On Fri, Jun 5, 2020 at 8:45 AM Jakub Jelinek <ja...@redhat.com> wrote:
>
> Hi!
>
> In January I've added patterns to optimize SImode -> DImode sign or zero
> extension of __builtin_popcount, this patch does the same for
> __builtin_c[lt]z. Like most other instructions, the [tl]zcntl instructions
> clear the upper 32 bits of the destination register and as the instructions
> only result in values 0 to 32 inclusive, both sign and zero extensions
> behave the same.
>
> Bootstrapped/regtested on x86_64-linux and i686-linux, ok for trunk?
>
> 2020-06-05 Jakub Jelinek <ja...@redhat.com>
>
> PR target/95535
> * config/i386/i386.md (*ctzsi2_zext, *clzsi2_lzcnt_zext): New
> define_insn_and_split patterns.
> (*ctzsi2_zext_falsedep, *clzsi2_lzcnt_zext_falsedep): New
> define_insn patterns.
>
> * gcc.target/i386/pr95535-1.c: New test.
> * gcc.target/i386/pr95535-2.c: New test.
OK.
Thanks,
Uros.
> --- gcc/config/i386/i386.md.jj 2020-05-25 10:06:59.882176002 +0200
> +++ gcc/config/i386/i386.md 2020-06-04 18:44:26.333963121 +0200
> @@ -13985,6 +13985,50 @@ (define_insn "*ctz<mode>2_falsedep"
> (set_attr "prefix_rep" "1")
> (set_attr "mode" "<MODE>")])
>
> +(define_insn_and_split "*ctzsi2_zext"
> + [(set (match_operand:DI 0 "register_operand" "=r")
> + (and:DI
> + (subreg:DI
> + (ctz:SI
> + (match_operand:SI 1 "nonimmediate_operand" "rm")) 0)
> + (const_int 63)))
> + (clobber (reg:CC FLAGS_REG))]
> + "TARGET_BMI && TARGET_64BIT"
> + "tzcnt{l}\t{%1, %k0|%k0, %1}"
> + "&& TARGET_AVOID_FALSE_DEP_FOR_BMI && epilogue_completed
> + && optimize_function_for_speed_p (cfun)
> + && !reg_mentioned_p (operands[0], operands[1])"
> + [(parallel
> + [(set (match_dup 0)
> + (and:DI (subreg:DI (ctz:SI (match_dup 1)) 0) (const_int 63)))
> + (unspec [(match_dup 0)] UNSPEC_INSN_FALSE_DEP)
> + (clobber (reg:CC FLAGS_REG))])]
> + "ix86_expand_clear (operands[0]);"
> + [(set_attr "type" "alu1")
> + (set_attr "prefix_0f" "1")
> + (set_attr "prefix_rep" "1")
> + (set_attr "mode" "SI")])
> +
> +; False dependency happens when destination is only updated by tzcnt,
> +; lzcnt or popcnt. There is no false dependency when destination is
> +; also used in source.
> +(define_insn "*ctzsi2_zext_falsedep"
> + [(set (match_operand:DI 0 "register_operand" "=r")
> + (and:DI
> + (subreg:DI
> + (ctz:SI
> + (match_operand:SI 1 "nonimmediate_operand" "rm")) 0)
> + (const_int 63)))
> + (unspec [(match_operand:DI 2 "register_operand" "0")]
> + UNSPEC_INSN_FALSE_DEP)
> + (clobber (reg:CC FLAGS_REG))]
> + "TARGET_BMI && TARGET_64BIT"
> + "tzcnt{l}\t{%1, %k0|%k0, %1}"
> + [(set_attr "type" "alu1")
> + (set_attr "prefix_0f" "1")
> + (set_attr "prefix_rep" "1")
> + (set_attr "mode" "SI")])
> +
> (define_insn "bsr_rex64"
> [(set (match_operand:DI 0 "register_operand" "=r")
> (minus:DI (const_int 63)
> @@ -14077,6 +14121,48 @@ (define_insn "*clz<mode>2_lzcnt_falsedep
> (set_attr "type" "bitmanip")
> (set_attr "mode" "<MODE>")])
>
> +(define_insn_and_split "*clzsi2_lzcnt_zext"
> + [(set (match_operand:DI 0 "register_operand" "=r")
> + (and:DI
> + (subreg:DI
> + (clz:SI
> + (match_operand:SI 1 "nonimmediate_operand" "rm")) 0)
> + (const_int 63)))
> + (clobber (reg:CC FLAGS_REG))]
> + "TARGET_LZCNT && TARGET_64BIT"
> + "lzcnt{l}\t{%1, %k0|%k0, %1}"
> + "&& TARGET_AVOID_FALSE_DEP_FOR_BMI && epilogue_completed
> + && optimize_function_for_speed_p (cfun)
> + && !reg_mentioned_p (operands[0], operands[1])"
> + [(parallel
> + [(set (match_dup 0)
> + (and:DI (subreg:DI (clz:SI (match_dup 1)) 0) (const_int 63)))
> + (unspec [(match_dup 0)] UNSPEC_INSN_FALSE_DEP)
> + (clobber (reg:CC FLAGS_REG))])]
> + "ix86_expand_clear (operands[0]);"
> + [(set_attr "prefix_rep" "1")
> + (set_attr "type" "bitmanip")
> + (set_attr "mode" "SI")])
> +
> +; False dependency happens when destination is only updated by tzcnt,
> +; lzcnt or popcnt. There is no false dependency when destination is
> +; also used in source.
> +(define_insn "*clzsi2_lzcnt_zext_falsedep"
> + [(set (match_operand:DI 0 "register_operand" "=r")
> + (and:DI
> + (subreg:DI
> + (clz:SI
> + (match_operand:SWI48 1 "nonimmediate_operand" "rm")) 0)
> + (const_int 63)))
> + (unspec [(match_operand:DI 2 "register_operand" "0")]
> + UNSPEC_INSN_FALSE_DEP)
> + (clobber (reg:CC FLAGS_REG))]
> + "TARGET_LZCNT"
> + "lzcnt{l}\t{%1, %k0|%k0, %1}"
> + [(set_attr "prefix_rep" "1")
> + (set_attr "type" "bitmanip")
> + (set_attr "mode" "SI")])
> +
> (define_int_iterator LT_ZCNT
> [(UNSPEC_TZCNT "TARGET_BMI")
> (UNSPEC_LZCNT "TARGET_LZCNT")])
> --- gcc/testsuite/gcc.target/i386/pr95535-1.c.jj 2020-06-04
> 18:47:07.642642830 +0200
> +++ gcc/testsuite/gcc.target/i386/pr95535-1.c 2020-06-04 18:47:34.011263537
> +0200
> @@ -0,0 +1,54 @@
> +/* PR target/95535 */
> +/* { dg-do compile { target lp64 } } */
> +/* { dg-options "-O2 -mbmi" } */
> +/* { dg-final { scan-assembler-not "cltq" } } */
> +
> +unsigned int foo (void);
> +
> +unsigned long
> +f1 (unsigned int x)
> +{
> + return __builtin_ctz (x);
> +}
> +
> +unsigned long
> +f2 (unsigned int x)
> +{
> + return (unsigned) __builtin_ctz (x);
> +}
> +
> +unsigned long
> +f3 (unsigned int x)
> +{
> + return __builtin_ctz (x) & 63ULL;
> +}
> +
> +unsigned long
> +f4 (unsigned int x)
> +{
> + return __builtin_ctz (x) & 1023ULL;
> +}
> +
> +unsigned long
> +f5 (void)
> +{
> + return __builtin_ctz (foo ());
> +}
> +
> +unsigned long
> +f6 (void)
> +{
> + return (unsigned) __builtin_ctz (foo ());
> +}
> +
> +unsigned long
> +f7 (void)
> +{
> + return __builtin_ctz (foo ()) & 63ULL;
> +}
> +
> +unsigned long
> +f8 (void)
> +{
> + return __builtin_ctz (foo ()) & 1023ULL;
> +}
> --- gcc/testsuite/gcc.target/i386/pr95535-2.c.jj 2020-06-04
> 18:47:10.774597782 +0200
> +++ gcc/testsuite/gcc.target/i386/pr95535-2.c 2020-06-04 18:47:50.576025269
> +0200
> @@ -0,0 +1,54 @@
> +/* PR target/95535 */
> +/* { dg-do compile { target lp64 } } */
> +/* { dg-options "-O2 -mlzcnt" } */
> +/* { dg-final { scan-assembler-not "cltq" } } */
> +
> +unsigned int foo (void);
> +
> +unsigned long
> +f1 (unsigned int x)
> +{
> + return __builtin_clz (x);
> +}
> +
> +unsigned long
> +f2 (unsigned int x)
> +{
> + return (unsigned) __builtin_clz (x);
> +}
> +
> +unsigned long
> +f3 (unsigned int x)
> +{
> + return __builtin_clz (x) & 63ULL;
> +}
> +
> +unsigned long
> +f4 (unsigned int x)
> +{
> + return __builtin_clz (x) & 1023ULL;
> +}
> +
> +unsigned long
> +f5 (void)
> +{
> + return __builtin_clz (foo ());
> +}
> +
> +unsigned long
> +f6 (void)
> +{
> + return (unsigned) __builtin_clz (foo ());
> +}
> +
> +unsigned long
> +f7 (void)
> +{
> + return __builtin_clz (foo ()) & 63ULL;
> +}
> +
> +unsigned long
> +f8 (void)
> +{
> + return __builtin_clz (foo ()) & 1023ULL;
> +}
>
> Jakub
>
