* [x86, PATCH] More effecient code for short unsigned conversion to float-point.
@ 2013-08-06 14:34 Yuri Rumyantsev
2013-08-06 14:47 ` Andreas Schwab
2013-08-06 21:03 ` Richard Henderson
0 siblings, 2 replies; 3+ messages in thread
From: Yuri Rumyantsev @ 2013-08-06 14:34 UTC (permalink / raw)
To: gcc-patches, Uros Bizjak, Igor Zamyatin, Areg Melik-Adamyan
[-- Attachment #1: Type: text/plain, Size: 448 bytes --]
Hi All,
Here is simple fix which produces more optimal code unsigned
char(short) to float(double) with direct rtl-generation.
Bootstrapping and regression testing were successful for x86-64.
Is it OK for trunk?
ChangeLog:
2013-08-06 Yuri Rumyantsev <ysrumyan@gmail.com>
* config/i386/i386.md (floatunssi<mode>2 expand): Add support
for QI/HImode operand to produce more effecient code for
unsigned char(short) --> float(double) conversion.
[-- Attachment #2: i386-short-uns-conversion-to-fp.patch --]
[-- Type: application/octet-stream, Size: 1612 bytes --]
Index: config/i386/i386.md
===================================================================
--- config/i386/i386.md (revision 201475)
+++ config/i386/i386.md (working copy)
@@ -5089,25 +5089,32 @@
operands[3] = simplify_gen_subreg (DImode, operands[3], SImode, 0);
})
-(define_expand "floatunssi<mode>2"
+(define_expand "floatuns<SWI124:mode><X87MODEF:mode>2"
[(parallel
[(set (match_operand:X87MODEF 0 "register_operand")
(unsigned_float:X87MODEF
- (match_operand:SI 1 "nonimmediate_operand")))
+ (match_operand:SWI124 1 "nonimmediate_operand")))
(clobber (match_dup 2))
(clobber (match_scratch:SI 3))])]
"!TARGET_64BIT
&& ((TARGET_80387 && X87_ENABLE_FLOAT (<X87MODEF:MODE>mode, DImode)
&& TARGET_SSE)
- || (SSE_FLOAT_MODE_P (<MODE>mode) && TARGET_SSE_MATH))"
+ || (SSE_FLOAT_MODE_P (<X87MODEF:MODE>mode) && TARGET_SSE_MATH))"
{
- if (SSE_FLOAT_MODE_P (<MODE>mode) && TARGET_SSE_MATH)
+ if (GET_MODE_SIZE (<SWI124:MODE>mode) < GET_MODE_SIZE (SImode))
+ {
+ operands[1] = convert_to_mode (SImode, operands[1], 1);
+ emit_insn (gen_floatsi<X87MODEF:mode>2 (operands[0], operands[1]));
+ DONE;
+ }
+
+ if (SSE_FLOAT_MODE_P (<X87MODEF:MODE>mode) && TARGET_SSE_MATH)
{
- ix86_expand_convert_uns_si<mode>_sse (operands[0], operands[1]);
+ ix86_expand_convert_uns_si<X87MODEF:mode>_sse (operands[0], operands[1]);
DONE;
}
- else
- operands[2] = assign_386_stack_local (DImode, SLOT_TEMP);
+
+ operands[2] = assign_386_stack_local (DImode, SLOT_TEMP);
})
(define_expand "floatunsdisf2"
^ permalink raw reply [flat|nested] 3+ messages in thread
* Re: [x86, PATCH] More effecient code for short unsigned conversion to float-point.
2013-08-06 14:34 [x86, PATCH] More effecient code for short unsigned conversion to float-point Yuri Rumyantsev
@ 2013-08-06 14:47 ` Andreas Schwab
2013-08-06 21:03 ` Richard Henderson
1 sibling, 0 replies; 3+ messages in thread
From: Andreas Schwab @ 2013-08-06 14:47 UTC (permalink / raw)
To: Yuri Rumyantsev
Cc: gcc-patches, Uros Bizjak, Igor Zamyatin, Areg Melik-Adamyan
Yuri Rumyantsev <ysrumyan@gmail.com> writes:
> * config/i386/i386.md (floatunssi<mode>2 expand): Add support
> for QI/HImode operand to produce more effecient code for
s/effecient/efficient/
Andreas.
--
Andreas Schwab, SUSE Labs, schwab@suse.de
GPG Key fingerprint = 0196 BAD8 1CE9 1970 F4BE 1748 E4D4 88E3 0EEA B9D7
"And now for something completely different."
^ permalink raw reply [flat|nested] 3+ messages in thread
* Re: [x86, PATCH] More effecient code for short unsigned conversion to float-point.
2013-08-06 14:34 [x86, PATCH] More effecient code for short unsigned conversion to float-point Yuri Rumyantsev
2013-08-06 14:47 ` Andreas Schwab
@ 2013-08-06 21:03 ` Richard Henderson
1 sibling, 0 replies; 3+ messages in thread
From: Richard Henderson @ 2013-08-06 21:03 UTC (permalink / raw)
To: Yuri Rumyantsev
Cc: gcc-patches, Uros Bizjak, Igor Zamyatin, Areg Melik-Adamyan
On 08/06/2013 04:33 AM, Yuri Rumyantsev wrote:
> -(define_expand "floatunssi<mode>2"
> +(define_expand "floatuns<SWI124:mode><X87MODEF:mode>2"
> [(parallel
> [(set (match_operand:X87MODEF 0 "register_operand")
> (unsigned_float:X87MODEF
> - (match_operand:SI 1 "nonimmediate_operand")))
> + (match_operand:SWI124 1 "nonimmediate_operand")))
> (clobber (match_dup 2))
> (clobber (match_scratch:SI 3))])]
> "!TARGET_64BIT
> && ((TARGET_80387 && X87_ENABLE_FLOAT (<X87MODEF:MODE>mode, DImode)
> && TARGET_SSE)
> - || (SSE_FLOAT_MODE_P (<MODE>mode) && TARGET_SSE_MATH))"
> + || (SSE_FLOAT_MODE_P (<X87MODEF:MODE>mode) && TARGET_SSE_MATH))"
> {
> - if (SSE_FLOAT_MODE_P (<MODE>mode) && TARGET_SSE_MATH)
> + if (GET_MODE_SIZE (<SWI124:MODE>mode) < GET_MODE_SIZE (SImode))
> + {
> + operands[1] = convert_to_mode (SImode, operands[1], 1);
> + emit_insn (gen_floatsi<X87MODEF:mode>2 (operands[0], operands[1]));
> + DONE;
> + }
> +
The idea is correct, but there's no need to share the same expand,
adding a runtime test against the mode.
r~
^ permalink raw reply [flat|nested] 3+ messages in thread
end of thread, other threads:[~2013-08-06 21:03 UTC | newest]
Thread overview: 3+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2013-08-06 14:34 [x86, PATCH] More effecient code for short unsigned conversion to float-point Yuri Rumyantsev
2013-08-06 14:47 ` Andreas Schwab
2013-08-06 21:03 ` Richard Henderson
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).