[llvm-commits] CVS: llvm/test/Regression/LLC/2003-07-06-BadIntCmp.ll
Vikram Adve
vadve at cs.uiuc.edu
Sun Jul 6 15:13:01 PDT 2003
Changes in directory llvm/test/Regression/LLC:
2003-07-06-BadIntCmp.ll added (r1.1)
---
Log message:
Regression test for bug in reg. allocation that was using %xcc instead
of %icc even for 32-bit and smaller comparisons.
---
Diffs of the changes:
Index: llvm/test/Regression/LLC/2003-07-06-BadIntCmp.ll
diff -c /dev/null llvm/test/Regression/LLC/2003-07-06-BadIntCmp.ll:1.1
*** /dev/null Sun Jul 6 15:12:28 2003
--- llvm/test/Regression/LLC/2003-07-06-BadIntCmp.ll Sun Jul 6 15:12:18 2003
***************
*** 0 ****
--- 1,77 ----
+ ;; Date: May 28, 2003.
+ ;; From: test/Programs/MultiSource/Olden-perimeter/maketree.c
+ ;; Function: int CheckOutside(int x, int y)
+ ;;
+ ;; Note: The .ll code below for this regression test has identical
+ ;; behavior to the above function up to the error, but then prints
+ ;; true/false on the two branches.
+ ;;
+ ;; Error: llc generates a branch-on-xcc instead of branch-on-icc, which
+ ;; is wrong because the value being compared (int euclid = x*x + y*y)
+ ;; overflows, so that the 64-bit and 32-bit compares are not equal.
+ ;;
+ ;; LLC Output:
+ ;;
+ ;; !****** Outputing Function: CheckOutside_34 ******
+ ;; .align 4
+ ;; .global CheckOutside_34
+ ;; .type CheckOutside_34, 2
+ ;; CheckOutside_34:
+ ;; .L_CheckOutside_34_LL_0:
+ ;; save %o6, -192, %o6
+ ;; mulx %i1, %i1, %i1
+ ;; mulx %i0, %i0, %i0
+ ;; sethi %lm(4194304), %o1
+ ;; or %g0, 1, %o0
+ ;; add %i0, %i1, %i0
+ ;; sra %o1, 0, %o1
+ ;; subcc %i0, %o1, %g0
+ ;; bg %xcc, .L_CheckOutside_34_LL_3
+ ;; nop
+ ;; ba .L_CheckOutside_34_LL_1
+ ;; nop
+ ;; ...
+ ;;
+ ;;
+
+
+ target endian = little
+ target pointersize = 32
+ %.str_1 = internal constant [6 x sbyte] c"true\0A\00" ; <[6 x sbyte]*> [#uses=1]
+ %.str_2 = internal constant [7 x sbyte] c"false\0A\00" ; <[7 x sbyte]*> [#uses=1]
+
+ implementation ; Functions:
+
+ declare int %printf(sbyte*, ...)
+
+ internal void %__main() {
+ entry: ; No predecessors!
+ ret void
+ }
+
+ internal void %CheckOutside(int %x.1, int %y.1) {
+ entry: ; No predecessors!
+ %tmp.2 = mul int %x.1, %x.1 ; <int> [#uses=1]
+ %tmp.5 = mul int %y.1, %y.1 ; <int> [#uses=1]
+ %tmp.6 = add int %tmp.2, %tmp.5 ; <int> [#uses=1]
+ %tmp.8 = setle int %tmp.6, 4194304 ; <bool> [#uses=1]
+ br bool %tmp.8, label %then, label %else
+
+ then: ; preds = %entry
+ %tmp.11 = call int (sbyte*, ...)* %printf( sbyte* getelementptr ([6 x sbyte]* %.str_1, long 0, long 0) ) ; <int> [#uses=0]
+ br label %UnifiedExitNode
+
+ else: ; preds = %entry
+ %tmp.13 = call int (sbyte*, ...)* %printf( sbyte* getelementptr ([7 x sbyte]* %.str_2, long 0, long 0) ) ; <int> [#uses=0]
+ br label %UnifiedExitNode
+
+ UnifiedExitNode: ; preds = %then, %else
+ ret void
+ }
+
+ int %main() {
+ entry: ; No predecessors!
+ call void %__main( )
+ call void %CheckOutside( int 2097152, int 2097152 )
+ ret int 0
+ }
More information about the llvm-commits
mailing list