target/1947: gcc for target xscale generates incorrect code when optimization is turned on

[vmoorthy@giganet.com]

>Number:         1947
>Category:       target
>Synopsis:       gcc for target xscale generates incorrect code when optimization is turned on
>Confidential:   no
>Severity:       serious
>Priority:       medium
>Responsible:    unassigned
>State:          open
>Class:          wrong-code
>Submitter-Id:   net
>Arrival-Date:   Mon Feb 12 07:56:00 PST 2001
>Closed-Date:
>Last-Modified:
>Originator:     Vijay Moorthy
>Release:        2.9-xscale-001130
>Organization:
>Environment:
gcc for target==xscale-elf build=sun-sparc-solaris2.8
>Description:
Here a test program.  Note that testa and testb should behave differently.
In particular is x is big positive and y is big negative, the subtraction
will overflow resulting in a negative number, whereas the "<" should be 
false.  The compiler however generates the same code for each when optimizations
are turned on.

int testa(int x,int y) {
        if ( x-y < 0 )
                return 1;
        else
                return 0;
}

int testb(int x, int y) {
        if ( x < y )
                return 1;
        else
                return 2;
}

This problem seems to be specific to the xscale target, and 
does not occur for the "strongarm" target.
>How-To-Repeat:
Here is complete C source :

-----------------------------
int testa(int x,int y) {
        if ( x-y < 0 )
                return 1;
        else
                return 0;
}

int testb(int x, int y) {
        if ( x < y )
                return 1;
        else
                return 2;
}

---------------------------------

Compile with the following command line:

xscale-elf-gcc -S -O2 test.c

The following is the (incorrect) assembly output:

-----------------------------------
@ Generated by gcc 2.9-xscale-001130 for ARM/elf
	.file	"test.c"
.gcc2_compiled.:
.text
	.align	2
	.global	testa
	.type	 testa,function
testa:
	@ args = 0, pretend = 0, frame = 0
	@ frame_needed = 0, current_function_anonymous_args = 0
	cmp	r0, r1
	movge	r0, #0
	movlt	r0, #1
	mov	pc, lr
.Lfe1:
	.size	 testa,.Lfe1-testa
	.align	2
	.global	testb
	.type	 testb,function
testb:
	@ args = 0, pretend = 0, frame = 0
	@ frame_needed = 0, current_function_anonymous_args = 0
	cmp	r0, r1
	movge	r0, #2
	movlt	r0, #1
	mov	pc, lr
.Lfe2:
	.size	 testb,.Lfe2-testb
--------------------------------------------

Without optimizations, the following (correct) code is genertated:

---------------------------------------------
@ Generated by gcc 2.9-xscale-001130 for ARM/elf
	.file	"test.c"
.gcc2_compiled.:
.text
	.align	2
	.global	testa
	.type	 testa,function
testa:
	@ args = 0, pretend = 0, frame = 8
	@ frame_needed = 1, current_function_anonymous_args = 0
	mov	ip, sp
	stmfd	sp!, {fp, ip, lr, pc}
	sub	fp, ip, #4
	sub	sp, sp, #8
	str	r0, [fp, #-16]
	str	r1, [fp, #-20]
	ldr	r3, [fp, #-16]
	ldr	r2, [fp, #-20]
	rsb	r3, r2, r3
	cmp	r3, #0
	bge	.L3
	mov	r0, #1
	b	.L2
	b	.L4
.L3:
	mov	r0, #0
	b	.L2
.L4:
.L2:
	ldmea	fp, {fp, sp, pc}
.Lfe1:
	.size	 testa,.Lfe1-testa
	.align	2
	.global	testb
	.type	 testb,function
testb:
	@ args = 0, pretend = 0, frame = 8
	@ frame_needed = 1, current_function_anonymous_args = 0
	mov	ip, sp
	stmfd	sp!, {fp, ip, lr, pc}
	sub	fp, ip, #4
	sub	sp, sp, #8
	str	r0, [fp, #-16]
	str	r1, [fp, #-20]
	ldr	r3, [fp, #-16]
	ldr	r2, [fp, #-20]
	cmp	r3, r2
	bge	.L6
	mov	r0, #1
	b	.L5
	b	.L7
.L6:
	mov	r0, #2
	b	.L5
.L7:
.L5:
	ldmea	fp, {fp, sp, pc}
.Lfe2:
	.size	 testb,.Lfe2-testb
------------------------------------------------
>Fix:

>Release-Note:
>Audit-Trail:
>Unformatted: