[Bug target/49263] SH Target: underutilized "TST #imm, R0" instruction

["klepikov.alex+bugs at gmail dot com" <gcc-bugzilla@gcc.gnu.org>]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=49263

Alexander Klepikov <klepikov.alex+bugs at gmail dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |klepikov.alex+bugs at gmail dot co
                   |                            |m

--- Comment #31 from Alexander Klepikov <klepikov.alex+bugs at gmail dot com> ---
I've found new cases for SH2 and SH2E CPUs only:

#define FLAG 0x40

unsigned int f(char v){
    return (v & FLAG) == FLAG;
}

For both big and little endian translates to dynamic shift call:

-O -m2 (or -m2e)

_f:
        sts.l   pr,@-r15
        mov.l   .L3,r1
        jsr     @r1
        exts.b  r4,r4
        mov     r4,r0
        and     #1,r0
        lds.l   @r15+,pr
        rts     
        nop
.L3:
        .long   ___ashiftrt_r4_6

And

#define FLAG 0x40
#define ADDR 0xFFFF0000
#define P ((unsigned char *)ADDR)

unsigned int f(void){
    return (*P & FLAG) == FLAG;
}

Translates to

_f:
        sts.l   pr,@-r15
        mov.l   .L3,r1
        mov.b   @r1,r4
        mov.l   .L4,r1
        jsr     @r1
        nop
        mov     r4,r0
        and     #1,r0
        lds.l   @r15+,pr
        rts     
        nop
.L3:
        .long   -65536
.L4:
        .long   ___ashiftrt_r4_6

Assembler output does not depend on ADDR value, but depends on variable (or
pointer) type. When type is integer, assembler code uses 'tst' for all options
'-m4', '-m2', '-m2e':

#define FLAG 0x40

unsigned int f(unsigned int v){
    return (v & FLAG) == FLAG;
}

translates to 

_f:
        mov     r4,r0
        tst     #64,r0
        mov     #-1,r0
        rts     
        negc    r0,r0

and

#define FLAG 0x40
#define ADDR 0xFFFF0000
#define P ((unsigned int *)ADDR)

unsigned int f(void){
    return (*P & FLAG) == FLAG;
}

translates to 

_f:
        mov.l   .L2,r1
        mov.l   @r1,r0
        tst     #64,r0
        mov     #-1,r0
        rts     
        negc    r0,r0
.L2:
        .long   -65536

Interesting that when '-m4' flag is specified, later GCC always translates to
code with 'tst' instruction. I played with godbolt and that's what I found. GCC
ver 4 uses dynamic shift 'shad' with '-m4' option and library call with '-m2'
or '-m2e' options. GCC 9.5 and later uses 'tst' with '-m4' and library call
with both '-m2' and '-m2e' when FLAG==0x40 and 'shll' instruction with both
'-m2' and '-m2e' when FLAG==0x80. I remind you that this is happening when char
type used only.

Maybe SH4 solution can be extended to support SH2/SH2E?