This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
[PATCH, i386]: Implement lrint and llrint as x87 intrinsic function
- From: Uros Bizjak <uros at kss-loka dot si>
- To: gcc-patches at gcc dot gnu dot org
- Date: Mon, 21 Mar 2005 13:23:45 +0100
- Subject: [PATCH, i386]: Implement lrint and llrint as x87 intrinsic function
Hello!
This is follow-up patch to
http://gcc.gnu.org/ml/gcc-patches/2005-03/msg01970.html. Attached patch
implements call to lrint and llrint with x87 insn. The testcase:
long int test11(double x)
{
return lrint(x);
}
Will produce (-O2 -ffast-math):
test11:
pushl %ebp
movl %esp, %ebp
subl $4, %esp
fldl 8(%ebp)
fistpl -4(%ebp)
movl -4(%ebp), %eax
leave
ret
Patch was regtested and bootstrapped on i686-pc-linux-gnu, for c and
c++, together with
http://gcc.gnu.org/ml/gcc-patches/2005-03/msg01970.html and
http://gcc.gnu.org/ml/gcc-patches/2005-03/msg01523.html.
OK for mainline? (I would like to commit a fix for target PR/20421
before commiting this patch...)
2005-03-21 Uros Bizjak <uros@kss-loka.si>
* reg-stack.c (subst_stack_regs_pat): Handle <UNSPEC_FIST> case.
* config/i386/i386.c (output_fix_trunc): Add new round_mode
variable. Output "fldcw" depending on round_mode.
* config/i386/i386.md (UNSPEC_FIST): New.
(fistdi2, fistdi2_with_temp, fist<mode>2, fist<mode>2_with_temp):
New isns patterns to implement lrint and llrint built-ins as x87
intrinsic function.
(fistdi2, fist<mode>2 splitters): New splitters.
(lrintsi2, llrintdi2): New expanders.
BTW: It is very hard ATM for me to split i386.md to different patches
(because of mgs01523.html), so the patch includes only minimal diff to
i386.md. The patterns themselves are:
(define_insn "fistdi2"
[(set (match_operand:DI 0 "memory_operand" "=m")
(unspec:DI [(match_operand:XF 1 "register_operand" "f")]
UNSPEC_FIST))
(clobber (match_scratch:XF 2 "=&1f"))]
"TARGET_USE_FANCY_MATH_387
&& flag_unsafe_math_optimizations"
"* return output_fix_trunc (insn, operands, 0);"
[(set_attr "type" "fpspc")
(set_attr "mode" "DI")])
(define_insn "fistdi2_with_temp"
[(set (match_operand:DI 0 "nonimmediate_operand" "=m,?r")
(unspec:DI [(match_operand:XF 1 "register_operand" "f,f")]
UNSPEC_FIST))
(clobber (match_operand:DI 2 "memory_operand" "=m,m"))
(clobber (match_scratch:XF 3 "=&1f,&1f"))]
"TARGET_USE_FANCY_MATH_387
&& flag_unsafe_math_optimizations"
"#"
[(set_attr "type" "fpspc")
(set_attr "mode" "DI")])
(define_split
[(set (match_operand:DI 0 "register_operand" "")
(unspec:DI [(match_operand:XF 1 "register_operand" "")]
UNSPEC_FIST))
(clobber (match_operand:DI 2 "memory_operand" ""))
(clobber (match_scratch 3 ""))]
"reload_completed"
[(parallel [(set (match_dup 2) (unspec:DI [(match_dup 1)] UNSPEC_FIST))
(clobber (match_dup 3))])
(set (match_dup 0) (match_dup 2))]
"")
(define_split
[(set (match_operand:DI 0 "memory_operand" "")
(unspec:DI [(match_operand:XF 1 "register_operand" "")]
UNSPEC_FIST))
(clobber (match_operand:DI 2 "memory_operand" ""))
(clobber (match_scratch 3 ""))]
"reload_completed"
[(parallel [(set (match_dup 0) (unspec:DI [(match_dup 1)] UNSPEC_FIST))
(clobber (match_dup 3))])]
"")
(define_insn "fist<mode>2"
[(set (match_operand:X87MODEI12 0 "memory_operand" "=m")
(unspec:X87MODEI12 [(match_operand:XF 1 "register_operand" "f")]
UNSPEC_FIST))]
"TARGET_USE_FANCY_MATH_387
&& flag_unsafe_math_optimizations"
"* return output_fix_trunc (insn, operands, 0);"
[(set_attr "type" "fpspc")
(set_attr "mode" "<MODE>")])
(define_insn "fist<mode>2_with_temp"
[(set (match_operand:X87MODEI12 0 "nonimmediate_operand" "=m,?r")
(unspec:X87MODEI12 [(match_operand:XF 1 "register_operand" "f,f")]
UNSPEC_FIST))
(clobber (match_operand:X87MODEI12 2 "memory_operand" "=m,m"))]
"TARGET_USE_FANCY_MATH_387
&& flag_unsafe_math_optimizations"
"#"
[(set_attr "type" "fpspc")
(set_attr "mode" "<MODE>")])
(define_split
[(set (match_operand:X87MODEI12 0 "register_operand" "")
(unspec:X87MODEI12 [(match_operand:XF 1 "register_operand" "")]
UNSPEC_FIST))
(clobber (match_operand:X87MODEI12 2 "memory_operand" ""))]
"reload_completed"
[(set (match_dup 2) (unspec:X87MODEI12 [(match_dup 1)]
UNSPEC_FIST))
(set (match_dup 0) (match_dup 2))]
"")
(define_split
[(set (match_operand:X87MODEI12 0 "memory_operand" "")
(unspec:X87MODEI12 [(match_operand:XF 1 "register_operand" "")]
UNSPEC_FIST))
(clobber (match_scratch 2 ""))]
"reload_completed"
[(set (match_dup 0) (unspec:X87MODEI12 [(match_dup 1)]
UNSPEC_FIST))]
"")
(define_expand "lrintsi2"
[(use (match_operand:SI 0 "nonimmediate_operand" ""))
(use (match_operand:XF 1 "register_operand" ""))]
"TARGET_USE_FANCY_MATH_387
&& (!TARGET_SSE_MATH || TARGET_MIX_SSE_I387)
&& flag_unsafe_math_optimizations"
{
if (memory_operand (operands[0], VOIDmode))
emit_insn (gen_fistsi2 (operands[0], operands[1]));
else
{
operands[2] = assign_386_stack_local (SImode, 0);
emit_insn (gen_fistsi2_with_temp (operands[0], operands[1],
operands[2]));
}
DONE;
})
(define_expand "llrintdi2"
[(use (match_operand:DI 0 "nonimmediate_operand" ""))
(use (match_operand:XF 1 "register_operand" ""))]
"TARGET_USE_FANCY_MATH_387
&& (!TARGET_SSE_MATH || TARGET_MIX_SSE_I387)
&& flag_unsafe_math_optimizations"
{
if (memory_operand (operands[0], VOIDmode))
emit_insn (gen_fistdi2 (operands[0], operands[1]));
else
{
operands[2] = assign_386_stack_local (DImode, 0);
emit_insn (gen_fistdi2_with_temp (operands[0], operands[1],
operands[2]));
}
DONE;
})
Uros.
Index: reg-stack.c
===================================================================
RCS file: /cvs/gcc/gcc/gcc/reg-stack.c,v
retrieving revision 1.173
diff -u -p -r1.173 reg-stack.c
--- reg-stack.c 13 Mar 2005 14:06:47 -0000 1.173
+++ reg-stack.c 21 Mar 2005 10:49:20 -0000
@@ -1672,6 +1672,27 @@ subst_stack_regs_pat (rtx insn, stack re
case UNSPEC:
switch (XINT (pat_src, 1))
{
+ case UNSPEC_FIST:
+ /* These insns only operate on the top of the stack. */
+
+ src1 = get_true_reg (&XVECEXP (pat_src, 0, 0));
+ emit_swap_insn (insn, regstack, *src1);
+
+ src1_note = find_regno_note (insn, REG_DEAD, REGNO (*src1));
+
+ if (STACK_REG_P (*dest))
+ replace_reg (dest, FIRST_STACK_REG);
+
+ if (src1_note)
+ {
+ replace_reg (&XEXP (src1_note, 0), FIRST_STACK_REG);
+ regstack->top--;
+ CLEAR_HARD_REG_BIT (regstack->reg_set, REGNO (*src1));
+ }
+
+ replace_reg (src1, FIRST_STACK_REG);
+ break;
+
case UNSPEC_SIN:
case UNSPEC_COS:
case UNSPEC_FRNDINT:
Index: config/i386/i386.c
===================================================================
RCS file: /cvs/gcc/gcc/gcc/config/i386/i386.c,v
retrieving revision 1.798
diff -u -p -r1.798 i386.c
--- config/i386/i386.c 15 Mar 2005 14:44:08 -0000 1.798
+++ config/i386/i386.c 21 Mar 2005 10:49:23 -0000
@@ -7287,6 +7286,7 @@ output_fix_trunc (rtx insn, rtx *operand
{
int stack_top_dies = find_regno_note (insn, REG_DEAD, FIRST_STACK_REG) != 0;
int dimode_p = GET_MODE (operands[0]) == DImode;
+ int round_mode = get_attr_i387_cw (insn);
/* Jump through a hoop or two for DImode, since the hardware has no
non-popping instruction. We used to do this a different way, but
@@ -7304,12 +7304,14 @@ output_fix_trunc (rtx insn, rtx *operand
output_asm_insn ("fisttp%z0\t%0", operands);
else
{
- output_asm_insn ("fldcw\t%3", operands);
+ if (round_mode != I387_CW_ANY)
+ output_asm_insn ("fldcw\t%3", operands);
if (stack_top_dies || dimode_p)
output_asm_insn ("fistp%z0\t%0", operands);
else
output_asm_insn ("fist%z0\t%0", operands);
- output_asm_insn ("fldcw\t%2", operands);
+ if (round_mode != I387_CW_ANY)
+ output_asm_insn ("fldcw\t%2", operands);
}
return "";
Index: config/i386/i386.md
===================================================================
RCS file: /cvs/gcc/gcc/gcc/config/i386/i386.md,v
retrieving revision 1.620
diff -u -p -r1.620 i386.md
--- config/i386/i386.md 15 Mar 2005 14:44:09 -0000 1.620
+++ config/i386/i386.md 21 Mar 2005 10:49:26 -0000
@@ -119,7 +119,8 @@
(UNSPEC_FYL2X 66)
(UNSPEC_FYL2XP1 67)
(UNSPEC_FRNDINT 68)
- (UNSPEC_F2XM1 69)
+ (UNSPEC_FIST 69)
+ (UNSPEC_F2XM1 70)
; x87 Double output FP
(UNSPEC_SINCOS_COS 80)