[llvm] r272360 - [LiveRangeEdit] Add a test case for r272314.
Quentin Colombet via llvm-commits
llvm-commits at lists.llvm.org
Thu Jun 9 18:57:48 PDT 2016
Author: qcolombet
Date: Thu Jun 9 20:57:48 2016
New Revision: 272360
URL: http://llvm.org/viewvc/llvm-project?rev=272360&view=rev
Log:
[LiveRangeEdit] Add a test case for r272314.
The test case is not great espicially because it is still cumbersome to
run the regalloc pass with run-pass. (We miss a bunch of initiliazier to
be properly implemented.)
Related to llvm.org/PR27983
Added:
llvm/trunk/test/CodeGen/X86/crash-lre-eliminate-dead-def.ll
Added: llvm/trunk/test/CodeGen/X86/crash-lre-eliminate-dead-def.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/crash-lre-eliminate-dead-def.ll?rev=272360&view=auto
==============================================================================
--- llvm/trunk/test/CodeGen/X86/crash-lre-eliminate-dead-def.ll (added)
+++ llvm/trunk/test/CodeGen/X86/crash-lre-eliminate-dead-def.ll Thu Jun 9 20:57:48 2016
@@ -0,0 +1,268 @@
+; RUN: llc %s -o - | FileCheck %s
+; This file checks some weird corner case in LiveRangeEdit.
+; We used to do crash when we eliminate the definition
+; of the product of splitting when the original live-range
+; has already been removed.
+; Basically, we have the following input.
+; v1 = loadimm cst
+; ...
+; = use v1
+;
+; We split the live-range like this:
+; v1 = loadimm cst
+; ...
+; v2 = copy v1
+; ...
+; = use v2
+;
+; We actually issue loadimm instead of the copy:
+; v1 = loadimm cst
+; ...
+; v2 = loadimm cst
+; ...
+; = use v2
+;
+; v1 is now dead so we remove its live-range.
+; Actually, we shrink it to empty to keep the
+; instruction around for futher remat opportunities
+; (accessbile via the origin pointer.)
+;
+; Later v2 gets remove as well (e.g., because we
+; remat it closer to its use) and the live-range
+; gets eliminated. We used to crash at this point
+; because we were looking for a VNI of origin (v1)
+; at the slot index of the definition of v2. However,
+; we do not have a VNI for v1 at this point, since the
+; live-range is now empty... crash!
+; PR27983
+
+source_filename = "bugpoint-output-1e29d28.bc"
+target datalayout = "e-m:e-p:32:32-f64:32:64-f80:32-n8:16:32-S128"
+target triple = "i386-unknown-linux-gnu"
+
+ at r = external global i32, align 4
+ at k = external global i32, align 4
+ at g = external global i32, align 4
+ at a = external global i16, align 2
+ at p = external global i32, align 4
+ at n = external global i16, align 2
+ at .str = external unnamed_addr constant [12 x i8], align 1
+ at .str.1 = external unnamed_addr constant [13 x i8], align 1
+ at s = external global i32, align 4
+ at z = external global i16, align 2
+
+; CHECK-LABEL: fn1:
+define void @fn1() #0 {
+entry:
+ br label %for.cond
+
+for.cond: ; preds = %cleanup100, %for.end29, %entry
+ %t7.0 = phi i16 [ undef, %entry ], [ %t7.1, %for.end29 ], [ %t7.19, %cleanup100 ]
+ %t9.0 = phi i32 [ undef, %entry ], [ %t9.1, %for.end29 ], [ 0, %cleanup100 ]
+ %t2.0 = phi i32 [ undef, %entry ], [ undef, %for.end29 ], [ %t2.18, %cleanup100 ]
+ %tmp = load i32, i32* @r, align 4
+ br i1 undef, label %if.then, label %if.end7
+
+if.then: ; preds = %for.cond
+ %tobool = icmp ne i32 %tmp, 0
+ %tobool1 = icmp ne i32 %t2.0, 0
+ %tmp1 = and i1 %tobool1, %tobool
+ %land.ext = zext i1 %tmp1 to i32
+ %tmp2 = load i32, i32* @k, align 4
+ %shr = lshr i32 %land.ext, %tmp2
+ %tobool4 = icmp ne i32 %shr, 0
+ %or.cond = and i1 false, %tobool4
+ br i1 %or.cond, label %L6, label %if.end7
+
+if.end7: ; preds = %if.then, %for.cond
+ %t2.1 = phi i32 [ %shr, %if.then ], [ %t2.0, %for.cond ]
+ %tobool8 = icmp eq i32 undef, 0
+ br i1 %tobool8, label %if.end11, label %for.cond10
+
+for.cond10: ; preds = %for.cond10, %if.end7
+ br label %for.cond10
+
+if.end11: ; preds = %if.end7
+ %tmp3 = load i32, i32* @g, align 4
+ %tmp4 = load i16, i16* @a, align 2
+ %conv = sext i16 %tmp4 to i32
+ %div = sdiv i32 %tmp3, %conv
+ %tobool12 = icmp eq i32 %div, 0
+ br i1 %tobool12, label %for.cond15, label %L5
+
+for.cond15: ; preds = %for.cond17, %if.end11
+ %t7.1 = phi i16 [ %t7.2, %for.cond17 ], [ %t7.0, %if.end11 ]
+ %t9.1 = phi i32 [ %t9.2, %for.cond17 ], [ %t9.0, %if.end11 ]
+ %tobool16 = icmp eq i32 undef, 0
+ br i1 %tobool16, label %for.end29, label %for.cond17
+
+for.cond17: ; preds = %for.cond20, %for.cond15
+ %t7.2 = phi i16 [ %t7.3, %for.cond20 ], [ %t7.1, %for.cond15 ]
+ %t9.2 = phi i32 [ undef, %for.cond20 ], [ %t9.1, %for.cond15 ]
+ %tobool18 = icmp eq i8 undef, 0
+ br i1 %tobool18, label %for.cond15, label %for.cond20
+
+for.cond20: ; preds = %for.cond23, %for.cond17
+ %t7.3 = phi i16 [ %t7.4, %for.cond23 ], [ %t7.2, %for.cond17 ]
+ %tobool21 = icmp eq i32 undef, 0
+ br i1 %tobool21, label %for.cond17, label %for.cond23
+
+for.cond23: ; preds = %L1, %for.cond20
+ %t7.4 = phi i16 [ %t7.5, %L1 ], [ %t7.3, %for.cond20 ]
+ %tobool24 = icmp eq i8 undef, 0
+ br i1 %tobool24, label %for.cond20, label %L1
+
+L1: ; preds = %cleanup100, %for.cond23
+ %t7.5 = phi i16 [ %t7.19, %cleanup100 ], [ %t7.4, %for.cond23 ]
+ %conv26 = sext i16 undef to i64
+ br label %for.cond23
+
+for.end29: ; preds = %for.cond15
+ br i1 undef, label %for.cond, label %for.cond32thread-pre-split
+
+for.cond32thread-pre-split: ; preds = %for.end29
+ %.pr = load i32, i32* @p, align 4
+ br label %for.cond32
+
+for.cond32: ; preds = %for.inc94, %for.cond32thread-pre-split
+ %t7.6 = phi i16 [ %t7.1, %for.cond32thread-pre-split ], [ %t7.17, %for.inc94 ]
+ %t3.4 = phi i64 [ 0, %for.cond32thread-pre-split ], [ 0, %for.inc94 ]
+ %t9.6 = phi i32 [ %t9.1, %for.cond32thread-pre-split ], [ 0, %for.inc94 ]
+ %t2.7 = phi i32 [ undef, %for.cond32thread-pre-split ], [ %t2.16, %for.inc94 ]
+ %tobool33 = icmp eq i32 0, 0
+ br i1 %tobool33, label %for.end95, label %for.body34
+
+for.body34: ; preds = %for.cond32
+ %tobool35 = icmp eq i16 undef, 0
+ br i1 %tobool35, label %for.inc94, label %if.then36
+
+if.then36: ; preds = %for.body34
+ %tmp5 = load i16, i16* @n, align 2
+ %tobool37 = icmp eq i32 undef, 0
+ br i1 %tobool37, label %if.end78, label %if.then38
+
+if.then38: ; preds = %if.then36
+ tail call void (i8*, ...) @printf(i8* getelementptr inbounds ([12 x i8], [12 x i8]* @.str, i32 0, i32 0), i64 undef)
+ %tobool40 = icmp eq i32 undef, 0
+ br i1 %tobool40, label %L3, label %cleanup100
+
+L3: ; preds = %while.end.split, %if.then38
+ %t7.7 = phi i16 [ %tmp5, %if.then38 ], [ %t7.15, %while.end.split ]
+ %t3.5 = phi i64 [ %t3.4, %if.then38 ], [ %t3.11, %while.end.split ]
+ %t2.8 = phi i32 [ %t2.7, %if.then38 ], [ %t2.14, %while.end.split ]
+ %tobool43 = icmp eq i32 undef, 0
+ br i1 %tobool43, label %if.end48, label %cleanup75
+
+if.end48: ; preds = %L3
+ tail call void (i8*, ...) @printf(i8* getelementptr inbounds ([13 x i8], [13 x i8]* @.str.1, i32 0, i32 0), i64 %t3.5)
+ br i1 undef, label %if.end61, label %for.cond52.preheader
+
+for.cond52.preheader: ; preds = %if.end48
+ %tobool57 = icmp eq i16 undef, 0
+ %.130 = select i1 %tobool57, i16 -8, i16 0
+ br label %if.end61
+
+if.end61: ; preds = %for.cond52.preheader, %if.end48
+ %t7.9 = phi i16 [ %t7.7, %if.end48 ], [ %.130, %for.cond52.preheader ]
+ %tobool62 = icmp eq i32 undef, 0
+ br i1 %tobool62, label %if.end71, label %if.then63
+
+if.then63: ; preds = %if.end61
+ br i1 undef, label %if.end67, label %L5
+
+L5: ; preds = %cleanup100.L5_crit_edge, %if.then63, %if.end11
+ %.pre = phi i32 [ %.pre.pre, %cleanup100.L5_crit_edge ], [ undef, %if.then63 ], [ %tmp, %if.end11 ]
+ %t7.10 = phi i16 [ %t7.19, %cleanup100.L5_crit_edge ], [ %t7.9, %if.then63 ], [ %t7.0, %if.end11 ]
+ %t3.6 = phi i64 [ 0, %cleanup100.L5_crit_edge ], [ %t3.5, %if.then63 ], [ 2, %if.end11 ]
+ %t9.8 = phi i32 [ 0, %cleanup100.L5_crit_edge ], [ undef, %if.then63 ], [ %t9.0, %if.end11 ]
+ %t2.9 = phi i32 [ %t2.18, %cleanup100.L5_crit_edge ], [ %t2.8, %if.then63 ], [ %t2.1, %if.end11 ]
+ store i32 %t9.8, i32* @s, align 4
+ br label %if.end67
+
+if.end67: ; preds = %L5, %if.then63
+ %tmp6 = phi i32 [ %.pre, %L5 ], [ undef, %if.then63 ]
+ %t7.11 = phi i16 [ %t7.10, %L5 ], [ %t7.9, %if.then63 ]
+ %t3.7 = phi i64 [ %t3.6, %L5 ], [ %t3.5, %if.then63 ]
+ %t9.9 = phi i32 [ %t9.8, %L5 ], [ undef, %if.then63 ]
+ %t2.10 = phi i32 [ %t2.9, %L5 ], [ %t2.8, %if.then63 ]
+ %tobool68 = icmp eq i32 %tmp6, 0
+ br i1 %tobool68, label %if.end71, label %for.end95
+
+if.end71: ; preds = %if.end67, %if.end61
+ %t7.12 = phi i16 [ %t7.11, %if.end67 ], [ %t7.9, %if.end61 ]
+ %t3.8 = phi i64 [ %t3.7, %if.end67 ], [ %t3.5, %if.end61 ]
+ %tobool72 = icmp eq i32 undef, 0
+ br i1 %tobool72, label %cleanup75.thread128, label %if.then73
+
+if.then73: ; preds = %if.end71
+ br label %cleanup100
+
+cleanup75.thread128: ; preds = %if.end71
+ br label %if.end78
+
+cleanup75: ; preds = %L3
+ br i1 false, label %for.cond98, label %for.end95
+
+if.end78: ; preds = %cleanup75.thread128, %if.then36
+ %t7.14 = phi i16 [ %tmp5, %if.then36 ], [ 0, %cleanup75.thread128 ]
+ %t3.10 = phi i64 [ %t3.4, %if.then36 ], [ %t3.8, %cleanup75.thread128 ]
+ %t9.12 = phi i32 [ %t9.6, %if.then36 ], [ undef, %cleanup75.thread128 ]
+ %t2.13 = phi i32 [ %t2.7, %if.then36 ], [ undef, %cleanup75.thread128 ]
+ store i16 %t7.14, i16* @z, align 2
+ br label %L6
+
+L6: ; preds = %if.end78, %if.then
+ %t7.15 = phi i16 [ %t7.0, %if.then ], [ %t7.14, %if.end78 ]
+ %t3.11 = phi i64 [ 2, %if.then ], [ %t3.10, %if.end78 ]
+ %t9.13 = phi i32 [ %t9.0, %if.then ], [ %t9.12, %if.end78 ]
+ %t2.14 = phi i32 [ %shr, %if.then ], [ %t2.13, %if.end78 ]
+ br i1 undef, label %while.condthread-pre-split, label %for.inc94
+
+while.condthread-pre-split: ; preds = %L6
+ %tobool83 = icmp eq i32 undef, 0
+ br i1 %tobool83, label %while.end.split, label %while.cond
+
+while.cond: ; preds = %while.cond, %while.condthread-pre-split
+ br label %while.cond
+
+while.end.split: ; preds = %while.condthread-pre-split
+ %tobool84 = icmp eq i16 undef, 0
+ br i1 %tobool84, label %for.inc94, label %L3
+
+for.inc94: ; preds = %while.end.split, %L6, %for.body34
+ %t7.17 = phi i16 [ %t7.6, %for.body34 ], [ %t7.15, %L6 ], [ %t7.15, %while.end.split ]
+ %t2.16 = phi i32 [ %t2.7, %for.body34 ], [ %t2.14, %L6 ], [ %t2.14, %while.end.split ]
+ store i32 undef, i32* @p, align 4
+ br label %for.cond32
+
+for.end95: ; preds = %cleanup75, %if.end67, %for.cond32
+ %t7.18 = phi i16 [ %t7.6, %for.cond32 ], [ %t7.7, %cleanup75 ], [ %t7.11, %if.end67 ]
+ %t2.17 = phi i32 [ %t2.7, %for.cond32 ], [ %t2.8, %cleanup75 ], [ %t2.10, %if.end67 ]
+ %tobool96 = icmp eq i32 undef, 0
+ br i1 %tobool96, label %cleanup100, label %for.cond98
+
+for.cond98: ; preds = %for.cond98, %for.end95, %cleanup75
+ br label %for.cond98
+
+cleanup100: ; preds = %for.end95, %if.then73, %if.then38
+ %t7.19 = phi i16 [ %t7.18, %for.end95 ], [ %tmp5, %if.then38 ], [ %t7.12, %if.then73 ]
+ %t2.18 = phi i32 [ %t2.17, %for.end95 ], [ %t2.7, %if.then38 ], [ undef, %if.then73 ]
+ switch i32 undef, label %unreachable [
+ i32 0, label %for.cond
+ i32 17, label %L1
+ i32 7, label %cleanup100.L5_crit_edge
+ ]
+
+cleanup100.L5_crit_edge: ; preds = %cleanup100
+ %.pre.pre = load i32, i32* @r, align 4
+ br label %L5
+
+unreachable: ; preds = %cleanup100
+ unreachable
+}
+
+; Function Attrs: nounwind
+declare void @printf(i8* nocapture readonly, ...) #1
+
+attributes #0 = { noreturn nounwind "disable-tail-calls"="false" "less-precise-fpmad"="false" "no-frame-pointer-elim"="false" "no-infs-fp-math"="false" "no-jump-tables"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "target-cpu"="pentium4" "target-features"="+fxsr,+mmx,+sse,+sse2,+x87" "unsafe-fp-math"="false" "use-soft-float"="false" }
+attributes #1 = { nounwind "disable-tail-calls"="false" "less-precise-fpmad"="false" "no-frame-pointer-elim"="false" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "target-cpu"="pentium4" "target-features"="+fxsr,+mmx,+sse,+sse2,+x87" "unsafe-fp-math"="false" "use-soft-float"="false" }
More information about the llvm-commits
mailing list