* Re: censored naked SSE reciprocals, -mrecip
@ 2007-12-29 16:11 Uros Bizjak
2007-12-30 2:06 ` tbp
0 siblings, 1 reply; 4+ messages in thread
From: Uros Bizjak @ 2007-12-29 16:11 UTC (permalink / raw)
To: GCC; +Cc: GCC Patches, tbp
[-- Attachment #1: Type: text/plain, Size: 1399 bytes --]
Hello!
> i lately had some use for -mrecip but it turned out to come with all
> sorts of strings attached and apparently no opt-out. Briefly, barring
> inline asm, i can't get gcc to emit those ops without a NR fixup.
>
<snip>
> Questions:
> a) is that really by design?
No.
Attached patch fixes these problems by using correct shortcuts when
generating intrinsic functions.
2007-12-29 Uros Bizjak <ubizjak@gmail.com>
* config/i386/sse.md ("*divv4sf3"): Rename to "sse_divv4sf3".
("*sse_rsqrtv4sf2"): Export.
("*sse_sqrtv4sf2"): Ditto.
* config/i386/i386.c (enum ix86_builtins) [IX86_BUILTIN_RSQRTPS_NR,
IX86_BUILTIN_SQRTPS_NR]: New constants.
(struct builtin_description) [IX86_BUILTIN_DIVPS]: Use
CODE_FOR_sse_divv4sf3.
[IX86_BUILTIN_SQRTPS]: Use CODE_FOR_sse_sqrtv4sf2.
[IX86_BUILTIN_SQRTPS_NR]: New.
[IX86_BUILTIN_RSQRTPS_NR]: Ditto.
(ix86_init_mmx_sse_builtins): Initialize
__builtin_ia32_rsqrtps_nr and
__builtin_ia32_sqrtps_nr.
(ix86_builtin_vectorized_function): Convert BUILT_IN_SQRTF to
IX86_BUILTIN_SQRTPS_NR.
(ix86_builtin_reciprocal): Convert IX86_BUILTIN_SQRTPS_NR to
IX86_BUILTIN_RSQRTPS_NR.
Patch was bootstrapped and regression tested with {,-m32} on
x86_64-pc-linux-gnu. Patch is committed to SVN.
Thanks a lot for your report,
Uros.
[-- Attachment #2: r.diff.txt --]
[-- Type: text/plain, Size: 6407 bytes --]
Index: sse.md
===================================================================
--- sse.md (revision 131218)
+++ sse.md (working copy)
@@ -490,7 +490,7 @@
}
})
-(define_insn "*divv4sf3"
+(define_insn "sse_divv4sf3"
[(set (match_operand:V4SF 0 "register_operand" "=x")
(div:V4SF (match_operand:V4SF 1 "register_operand" "0")
(match_operand:V4SF 2 "nonimmediate_operand" "xm")))]
@@ -532,16 +532,7 @@
[(set_attr "type" "sse")
(set_attr "mode" "SF")])
-(define_insn "*sse_rsqrtv4sf2"
- [(set (match_operand:V4SF 0 "register_operand" "=x")
- (unspec:V4SF
- [(match_operand:V4SF 1 "nonimmediate_operand" "xm")] UNSPEC_RSQRT))]
- "TARGET_SSE"
- "rsqrtps\t{%1, %0|%0, %1}"
- [(set_attr "type" "sse")
- (set_attr "mode" "V4SF")])
-
-(define_expand "sse_rsqrtv4sf2"
+(define_expand "rsqrtv4sf2"
[(set (match_operand:V4SF 0 "register_operand" "")
(unspec:V4SF
[(match_operand:V4SF 1 "nonimmediate_operand" "")] UNSPEC_RSQRT))]
@@ -556,6 +547,15 @@
}
})
+(define_insn "sse_rsqrtv4sf2"
+ [(set (match_operand:V4SF 0 "register_operand" "=x")
+ (unspec:V4SF
+ [(match_operand:V4SF 1 "nonimmediate_operand" "xm")] UNSPEC_RSQRT))]
+ "TARGET_SSE"
+ "rsqrtps\t{%1, %0|%0, %1}"
+ [(set_attr "type" "sse")
+ (set_attr "mode" "V4SF")])
+
(define_insn "sse_vmrsqrtv4sf2"
[(set (match_operand:V4SF 0 "register_operand" "=x")
(vec_merge:V4SF
@@ -568,14 +568,6 @@
[(set_attr "type" "sse")
(set_attr "mode" "SF")])
-(define_insn "*sqrtv4sf2"
- [(set (match_operand:V4SF 0 "register_operand" "=x")
- (sqrt:V4SF (match_operand:V4SF 1 "nonimmediate_operand" "xm")))]
- "TARGET_SSE"
- "sqrtps\t{%1, %0|%0, %1}"
- [(set_attr "type" "sse")
- (set_attr "mode" "V4SF")])
-
(define_expand "sqrtv4sf2"
[(set (match_operand:V4SF 0 "register_operand" "=")
(sqrt:V4SF (match_operand:V4SF 1 "nonimmediate_operand" "")))]
@@ -590,6 +582,14 @@
}
})
+(define_insn "sse_sqrtv4sf2"
+ [(set (match_operand:V4SF 0 "register_operand" "=x")
+ (sqrt:V4SF (match_operand:V4SF 1 "nonimmediate_operand" "xm")))]
+ "TARGET_SSE"
+ "sqrtps\t{%1, %0|%0, %1}"
+ [(set_attr "type" "sse")
+ (set_attr "mode" "V4SF")])
+
(define_insn "sse_vmsqrtv4sf2"
[(set (match_operand:V4SF 0 "register_operand" "=x")
(vec_merge:V4SF
Index: i386.c
===================================================================
--- i386.c (revision 131218)
+++ i386.c (working copy)
@@ -17093,9 +17093,11 @@ enum ix86_builtins
IX86_BUILTIN_RCPPS,
IX86_BUILTIN_RCPSS,
IX86_BUILTIN_RSQRTPS,
+ IX86_BUILTIN_RSQRTPS_NR,
IX86_BUILTIN_RSQRTSS,
IX86_BUILTIN_RSQRTF,
IX86_BUILTIN_SQRTPS,
+ IX86_BUILTIN_SQRTPS_NR,
IX86_BUILTIN_SQRTSS,
IX86_BUILTIN_UNPCKHPS,
@@ -17849,7 +17851,7 @@ static const struct builtin_description
{ OPTION_MASK_ISA_SSE, CODE_FOR_addv4sf3, "__builtin_ia32_addps", IX86_BUILTIN_ADDPS, UNKNOWN, 0 },
{ OPTION_MASK_ISA_SSE, CODE_FOR_subv4sf3, "__builtin_ia32_subps", IX86_BUILTIN_SUBPS, UNKNOWN, 0 },
{ OPTION_MASK_ISA_SSE, CODE_FOR_mulv4sf3, "__builtin_ia32_mulps", IX86_BUILTIN_MULPS, UNKNOWN, 0 },
- { OPTION_MASK_ISA_SSE, CODE_FOR_divv4sf3, "__builtin_ia32_divps", IX86_BUILTIN_DIVPS, UNKNOWN, 0 },
+ { OPTION_MASK_ISA_SSE, CODE_FOR_sse_divv4sf3, "__builtin_ia32_divps", IX86_BUILTIN_DIVPS, UNKNOWN, 0 },
{ OPTION_MASK_ISA_SSE, CODE_FOR_sse_vmaddv4sf3, "__builtin_ia32_addss", IX86_BUILTIN_ADDSS, UNKNOWN, 0 },
{ OPTION_MASK_ISA_SSE, CODE_FOR_sse_vmsubv4sf3, "__builtin_ia32_subss", IX86_BUILTIN_SUBSS, UNKNOWN, 0 },
{ OPTION_MASK_ISA_SSE, CODE_FOR_sse_vmmulv4sf3, "__builtin_ia32_mulss", IX86_BUILTIN_MULSS, UNKNOWN, 0 },
@@ -18158,8 +18160,10 @@ static const struct builtin_description
{ OPTION_MASK_ISA_SSE | OPTION_MASK_ISA_3DNOW_A, CODE_FOR_mmx_pmovmskb, 0, IX86_BUILTIN_PMOVMSKB, UNKNOWN, 0 },
{ OPTION_MASK_ISA_SSE, CODE_FOR_sse_movmskps, 0, IX86_BUILTIN_MOVMSKPS, UNKNOWN, 0 },
- { OPTION_MASK_ISA_SSE, CODE_FOR_sqrtv4sf2, 0, IX86_BUILTIN_SQRTPS, UNKNOWN, 0 },
+ { OPTION_MASK_ISA_SSE, CODE_FOR_sse_sqrtv4sf2, 0, IX86_BUILTIN_SQRTPS, UNKNOWN, 0 },
+ { OPTION_MASK_ISA_SSE, CODE_FOR_sqrtv4sf2, 0, IX86_BUILTIN_SQRTPS_NR, UNKNOWN, 0 },
{ OPTION_MASK_ISA_SSE, CODE_FOR_sse_rsqrtv4sf2, 0, IX86_BUILTIN_RSQRTPS, UNKNOWN, 0 },
+ { OPTION_MASK_ISA_SSE, CODE_FOR_rsqrtv4sf2, 0, IX86_BUILTIN_RSQRTPS_NR, UNKNOWN, 0 },
{ OPTION_MASK_ISA_SSE, CODE_FOR_sse_rcpv4sf2, 0, IX86_BUILTIN_RCPPS, UNKNOWN, 0 },
{ OPTION_MASK_ISA_SSE, CODE_FOR_sse_cvtps2pi, 0, IX86_BUILTIN_CVTPS2PI, UNKNOWN, 0 },
@@ -19279,12 +19283,14 @@ ix86_init_mmx_sse_builtins (void)
def_builtin_const (OPTION_MASK_ISA_SSE, "__builtin_ia32_rcpps", v4sf_ftype_v4sf, IX86_BUILTIN_RCPPS);
def_builtin_const (OPTION_MASK_ISA_SSE, "__builtin_ia32_rcpss", v4sf_ftype_v4sf, IX86_BUILTIN_RCPSS);
def_builtin_const (OPTION_MASK_ISA_SSE, "__builtin_ia32_rsqrtps", v4sf_ftype_v4sf, IX86_BUILTIN_RSQRTPS);
+ def_builtin_const (OPTION_MASK_ISA_SSE, "__builtin_ia32_rsqrtps_nr", v4sf_ftype_v4sf, IX86_BUILTIN_RSQRTPS_NR);
def_builtin_const (OPTION_MASK_ISA_SSE, "__builtin_ia32_rsqrtss", v4sf_ftype_v4sf, IX86_BUILTIN_RSQRTSS);
ftype = build_function_type_list (float_type_node,
float_type_node,
NULL_TREE);
def_builtin_const (OPTION_MASK_ISA_SSE, "__builtin_ia32_rsqrtf", ftype, IX86_BUILTIN_RSQRTF);
def_builtin_const (OPTION_MASK_ISA_SSE, "__builtin_ia32_sqrtps", v4sf_ftype_v4sf, IX86_BUILTIN_SQRTPS);
+ def_builtin_const (OPTION_MASK_ISA_SSE, "__builtin_ia32_sqrtps_nr", v4sf_ftype_v4sf, IX86_BUILTIN_SQRTPS_NR);
def_builtin_const (OPTION_MASK_ISA_SSE, "__builtin_ia32_sqrtss", v4sf_ftype_v4sf, IX86_BUILTIN_SQRTSS);
def_builtin_const (OPTION_MASK_ISA_SSE, "__builtin_ia32_shufps", v4sf_ftype_v4sf_v4sf_int, IX86_BUILTIN_SHUFPS);
@@ -21301,7 +21307,7 @@ ix86_builtin_vectorized_function (unsign
case BUILT_IN_SQRTF:
if (out_mode == SFmode && out_n == 4
&& in_mode == SFmode && in_n == 4)
- return ix86_builtins[IX86_BUILTIN_SQRTPS];
+ return ix86_builtins[IX86_BUILTIN_SQRTPS_NR];
break;
case BUILT_IN_LRINT:
@@ -21463,8 +21469,8 @@ ix86_builtin_reciprocal (unsigned int fn
switch (fn)
{
/* Vectorized version of sqrt to rsqrt conversion. */
- case IX86_BUILTIN_SQRTPS:
- return ix86_builtins[IX86_BUILTIN_RSQRTPS];
+ case IX86_BUILTIN_SQRTPS_NR:
+ return ix86_builtins[IX86_BUILTIN_RSQRTPS_NR];
default:
return NULL_TREE;
^ permalink raw reply [flat|nested] 4+ messages in thread
* Re: censored naked SSE reciprocals, -mrecip
2007-12-29 16:11 censored naked SSE reciprocals, -mrecip Uros Bizjak
@ 2007-12-30 2:06 ` tbp
2007-12-30 9:16 ` Dave Korn
0 siblings, 1 reply; 4+ messages in thread
From: tbp @ 2007-12-30 2:06 UTC (permalink / raw)
To: Uros Bizjak; +Cc: GCC, GCC Patches
On Dec 29, 2007 4:35 PM, Uros Bizjak <ubizjak@gmail.com> wrote:
> Attached patch fixes these problems by using correct shortcuts when
> generating intrinsic functions.
>
> Patch was bootstrapped and regression tested with {,-m32} on
> x86_64-pc-linux-gnu. Patch is committed to SVN.
>
> Thanks a lot for your report,
Now that's blazing fast after-sales service. And i get no less than
two undocumented but functional builtins (as opposed to, say
__builtin_ia32_movddup, which is documented but dysfunctional) for the
same price.
As an extremely satisfied customer, i want to nominate you for the
2007 man of the year short list.
^ permalink raw reply [flat|nested] 4+ messages in thread
* RE: censored naked SSE reciprocals, -mrecip
2007-12-30 2:06 ` tbp
@ 2007-12-30 9:16 ` Dave Korn
0 siblings, 0 replies; 4+ messages in thread
From: Dave Korn @ 2007-12-30 9:16 UTC (permalink / raw)
To: 'tbp', 'Uros Bizjak'; +Cc: 'GCC', 'GCC Patches'
On 29 December 2007 23:04, tbp wrote:
> Now that's blazing fast after-sales service.
> As an extremely satisfied customer, i want to nominate you for the
> 2007 man of the year short list.
Hear hear! Uros works very hard and contributes a lot. Thank you, Uros!
cheers,
DaveK
--
Can't think of a witty .sigline today....
^ permalink raw reply [flat|nested] 4+ messages in thread
* censored naked SSE reciprocals, -mrecip
@ 2007-12-29 15:35 tbp
0 siblings, 0 replies; 4+ messages in thread
From: tbp @ 2007-12-29 15:35 UTC (permalink / raw)
To: GCC
[-- Attachment #1: Type: text/plain, Size: 1481 bytes --]
Merry xmas,
i lately had some use for -mrecip but it turned out to come with all
sorts of strings attached and apparently no opt-out. Briefly, barring
inline asm, i can't get gcc to emit those ops without a NR fixup.
# cat src/pr-recip.c
#include <xmmintrin.h>
typedef float v4sf_t __attribute__ ((__vector_size__ (16)));
__m128 foo(__m128 a) { return _mm_sqrt_ps(a); }
__m128 bar(__m128 a) { return _mm_rsqrt_ps(a); }
__m128 baz(__m128 a) { return _mm_rcp_ps(a); }
v4sf_t nope1(v4sf_t a) { return __builtin_ia32_sqrtps(a); }
v4sf_t nope2(v4sf_t a) { return __builtin_ia32_rsqrtps(a); }
v4sf_t allright(v4sf_t a) { return __builtin_ia32_rcpps(a); }
int main() { return 0; }
# /usr/local/gcc-4.3-20071221/bin/gcc -march=native -ffast-math
-mrecip -O2 src/pr-recip.c
... and as can be witnessed in the attached asm dump foo, bar, nope1,
nope2 get mangled (at least on x86-64 linux).
While i can somehow understand the logic behind the automatic
transformation of _mm_sqrt_ps - it can be argued that's what the user
has asked for - there's no obvious way to opt out. But then i really
don't understand why gcc feels the urge to tinker when i specifically
ask for a rsqrt.
To add insult to injury -mrecip, unlike fast-math, doesn't set any
macro so kludging around is a cat / mouse game.
Questions:
a) is that really by design?
b) what's the official way to dodge fixups when -mrecip is active?
c) any chance for -mrecip to set __FAST_MATH_NONE_SHALL_PASS__ or something?
[-- Attachment #2: dump.asm --]
[-- Type: application/octet-stream, Size: 3705 bytes --]
0000000000400470 <allright>:
400470: 0f 53 c0 rcpps %xmm0,%xmm0
400473: c3 retq
400474: 66 66 66 2e 0f 1f 84 nopw %cs:0x0(%rax,%rax,1)
40047b: 00 00 00 00 00
0000000000400480 <baz>:
400480: 0f 53 c0 rcpps %xmm0,%xmm0
400483: c3 retq
400484: 66 66 66 2e 0f 1f 84 nopw %cs:0x0(%rax,%rax,1)
40048b: 00 00 00 00 00
0000000000400490 <nope2>:
400490: 0f 28 d0 movaps %xmm0,%xmm2
400493: 0f 57 c9 xorps %xmm1,%xmm1
400496: 0f 28 05 d3 01 00 00 movaps 0x1d3(%rip),%xmm0 # 400670 <_IO_stdin_used+0x10>
40049d: 0f 28 da movaps %xmm2,%xmm3
4004a0: 0f c2 d9 04 cmpneqps %xmm1,%xmm3
4004a4: 0f 52 ca rsqrtps %xmm2,%xmm1
4004a7: 0f 54 cb andps %xmm3,%xmm1
4004aa: 0f 59 d1 mulps %xmm1,%xmm2
4004ad: 0f 59 d1 mulps %xmm1,%xmm2
4004b0: 0f 59 0d c9 01 00 00 mulps 0x1c9(%rip),%xmm1 # 400680 <_IO_stdin_used+0x20>
4004b7: 0f 5c c2 subps %xmm2,%xmm0
4004ba: 0f 59 c1 mulps %xmm1,%xmm0
4004bd: c3 retq
4004be: 66 90 xchg %ax,%ax
00000000004004c0 <bar>:
4004c0: 0f 28 d0 movaps %xmm0,%xmm2
4004c3: 0f 57 c9 xorps %xmm1,%xmm1
4004c6: 0f 28 05 a3 01 00 00 movaps 0x1a3(%rip),%xmm0 # 400670 <_IO_stdin_used+0x10>
4004cd: 0f 28 da movaps %xmm2,%xmm3
4004d0: 0f c2 d9 04 cmpneqps %xmm1,%xmm3
4004d4: 0f 52 ca rsqrtps %xmm2,%xmm1
4004d7: 0f 54 cb andps %xmm3,%xmm1
4004da: 0f 59 d1 mulps %xmm1,%xmm2
4004dd: 0f 59 d1 mulps %xmm1,%xmm2
4004e0: 0f 59 0d 99 01 00 00 mulps 0x199(%rip),%xmm1 # 400680 <_IO_stdin_used+0x20>
4004e7: 0f 5c c2 subps %xmm2,%xmm0
4004ea: 0f 59 c1 mulps %xmm1,%xmm0
4004ed: c3 retq
4004ee: 66 90 xchg %ax,%ax
00000000004004f0 <nope1>:
4004f0: 0f 28 d0 movaps %xmm0,%xmm2
4004f3: 0f 57 c9 xorps %xmm1,%xmm1
4004f6: 0f 28 05 73 01 00 00 movaps 0x173(%rip),%xmm0 # 400670 <_IO_stdin_used+0x10>
4004fd: 0f 28 da movaps %xmm2,%xmm3
400500: 0f c2 d9 04 cmpneqps %xmm1,%xmm3
400504: 0f 52 ca rsqrtps %xmm2,%xmm1
400507: 0f 54 cb andps %xmm3,%xmm1
40050a: 0f 59 d1 mulps %xmm1,%xmm2
40050d: 0f 59 ca mulps %xmm2,%xmm1
400510: 0f 59 15 69 01 00 00 mulps 0x169(%rip),%xmm2 # 400680 <_IO_stdin_used+0x20>
400517: 0f 5c c1 subps %xmm1,%xmm0
40051a: 0f 59 c2 mulps %xmm2,%xmm0
40051d: c3 retq
40051e: 66 90 xchg %ax,%ax
0000000000400520 <foo>:
400520: 0f 28 d0 movaps %xmm0,%xmm2
400523: 0f 57 c9 xorps %xmm1,%xmm1
400526: 0f 28 05 43 01 00 00 movaps 0x143(%rip),%xmm0 # 400670 <_IO_stdin_used+0x10>
40052d: 0f 28 da movaps %xmm2,%xmm3
400530: 0f c2 d9 04 cmpneqps %xmm1,%xmm3
400534: 0f 52 ca rsqrtps %xmm2,%xmm1
400537: 0f 54 cb andps %xmm3,%xmm1
40053a: 0f 59 d1 mulps %xmm1,%xmm2
40053d: 0f 59 ca mulps %xmm2,%xmm1
400540: 0f 59 15 39 01 00 00 mulps 0x139(%rip),%xmm2 # 400680 <_IO_stdin_used+0x20>
400547: 0f 5c c1 subps %xmm1,%xmm0
40054a: 0f 59 c2 mulps %xmm2,%xmm0
40054d: c3 retq
40054e: 90 nop
40054f: 90 nop
^ permalink raw reply [flat|nested] 4+ messages in thread
end of thread, other threads:[~2007-12-30 2:06 UTC | newest]
Thread overview: 4+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2007-12-29 16:11 censored naked SSE reciprocals, -mrecip Uros Bizjak
2007-12-30 2:06 ` tbp
2007-12-30 9:16 ` Dave Korn
-- strict thread matches above, loose matches on Subject: below --
2007-12-29 15:35 tbp
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).