEXT: [[LOCAL_CMP:%.*]] = alloca [1024 x i32], align 4
+; CHECK-NEXT: call void @init(ptr [[LOCAL_SRC]])
+; CHECK-NEXT: call void @init(ptr [[LOCAL_CMP]])
+; CHECK-NEXT: br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
+; CHECK: vector.ph:
+; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]
+; CHECK: vector.body:
+; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[PRED_STORE_CONTINUE2:%.*]] ]
+; CHECK-NEXT: [[VEC_IND:%.*]] = phi <2 x i64> [ <i64 1023, i64 1022>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[PRED_STORE_CONTINUE2]] ]
+; CHECK-NEXT: [[TMP0:%.*]] = add <2 x i64> [[VEC_IND]], splat (i64 -1)
+; CHECK-NEXT: [[TMP1:%.*]] = extractelement <2 x i64> [[TMP0]], i32 0
+; CHECK-NEXT: [[TMP2:%.*]] = getelementptr inbounds [1024 x i32], ptr [[LOCAL_CMP]], i64 0, i64 [[TMP1]]
+; CHECK-NEXT: [[TMP3:%.*]] = getelementptr inbounds i32, ptr [[TMP2]], i32 0
+; CHECK-NEXT: [[TMP4:%.*]] = getelementptr inbounds i32, ptr [[TMP3]], i32 -1
+; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <2 x i32>, ptr [[TMP4]], align 4
+; CHECK-NEXT: [[REVERSE:%.*]] = shufflevector <2 x i32> [[WIDE_LOAD]], <2 x i32> poison, <2 x i32> <i32 1, i32 0>
+; CHECK-NEXT: [[TMP5:%.*]] = icmp eq <2 x i32> [[REVERSE]], splat (i32 3)
+; CHECK-NEXT: [[TMP6:%.*]] = xor <2 x i1> [[TMP5]], splat (i1 true)
+; CHECK-NEXT: [[TMP7:%.*]] = extractelement <2 x i1> [[TMP6]], i32 0
+; CHECK-NEXT: br i1 [[TMP7]], label [[PRED_STORE_IF:%.*]], label [[PRED_STORE_CONTINUE:%.*]]
+; CHECK: pred.store.if:
+; CHECK-NEXT: [[TMP8:%.*]] = extractelement <2 x i64> [[TMP0]], i32 0
+; CHECK-NEXT: [[TMP9:%.*]] = getelementptr inbounds [1024 x i32], ptr [[LOCAL_SRC]], i64 0, i64 [[TMP8]]
+; CHECK-NEXT: [[TMP10:%.*]] = load i32, ptr [[TMP9]], align 4
+; CHECK-NEXT: [[TMP11:%.*]] = extractelement <2 x i64> [[TMP0]], i32 0
+; CHECK-NEXT: [[TMP12:%.*]] = getelementptr inbounds [1024 x i32], ptr [[LOCAL_DEST]], i64 0, i64 [[TMP11]]
+; CHECK-NEXT: [[TMP13:%.*]] = shl nsw i32 [[TMP10]], 2
+; CHECK-NEXT: store i32 [[TMP13]], ptr [[TMP12]], align 4
+; CHECK-NEXT: br label [[PRED_STORE_CONTINUE]]
+; CHECK: pred.store.continue:
+; CHECK-NEXT: [[TMP14:%.*]] = extractelement <2 x i1> [[TMP6]], i32 1
+; CHECK-NEXT: br i1 [[TMP14]], label [[PRED_STORE_IF1:%.*]], label [[PRED_STORE_CONTINUE2]]
+; CHECK: pred.store.if1:
+; CHECK-NEXT: [[TMP15:%.*]] = extractelement <2 x i64> [[TMP0]], i32 1
+; CHECK-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1024 x i32], ptr [[LOCAL_SRC]], i64 0, i64 [[TMP15]]
+; CHECK-NEXT: [[TMP17:%.*]] = load i32, ptr [[TMP16]], align 4
+; CHECK-NEXT: [[TMP18:%.*]] = extractelement <2 x i64> [[TMP0]], i32 1
+; CHECK-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1024 x i32], ptr [[LOCAL_DEST]], i64 0, i64 [[TMP18]]
+; CHECK-NEXT: [[TMP20:%.*]] = shl nsw i32 [[TMP17]], 2
+; CHECK-NEXT: store i32 [[TMP20]], ptr [[TMP19]], align 4
+; CHECK-NEXT: br label [[PRED_STORE_CONTINUE2]]
+; CHECK: pred.store.continue2:
+; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2
+; CHECK-NEXT: [[VEC_IND_NEXT]] = add <2 x i64> [[VEC_IND]], splat (i64 -2)
+; CHECK-NEXT: [[TMP21:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1024
+; CHECK-NEXT: br i1 [[TMP21]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP12:![0-9]+]]
+; CHECK: middle.block:
+; CHECK-NEXT: br i1 true, label [[EXIT:%.*]], label [[SCALAR_PH]]
+; CHECK: scalar.ph:
+; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ -1, [[MIDDLE_BLOCK]] ], [ 1023, [[ENTRY:%.*]] ]
+; CHECK-NEXT: br label [[FOR_BODY:%.*]]
+; CHECK: for.body:
+; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[FOR_INC:%.*]] ]
+; CHECK-NEXT: [[OFF:%.*]] = add i64 [[IV]], -1
+; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [1024 x i32], ptr [[LOCAL_CMP]], i64 0, i64 [[OFF]]
+; CHECK-NEXT: [[TMP22:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
+; CHECK-NEXT: [[CMP3_NOT:%.*]] = icmp eq i32 [[TMP22]], 3
+; CHECK-NEXT: br i1 [[CMP3_NOT]], label [[FOR_INC]], label [[IF_THEN:%.*]]
+; CHECK: if.then:
+; CHECK-NEXT: [[ARRAYIDX5:%.*]] = getelementptr inbounds [1024 x i32], ptr [[LOCAL_SRC]], i64 0, i64 [[OFF]]
+; CHECK-NEXT: [[TMP23:%.*]] = load i32, ptr [[ARRAYIDX5]], align 4
+; CHECK-NEXT: [[MUL:%.*]] = shl nsw i32 [[TMP23]], 2
+; CHECK-NEXT: [[ARRAYIDX7:%.*]] = getelementptr inbounds [1024 x i32], ptr [[LOCAL_DEST]], i64 0, i64 [[OFF]]
+; CHECK-NEXT: store i32 [[MUL]], ptr [[ARRAYIDX7]], align 4
+; CHECK-NEXT: br label [[FOR_INC]]
+; CHECK: for.inc:
+; CHECK-NEXT: [[IV_NEXT]] = add nsw i64 [[IV]], -1
+; CHECK-NEXT: [[CMP2_NOT:%.*]] = icmp eq i64 [[IV]], 0
+; CHECK-NEXT: br i1 [[CMP2_NOT]], label [[EXIT]], label [[FOR_BODY]], !llvm.loop [[LOOP13:![0-9]+]]
+; CHECK: exit:
+; CHECK-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr [[DEST:%.*]], ptr [[LOCAL_DEST]], i64 1024, i1 false)
+; CHECK-NEXT: ret void
+;
+entry:
+ %local_dest = alloca [1024 x i32], align 4
+ %local_src = alloca [1024 x i32], align 4
+ %local_cmp = alloca [1024 x i32], align 4
+ call void @init(ptr %local_src)
+ call void @init(ptr %local_cmp)
+ br label %for.body
+
+for.body:
+ %iv = phi i64 [ 1023, %entry ], [ %iv.next, %for.inc ]
+ %off = add i64 %iv, -1
+ %arrayidx = getelementptr inbounds [1024 x i32], ptr %local_cmp, i64 0, i64 %off
+ %0 = load i32, ptr %arrayidx, align 4
+ %cmp3.not = icmp eq i32 %0, 3
+ br i1 %cmp3.not, label %for.inc, label %if.then
+
+if.then:
+ %arrayidx5 = getelementptr inbounds [1024 x i32], ptr %local_src, i64 0, i64 %off
+ %1 = load i32, ptr %arrayidx5, align 4
+ %mul = shl nsw i32 %1, 2
+ %arrayidx7 = getelementptr inbounds [1024 x i32], ptr %local_dest, i64 0, i64 %off
+ store i32 %mul, ptr %arrayidx7, align 4
+ br label %for.inc
+
+for.inc:
+ %iv.next = add nsw i64 %iv, -1
+ %cmp2.not = icmp eq i64 %iv, 0
+ br i1 %cmp2.not, label %exit, label %for.body
+
+exit:
+ call void @llvm.memcpy.p0.p0.i64(ptr %dest, ptr %local_dest, i64 1024, i1 false)
+ ret void
+}
+
+
+; Test a loop with a positive step recurrence that has a strided access
+define i16 @test_strided_access(i64 %len, ptr %test_base) {
+; CHECK-LABEL: @test_strided_access(
+; CHECK-NEXT: entry:
+; CHECK-NEXT: [[ALLOCA:%.*]] = alloca [163840 x i16], align 4
+; CHECK-NEXT: call void @init(ptr [[ALLOCA]])
+; CHECK-NEXT: br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
+; CHECK: vector.ph:
+; 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 <2 x i64> [ <i64 0, i64 1>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
+; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <2 x i16> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP13:%.*]], [[VECTOR_BODY]] ]
+; CHECK-NEXT: [[TMP0:%.*]] = add i64 [[INDEX]], 0
+; CHECK-NEXT: [[TMP1:%.*]] = getelementptr inbounds i8, ptr [[TEST_BASE:%.*]], i64 [[TMP0]]
+; CHECK-NEXT: [[TMP2:%.*]] = getelementptr inbounds i8, ptr [[TMP1]], i32 0
+; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <2 x i8>, ptr [[TMP2]], align 1
+; CHECK-NEXT: [[TMP3:%.*]] = icmp sge <2 x i8> [[WIDE_LOAD]], zeroinitializer
+; CHECK-NEXT: [[TMP4:%.*]] = mul <2 x i64> [[VEC_IND]], splat (i64 2)
+; CHECK-NEXT: [...
[truncated]
|
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
As part of this I've moved one of the negative tests from
load-deref-pred-align.ll into a new file
(load-deref-pred-neg-off.ll) because the pointer type had a
size of 16 bits and I realised it's probably not sensible for
allocas that are >16 bits in size!