[Sdcc-user] another pic16 offset related code gen bug

[Ben J. <be...@be...>]

Using:

$ bin/sdcc -v
SDCC : mcs51/gbz80/z80/avr/ds390/pic16/pic14/TININative/xa51/ds400/hc08 2.5.1 #1058 (Aug  6 2005) (UNIX)

int
shiftbug()
{
        static int foo;
        int bar;

        bar = foo >> 8;
        return bar;
}

The code correctly sees that a shift by 8 involves taking the top 8
bits and then sign extending, but it tests for the sign bit in the
lower word instead of the upper:

        BANKSEL (_shiftbug_foo_1_1 + 1)
;       .line   7; bug.c        bar = foo >> 8;
        MOVF    (_shiftbug_foo_1_1 + 1), W, B
        MOVWF   r0x00
        CLRF    r0x01
        BANKSEL _shiftbug_foo_1_1
        BTFSC   _shiftbug_foo_1_1, 7, B		<<< should operate on +1
        SETF    r0x01
;       .line   8; bug.c        return bar;

This is a bug in gen.c genrshTwo(), either the BTFSC should use MSB16
instead of LSB or (maybe better?) test the LSB of the new *result*
instead of the lhs.

-- 
Ben Jackson
<be...@be...>
http://www.ben.com/