On Tue, Nov 24, 2020 at 9:20 AM Jakub Jelinek <ja...@redhat.com> wrote:
>
> Hi!
>
> As the following testcase shows, unlike char, int or long long sized
> __builtin_*_overflow{,_p}, for short sized one in most cases the ce1 pass
> doesn't optimize the jo/jno or jc/jnc jumps with setting of a pseudo to 0/1
> into seto/setc. The reason is missing *setcc_hi_1* pattern. The following
> patch implements it using mode iterators so that on i486 and pentium?
> one can get the zero extension through and instead of movzbw.
>
> Bootstrapped/regtested on x86_64-linux and i686-linux, ok for trunk?
>
> 2020-11-24 Jakub Jelinek <ja...@redhat.com>
>
> PR target/97950
> * config/i386/i386.md (*setcc_si_1_and): Macroize into...
> (*setcc_<mode>_1_and): New define_insn_and_split with SWI24 iterator.
> (*setcc_si_1_movzbl): Macroize into...
> (*setcc_<mode>_1_movzbl): New define_insn_and_split with SWI24
> iterator.
>
> * gcc.target/i386/pr97950.c: New test.
OK.
Thanks,
Uros.
>
> --- gcc/config/i386/i386.md.jj 2020-11-23 17:01:48.235055044 +0100
> +++ gcc/config/i386/i386.md 2020-11-23 21:29:43.425842870 +0100
> @@ -12714,9 +12714,9 @@ (define_insn_and_split "*setcc_di_1"
> operands[2] = gen_lowpart (QImode, operands[0]);
> })
>
> -(define_insn_and_split "*setcc_si_1_and"
> - [(set (match_operand:SI 0 "register_operand" "=q")
> - (match_operator:SI 1 "ix86_comparison_operator"
> +(define_insn_and_split "*setcc_<mode>_1_and"
> + [(set (match_operand:SWI24 0 "register_operand" "=q")
> + (match_operator:SWI24 1 "ix86_comparison_operator"
> [(reg FLAGS_REG) (const_int 0)]))
> (clobber (reg:CC FLAGS_REG))]
> "!TARGET_PARTIAL_REG_STALL
> @@ -12724,7 +12724,7 @@ (define_insn_and_split "*setcc_si_1_and"
> "#"
> "&& reload_completed"
> [(set (match_dup 2) (match_dup 1))
> - (parallel [(set (match_dup 0) (zero_extend:SI (match_dup 2)))
> + (parallel [(set (match_dup 0) (zero_extend:SWI24 (match_dup 2)))
> (clobber (reg:CC FLAGS_REG))])]
> {
> operands[1] = shallow_copy_rtx (operands[1]);
> @@ -12732,16 +12732,16 @@ (define_insn_and_split "*setcc_si_1_and"
> operands[2] = gen_lowpart (QImode, operands[0]);
> })
>
> -(define_insn_and_split "*setcc_si_1_movzbl"
> - [(set (match_operand:SI 0 "register_operand" "=q")
> - (match_operator:SI 1 "ix86_comparison_operator"
> +(define_insn_and_split "*setcc_<mode>_1_movzbl"
> + [(set (match_operand:SWI24 0 "register_operand" "=q")
> + (match_operator:SWI24 1 "ix86_comparison_operator"
> [(reg FLAGS_REG) (const_int 0)]))]
> "!TARGET_PARTIAL_REG_STALL
> && (!TARGET_ZERO_EXTEND_WITH_AND || optimize_function_for_size_p (cfun))"
> "#"
> "&& reload_completed"
> [(set (match_dup 2) (match_dup 1))
> - (set (match_dup 0) (zero_extend:SI (match_dup 2)))]
> + (set (match_dup 0) (zero_extend:SWI24 (match_dup 2)))]
> {
> operands[1] = shallow_copy_rtx (operands[1]);
> PUT_MODE (operands[1], QImode);
> --- gcc/testsuite/gcc.target/i386/pr97950.c.jj 2020-11-23 17:20:33.481605139
> +0100
> +++ gcc/testsuite/gcc.target/i386/pr97950.c 2020-11-23 21:32:53.593734242
> +0100
> @@ -0,0 +1,153 @@
> +/* PR target/95950 */
> +/* { dg-do compile } */
> +/* { dg-options "-O2 -mtune=generic" } */
> +/* { dg-final { scan-assembler-times "\tseta\t" 4 } } */
> +/* { dg-final { scan-assembler-times "\tseto\t" 16 } } */
> +/* { dg-final { scan-assembler-times "\tsetc\t" 4 } } */
> +/* { dg-final { scan-assembler-not "\tjn?a\t" } } */
> +/* { dg-final { scan-assembler-not "\tjn?o\t" } } */
> +/* { dg-final { scan-assembler-not "\tjn?c\t" } } */
> +
> +char
> +f1 (short a, short b)
> +{
> + return __builtin_mul_overflow_p (a, b, (short) 0);
> +}
> +
> +char
> +f2 (short a, short b)
> +{
> + return __builtin_add_overflow_p (a, b, (short) 0);
> +}
> +
> +char
> +f3 (short a, short b)
> +{
> + return __builtin_sub_overflow_p (a, b, (short) 0);
> +}
> +
> +char
> +f4 (unsigned short a, unsigned short b)
> +{
> + return __builtin_mul_overflow_p (a, b, (unsigned short) 0);
> +}
> +
> +char
> +f5 (unsigned short a, unsigned short b)
> +{
> + return __builtin_add_overflow_p (a, b, (unsigned short) 0);
> +}
> +
> +char
> +f6 (unsigned short a, unsigned short b)
> +{
> + return __builtin_sub_overflow_p (a, b, (unsigned short) 0);
> +}
> +
> +char
> +f7 (short a, short b)
> +{
> + return __builtin_mul_overflow_p (a, b, (short) 0);
> +}
> +
> +char
> +f8 (short a, short b)
> +{
> + return __builtin_add_overflow_p (a, b, (short) 0);
> +}
> +
> +char
> +f9 (short a, short b)
> +{
> + return __builtin_sub_overflow_p (a, b, (short) 0);
> +}
> +
> +char
> +f10 (unsigned short a, unsigned short b)
> +{
> + return __builtin_mul_overflow_p (a, b, (unsigned short) 0);
> +}
> +
> +char
> +f11 (unsigned short a, unsigned short b)
> +{
> + return __builtin_add_overflow_p (a, b, (unsigned short) 0);
> +}
> +
> +char
> +f12 (unsigned short a, unsigned short b)
> +{
> + return __builtin_sub_overflow_p (a, b, (unsigned short) 0);
> +}
> +
> +unsigned short
> +f13 (short a, short b)
> +{
> + return __builtin_mul_overflow_p (a, b, (short) 0);
> +}
> +
> +unsigned short
> +f14 (short a, short b)
> +{
> + return __builtin_add_overflow_p (a, b, (short) 0);
> +}
> +
> +unsigned short
> +f15 (short a, short b)
> +{
> + return __builtin_sub_overflow_p (a, b, (short) 0);
> +}
> +
> +unsigned short
> +f16 (unsigned short a, unsigned short b)
> +{
> + return __builtin_mul_overflow_p (a, b, (unsigned short) 0);
> +}
> +
> +unsigned short
> +f17 (unsigned short a, unsigned short b)
> +{
> + return __builtin_add_overflow_p (a, b, (unsigned short) 0);
> +}
> +
> +unsigned short
> +f18 (unsigned short a, unsigned short b)
> +{
> + return __builtin_sub_overflow_p (a, b, (unsigned short) 0);
> +}
> +
> +unsigned short
> +f19 (short a, short b)
> +{
> + return __builtin_mul_overflow_p (a, b, (short) 0);
> +}
> +
> +unsigned short
> +f20 (short a, short b)
> +{
> + return __builtin_add_overflow_p (a, b, (short) 0);
> +}
> +
> +unsigned short
> +f21 (short a, short b)
> +{
> + return __builtin_sub_overflow_p (a, b, (short) 0);
> +}
> +
> +unsigned short
> +f22 (unsigned short a, unsigned short b)
> +{
> + return __builtin_mul_overflow_p (a, b, (unsigned short) 0);
> +}
> +
> +unsigned short
> +f23 (unsigned short a, unsigned short b)
> +{
> + return __builtin_add_overflow_p (a, b, (unsigned short) 0);
> +}
> +
> +unsigned short
> +f24 (unsigned short a, unsigned short b)
> +{
> + return __builtin_sub_overflow_p (a, b, (unsigned short) 0);
> +}
>
> Jakub
>
