* [PATCH, i386]: Committed: Fix PR target/22152
@ 2008-03-08 7:01 Uros Bizjak
2008-03-08 9:41 ` Paolo Bonzini
2008-03-08 10:21 ` Richard Guenther
0 siblings, 2 replies; 5+ messages in thread
From: Uros Bizjak @ 2008-03-08 7:01 UTC (permalink / raw)
To: GCC Patches
[-- Attachment #1: Type: text/plain, Size: 9004 bytes --]
Hello!
Attached patch substantially improves handling of MMX DImode values. It
turns out, that in order to reliably separate "native" and MMX DImode
values, a new mode is needed. Since we are using vector registers, the
natural choice is V1DImode, vector mode with one DImode element. After
this change, V1DI mode is considered to be native MMX mode (native MMX
modes were successfully separated out from non-vector modes by the
patch, committed a couple of weeks ago).
Patch rewrites all MMX DImode patterns into V1DImode, also updating
mmintrin.h for the new/changed builtins on the way. The patch also
rewrites MMX shift patterns in the same way as SSE shift patterns were
changed some time ago.
The benefits of the patch can be seen from the original testcase from PR
target/22152. For a slightly changed testcase (to avoid uninitialized
variables), non-patched gcc creates barely recognizable code:
unsigned_add3:
pushl %ebp
pxor %mm1, %mm1
movl %esp, %ebp
pushl %edi
pushl %esi
pushl %ebx
xorl %ebx, %ebx
subl $84, %esp
movl 12(%ebp), %edi
movl 8(%ebp), %eax
movl 20(%ebp), %esi
movq (%eax), %mm0
cmpl $1, %esi
movq %mm0, -56(%ebp)
movl (%edi), %eax
movl 4(%edi), %edx
movq -56(%ebp), %mm0
movl %eax, -48(%ebp)
movl %edx, -44(%ebp)
pcmpeqb %mm0, %mm0
movq -56(%ebp), %mm3
psubq %mm0, %mm1
movl $1, %eax
movq %mm1, %mm2
movq -56(%ebp), %mm0
movq -48(%ebp), %mm1
paddq %mm0, %mm1
psrlq $1, %mm0
movq %mm1, -72(%ebp)
movq -48(%ebp), %mm1
pand (%edi), %mm3
psrlq $1, %mm1
pand %mm2, %mm3
paddq %mm1, %mm0
movl -72(%ebp), %edx
paddq %mm3, %mm0
movl -68(%ebp), %ecx
psrlq $63, %mm0
movq %mm0, -32(%ebp)
jbe .L3
.p2align 4,,10
.p2align 3
.L6:
movl 16(%ebp), %ebx
movl %edx, -8(%ebx,%eax,8)
movl %ecx, -4(%ebx,%eax,8)
movl 8(%ebp), %edx
movq (%edx,%eax,8), %mm0
movq %mm0, -88(%ebp)
pand (%edi,%eax,8), %mm0
movl -88(%ebp), %edx
pand %mm2, %mm0
movl %edx, -40(%ebp)
movl -84(%ebp), %ecx
movl %ecx, -36(%ebp)
movq -40(%ebp), %mm1
movq -40(%ebp), %mm3
paddq (%edi,%eax,8), %mm1
paddq %mm0, %mm3
paddq -32(%ebp), %mm1
movq -32(%ebp), %mm0
movq %mm1, -96(%ebp)
pand %mm3, %mm0
incl %eax
paddq %mm0, %mm3
movl -96(%ebp), %edx
movl -92(%ebp), %ecx
movq %mm3, -32(%ebp)
cmpl %eax, %esi
ja .L6
leal -8(,%esi,8), %ebx
.L3:
movl 16(%ebp), %eax
movl %edx, (%eax,%ebx)
movl %ecx, 4(%eax,%ebx)
movq -32(%ebp), %mm0
addl $84, %esp
popl %ebx
popl %esi
popl %edi
leave
ret
The difference with the patch is noticeable (please note the code in the
loop):
unsigned_add3:
pushl %ebp
pxor %mm0, %mm0
movl %esp, %ebp
xorl %eax, %eax
pushl %edi
movl 20(%ebp), %edx
pushl %esi
leal -1(%edx), %ecx
pushl %ebx
movl 8(%ebp), %esi
movl 12(%ebp), %ebx
movq (%esi), %mm2
movq (%ebx), %mm4
movq %mm2, %mm1
movq %mm4, %mm3
pcmpeqb %mm2, %mm1
psrlq $1, %mm3
psubq %mm1, %mm0
xorl %edx, %edx
movq %mm0, %mm6
movq %mm2, %mm1
movq %mm2, %mm0
paddq %mm4, %mm1
pand %mm4, %mm2
psrlq $1, %mm0
pand %mm6, %mm2
paddq %mm3, %mm0
movl 16(%ebp), %edi
paddq %mm2, %mm0
movq %mm1, %mm5
psrlq $63, %mm0
cmpl $1, 20(%ebp)
movq %mm0, %mm3
jbe .L3
.align 16
.L6:
movq %mm5, (%edi,%eax,8)
movq 8(%esi,%eax,8), %mm0
movq 8(%ebx,%eax,8), %mm2
movq %mm0, %mm1
incl %eax
pand %mm2, %mm1
cmpl %ecx, %eax
pand %mm6, %mm1
paddq %mm0, %mm2
paddq %mm1, %mm0
paddq %mm3, %mm2
movq %mm3, %mm1
movq %mm2, %mm5
pand %mm0, %mm1
paddq %mm1, %mm0
movq %mm0, %mm3
jne .L6
movl 20(%ebp), %eax
leal -8(,%eax,8), %edx
.L3:
movq %mm5, (%edi,%edx)
movq %mm3, %mm0
popl %ebx
popl %esi
popl %edi
leave
ret
(FWIW, the result is 50% shorter object code).
Fortunately, none of the changed builtins were documented in the
documentation, so IMO we are free to change arguments to builtin
functions, whereas intrinsic function were not changed at all.
2008-03-07 Uros Bizjak <ubizjak@gmail.com>
PR target/22152
* config/i386/i386-modes.def (V1DI): New vector mode.
* config/i386/i386.h (VALID_MMX_REG_MODE): Add V1DImode.
* config/i386/mmx.md (MMXMODEI8): New mode iterator.
(MMXMODE248): Ditto.
(MMXMODE): Add V1DI mode.
(mmxvecsize): Change DI mode to V1DI mode.
("mov<mode>): Use MMXMODEI8 mode iterator.
("*mov<mode>_internal_rex64"): Ditto.
("*mov<mode>_internal"): Ditto.
("mmx_add<mode>3"): Ditto. Handle V1DImode for TARGET_SSE2.
("mmx_sub<mode>3"): Ditto.
("mmx_adddi3"): Remove insn pattern.
("mmx_subdi3"): Ditto.
("mmx_ashr<mode>3"): Use SImode and "yN" constraint for operand 2.
("mmx_lshr<mode>3"): Ditto. Use MMXMODE248 mode iterator.
("mmx_ashl<mode>3"): Ditto.
("mmx_lshrdi3"): Remove insn pattern.
("mmx_ashldi3"): Ditto.
* config/i386/i386.c (classify_argument): Handle V1DImode.
(function_arg_advance_32): Ditto.
(function_arg_32): Ditto.
(struct builtin_description) [IX86_BUILTIN_PADDQ]: Use
mmx_addv1di3 insn pattern.
[IX86_BUILTIN_PSUBQ]: Use mmx_subv1di3 insn pattern.
[IX86_BUILTIN_PSLL?, IX86_BUILTIN_PSRL?, IX86_BUILTIN_PSRA?,
IX86_BUILTIN_PSLL?I, IX86_BUILTIN_PSRL?I, IX86_BUILTIN_PSRA?I,
IX86_BUILTIN_PSLL?I128, IX86_BUILTIN_PSRL?I128,
IX86_BUILTIN_PSRA?I128]:
Remove definitions.
(V1DI_type_node): New node.
(v1di_ftype_v1di_int): Ditto.
(v1di_ftype_v1di_v1di): Ditto.
(v2si_ftype_v2si_si): Ditto.
(v4hi_ftype_v4hi_di): Remove node.
(v2si_ftype_v2si_di): Ditto.
(ix86_init_mmx_sse_builtins): Handle V1DImode.
(__builtin_ia32_psll?, __builtin_ia32_psrl?, __builtin_ia32_psra?):
Redefine builtins using def_builtin_const with *_ftype_*_int node.
(__builtin_ia32_psll?i, __builtin_ia32_psrl?i,
__builtin_ia32_psra?i):
Add new builtins using def_builtin_const.
(ix86_expand_builtin) [IX86_BUILTIN_PSLL?, IX86_BUILTIN_PSRL?,
IX86_BUILTIN_PSRA?, IX86_BUILTIN_PSLL?I, IX86_BUILTIN_PSRL?I,
IX86_BUILTIN_PSRA?I]: Handle builtin definitions.
* config/i386/mmintrin.h (__v1di): New typedef.
(_mm_add_si64): Cast arguments to __v1di type.
(_mm_sub_si64): Ditto.
(_mm_sll_pi16): Cast __count to __v4hi type.
(_mm_sll_pi32): Cast __count to __v2si type.
(_mm_sll_si64): Cast arguments to __v1di type.
(_mm_srl_pi16): Cast __count to __v4hi type.
(_mm_srl_pi32): Cast __count to __v2si type.
(_mm_srl_si64): Cast arguments to __v1di type.
(_mm_sra_pi16): Cast __count to __v4hi type.
(_mm_sra_pi32): Cast __count to __v2si type.
(_mm_slli_pi16): Use __builtin_ia32_psllwi.
(_mm_slli_pi32): Use __builtin_ia32_pslldi.
(_mm_slli_si64): Use __builtin_ia32_psllqi. Cast __m to __v1di type.
(_mm_srli_pi16): Use __builtin_ia32_psrlwi.
(_mm_srli_pi32): Use __builtin_ia32_psrldi.
(_mm_srli_si64): Use __builtin_ia32_psrlqi. Cast __m to __v1di type.
(_mm_srai_pi16): Use __builtin_ia32_psrawi.
(_mm_srai_pi32): Use __builtin_ia32_psradi.
* config/i386/i386.md (UNSPEC_NOP): Remove unspec definition.
* doc/extend.texi (X86 Built-in Functions) [__builtin_ia32_psll?,
__builtin_ia32_psrl?, __builtin_ia32_psra?, __builtin_ia32_psll?i,
__builtin_ia32_psrl?i, __builtin_ia32_psra?i]: Add new builtins.
Patch was bootstrapped and regression tested on i686-pc-linux-gnu and
x86_64-pc-linux-gnu {,-m32}. The testcase will be committed in a
separate commit, as I have to clean it a bit.
The patch is committed to SVN.
Uros.
[-- Attachment #2: p.diff.txt --]
[-- Type: text/plain, Size: 30224 bytes --]
Index: config/i386/i386.h
===================================================================
--- config/i386/i386.h (revision 133009)
+++ config/i386/i386.h (working copy)
@@ -1120,8 +1120,9 @@ do { \
((MODE) == V2SFmode || (MODE) == SFmode)
#define VALID_MMX_REG_MODE(MODE) \
- ((MODE) == DImode || (MODE) == V8QImode || (MODE) == V4HImode \
- || (MODE) == V2SImode || (MODE) == SImode)
+ ((MODE == V1DImode) || (MODE) == DImode \
+ || (MODE) == V2SImode || (MODE) == SImode \
+ || (MODE) == V4HImode || (MODE) == V8QImode)
/* ??? No autovectorization into MMX or 3DNOW until we can reliably
place emms and femms instructions. */
Index: config/i386/i386.md
===================================================================
--- config/i386/i386.md (revision 133009)
+++ config/i386/i386.md (working copy)
@@ -95,7 +95,6 @@
(UNSPEC_RCP 45)
(UNSPEC_RSQRT 46)
(UNSPEC_SFENCE 47)
- (UNSPEC_NOP 48) ; prevents combiner cleverness
(UNSPEC_PFRCP 49)
(UNSPEC_PFRCPIT1 40)
(UNSPEC_PFRCPIT2 41)
Index: config/i386/mmx.md
===================================================================
--- config/i386/mmx.md (revision 133009)
+++ config/i386/mmx.md (working copy)
@@ -32,16 +32,18 @@
;; 8 byte integral modes handled by MMX (and by extension, SSE)
(define_mode_iterator MMXMODEI [V8QI V4HI V2SI])
+(define_mode_iterator MMXMODEI8 [V8QI V4HI V2SI V1DI])
;; All 8-byte vector modes handled by MMX
-(define_mode_iterator MMXMODE [V8QI V4HI V2SI V2SF])
+(define_mode_iterator MMXMODE [V8QI V4HI V2SI V1DI V2SF])
;; Mix-n-match
(define_mode_iterator MMXMODE12 [V8QI V4HI])
(define_mode_iterator MMXMODE24 [V4HI V2SI])
+(define_mode_iterator MMXMODE248 [V4HI V2SI V1DI])
;; Mapping from integer vector mode to mnemonic suffix
-(define_mode_attr mmxvecsize [(V8QI "b") (V4HI "w") (V2SI "d") (DI "q")])
+(define_mode_attr mmxvecsize [(V8QI "b") (V4HI "w") (V2SI "d") (V1DI "q")])
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
;;
@@ -53,8 +55,8 @@
;; This is essential for maintaining stable calling conventions.
(define_expand "mov<mode>"
- [(set (match_operand:MMXMODEI 0 "nonimmediate_operand" "")
- (match_operand:MMXMODEI 1 "nonimmediate_operand" ""))]
+ [(set (match_operand:MMXMODEI8 0 "nonimmediate_operand" "")
+ (match_operand:MMXMODEI8 1 "nonimmediate_operand" ""))]
"TARGET_MMX"
{
ix86_expand_vector_move (<MODE>mode, operands);
@@ -62,9 +64,9 @@
})
(define_insn "*mov<mode>_internal_rex64"
- [(set (match_operand:MMXMODEI 0 "nonimmediate_operand"
+ [(set (match_operand:MMXMODEI8 0 "nonimmediate_operand"
"=rm,r,!y,!y ,m ,!y,Y2,x,x ,m,r,x")
- (match_operand:MMXMODEI 1 "vector_move_operand"
+ (match_operand:MMXMODEI8 1 "vector_move_operand"
"Cr ,m,C ,!ym,!y,Y2,!y,C,xm,x,x,r"))]
"TARGET_64BIT && TARGET_MMX
&& !(MEM_P (operands[0]) && MEM_P (operands[1]))"
@@ -86,9 +88,9 @@
(set_attr "mode" "DI")])
(define_insn "*mov<mode>_internal"
- [(set (match_operand:MMXMODEI 0 "nonimmediate_operand"
+ [(set (match_operand:MMXMODEI8 0 "nonimmediate_operand"
"=!y,!y ,m ,!y ,*Y2,*Y2,*Y2 ,m ,*x,*x,*x,m ,?r ,?m")
- (match_operand:MMXMODEI 1 "vector_move_operand"
+ (match_operand:MMXMODEI8 1 "vector_move_operand"
"C ,!ym,!y,*Y2,!y ,C ,*Y2m,*Y2,C ,*x,m ,*x,irm,r"))]
"TARGET_MMX
&& !(MEM_P (operands[0]) && MEM_P (operands[1]))"
@@ -557,26 +559,16 @@
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
(define_insn "mmx_add<mode>3"
- [(set (match_operand:MMXMODEI 0 "register_operand" "=y")
- (plus:MMXMODEI
- (match_operand:MMXMODEI 1 "nonimmediate_operand" "%0")
- (match_operand:MMXMODEI 2 "nonimmediate_operand" "ym")))]
- "TARGET_MMX && ix86_binary_operator_ok (PLUS, <MODE>mode, operands)"
+ [(set (match_operand:MMXMODEI8 0 "register_operand" "=y")
+ (plus:MMXMODEI8
+ (match_operand:MMXMODEI8 1 "nonimmediate_operand" "%0")
+ (match_operand:MMXMODEI8 2 "nonimmediate_operand" "ym")))]
+ "(TARGET_MMX || (TARGET_SSE2 && <MODE>mode == V1DImode))
+ && ix86_binary_operator_ok (PLUS, <MODE>mode, operands)"
"padd<mmxvecsize>\t{%2, %0|%0, %2}"
[(set_attr "type" "mmxadd")
(set_attr "mode" "DI")])
-(define_insn "mmx_adddi3"
- [(set (match_operand:DI 0 "register_operand" "=y")
- (unspec:DI
- [(plus:DI (match_operand:DI 1 "nonimmediate_operand" "%0")
- (match_operand:DI 2 "nonimmediate_operand" "ym"))]
- UNSPEC_NOP))]
- "TARGET_SSE2 && ix86_binary_operator_ok (PLUS, DImode, operands)"
- "paddq\t{%2, %0|%0, %2}"
- [(set_attr "type" "mmxadd")
- (set_attr "mode" "DI")])
-
(define_insn "mmx_ssadd<mode>3"
[(set (match_operand:MMXMODE12 0 "register_operand" "=y")
(ss_plus:MMXMODE12
@@ -598,26 +590,15 @@
(set_attr "mode" "DI")])
(define_insn "mmx_sub<mode>3"
- [(set (match_operand:MMXMODEI 0 "register_operand" "=y")
- (minus:MMXMODEI
- (match_operand:MMXMODEI 1 "register_operand" "0")
- (match_operand:MMXMODEI 2 "nonimmediate_operand" "ym")))]
- "TARGET_MMX"
+ [(set (match_operand:MMXMODEI8 0 "register_operand" "=y")
+ (minus:MMXMODEI8
+ (match_operand:MMXMODEI8 1 "register_operand" "0")
+ (match_operand:MMXMODEI8 2 "nonimmediate_operand" "ym")))]
+ "(TARGET_MMX || (TARGET_SSE2 && <MODE>mode == V1DImode))"
"psub<mmxvecsize>\t{%2, %0|%0, %2}"
[(set_attr "type" "mmxadd")
(set_attr "mode" "DI")])
-(define_insn "mmx_subdi3"
- [(set (match_operand:DI 0 "register_operand" "=y")
- (unspec:DI
- [(minus:DI (match_operand:DI 1 "register_operand" "0")
- (match_operand:DI 2 "nonimmediate_operand" "ym"))]
- UNSPEC_NOP))]
- "TARGET_SSE2"
- "psubq\t{%2, %0|%0, %2}"
- [(set_attr "type" "mmxadd")
- (set_attr "mode" "DI")])
-
(define_insn "mmx_sssub<mode>3"
[(set (match_operand:MMXMODE12 0 "register_operand" "=y")
(ss_minus:MMXMODE12
@@ -778,54 +759,32 @@
[(set (match_operand:MMXMODE24 0 "register_operand" "=y")
(ashiftrt:MMXMODE24
(match_operand:MMXMODE24 1 "register_operand" "0")
- (match_operand:DI 2 "nonmemory_operand" "yi")))]
+ (match_operand:SI 2 "nonmemory_operand" "yN")))]
"TARGET_MMX"
"psra<mmxvecsize>\t{%2, %0|%0, %2}"
[(set_attr "type" "mmxshft")
(set_attr "mode" "DI")])
(define_insn "mmx_lshr<mode>3"
- [(set (match_operand:MMXMODE24 0 "register_operand" "=y")
- (lshiftrt:MMXMODE24
- (match_operand:MMXMODE24 1 "register_operand" "0")
- (match_operand:DI 2 "nonmemory_operand" "yi")))]
+ [(set (match_operand:MMXMODE248 0 "register_operand" "=y")
+ (lshiftrt:MMXMODE248
+ (match_operand:MMXMODE248 1 "register_operand" "0")
+ (match_operand:SI 2 "nonmemory_operand" "yN")))]
"TARGET_MMX"
"psrl<mmxvecsize>\t{%2, %0|%0, %2}"
[(set_attr "type" "mmxshft")
(set_attr "mode" "DI")])
-(define_insn "mmx_lshrdi3"
- [(set (match_operand:DI 0 "register_operand" "=y")
- (unspec:DI
- [(lshiftrt:DI (match_operand:DI 1 "register_operand" "0")
- (match_operand:DI 2 "nonmemory_operand" "yi"))]
- UNSPEC_NOP))]
- "TARGET_MMX"
- "psrlq\t{%2, %0|%0, %2}"
- [(set_attr "type" "mmxshft")
- (set_attr "mode" "DI")])
-
(define_insn "mmx_ashl<mode>3"
- [(set (match_operand:MMXMODE24 0 "register_operand" "=y")
- (ashift:MMXMODE24
- (match_operand:MMXMODE24 1 "register_operand" "0")
- (match_operand:DI 2 "nonmemory_operand" "yi")))]
+ [(set (match_operand:MMXMODE248 0 "register_operand" "=y")
+ (ashift:MMXMODE248
+ (match_operand:MMXMODE248 1 "register_operand" "0")
+ (match_operand:SI 2 "nonmemory_operand" "yN")))]
"TARGET_MMX"
"psll<mmxvecsize>\t{%2, %0|%0, %2}"
[(set_attr "type" "mmxshft")
(set_attr "mode" "DI")])
-(define_insn "mmx_ashldi3"
- [(set (match_operand:DI 0 "register_operand" "=y")
- (unspec:DI
- [(ashift:DI (match_operand:DI 1 "register_operand" "0")
- (match_operand:DI 2 "nonmemory_operand" "yi"))]
- UNSPEC_NOP))]
- "TARGET_MMX"
- "psllq\t{%2, %0|%0, %2}"
- [(set_attr "type" "mmxshft")
- (set_attr "mode" "DI")])
-
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
;;
;; Parallel integral comparisons
Index: config/i386/mmintrin.h
===================================================================
--- config/i386/mmintrin.h (revision 133009)
+++ config/i386/mmintrin.h (working copy)
@@ -42,6 +42,7 @@ typedef int __m64 __attribute__ ((__vect
typedef int __v2si __attribute__ ((__vector_size__ (8)));
typedef short __v4hi __attribute__ ((__vector_size__ (8)));
typedef char __v8qi __attribute__ ((__vector_size__ (8)));
+typedef long long __v1di __attribute__ ((__vector_size__ (8)));
/* Empty the multimedia state. */
static __inline void __attribute__((__always_inline__, __artificial__))
@@ -309,7 +310,7 @@ _m_paddd (__m64 __m1, __m64 __m2)
static __inline __m64 __attribute__((__always_inline__, __artificial__))
_mm_add_si64 (__m64 __m1, __m64 __m2)
{
- return (__m64) __builtin_ia32_paddq ((long long)__m1, (long long)__m2);
+ return (__m64) __builtin_ia32_paddq ((__v1di)__m1, (__v1di)__m2);
}
#endif
@@ -413,7 +414,7 @@ _m_psubd (__m64 __m1, __m64 __m2)
static __inline __m64 __attribute__((__always_inline__, __artificial__))
_mm_sub_si64 (__m64 __m1, __m64 __m2)
{
- return (__m64) __builtin_ia32_psubq ((long long)__m1, (long long)__m2);
+ return (__m64) __builtin_ia32_psubq ((__v1di)__m1, (__v1di)__m2);
}
#endif
@@ -520,7 +521,7 @@ _m_pmullw (__m64 __m1, __m64 __m2)
static __inline __m64 __attribute__((__always_inline__, __artificial__))
_mm_sll_pi16 (__m64 __m, __m64 __count)
{
- return (__m64) __builtin_ia32_psllw ((__v4hi)__m, (long long)__count);
+ return (__m64) __builtin_ia32_psllw ((__v4hi)__m, (__v4hi)__count);
}
static __inline __m64 __attribute__((__always_inline__, __artificial__))
@@ -532,7 +533,7 @@ _m_psllw (__m64 __m, __m64 __count)
static __inline __m64 __attribute__((__always_inline__, __artificial__))
_mm_slli_pi16 (__m64 __m, int __count)
{
- return (__m64) __builtin_ia32_psllw ((__v4hi)__m, __count);
+ return (__m64) __builtin_ia32_psllwi ((__v4hi)__m, __count);
}
static __inline __m64 __attribute__((__always_inline__, __artificial__))
@@ -545,7 +546,7 @@ _m_psllwi (__m64 __m, int __count)
static __inline __m64 __attribute__((__always_inline__, __artificial__))
_mm_sll_pi32 (__m64 __m, __m64 __count)
{
- return (__m64) __builtin_ia32_pslld ((__v2si)__m, (long long)__count);
+ return (__m64) __builtin_ia32_pslld ((__v2si)__m, (__v2si)__count);
}
static __inline __m64 __attribute__((__always_inline__, __artificial__))
@@ -557,7 +558,7 @@ _m_pslld (__m64 __m, __m64 __count)
static __inline __m64 __attribute__((__always_inline__, __artificial__))
_mm_slli_pi32 (__m64 __m, int __count)
{
- return (__m64) __builtin_ia32_pslld ((__v2si)__m, __count);
+ return (__m64) __builtin_ia32_pslldi ((__v2si)__m, __count);
}
static __inline __m64 __attribute__((__always_inline__, __artificial__))
@@ -570,7 +571,7 @@ _m_pslldi (__m64 __m, int __count)
static __inline __m64 __attribute__((__always_inline__, __artificial__))
_mm_sll_si64 (__m64 __m, __m64 __count)
{
- return (__m64) __builtin_ia32_psllq ((long long)__m, (long long)__count);
+ return (__m64) __builtin_ia32_psllq ((__v1di)__m, (__v1di)__count);
}
static __inline __m64 __attribute__((__always_inline__, __artificial__))
@@ -582,7 +583,7 @@ _m_psllq (__m64 __m, __m64 __count)
static __inline __m64 __attribute__((__always_inline__, __artificial__))
_mm_slli_si64 (__m64 __m, int __count)
{
- return (__m64) __builtin_ia32_psllq ((long long)__m, (long long)__count);
+ return (__m64) __builtin_ia32_psllqi ((__v1di)__m, __count);
}
static __inline __m64 __attribute__((__always_inline__, __artificial__))
@@ -595,7 +596,7 @@ _m_psllqi (__m64 __m, int __count)
static __inline __m64 __attribute__((__always_inline__, __artificial__))
_mm_sra_pi16 (__m64 __m, __m64 __count)
{
- return (__m64) __builtin_ia32_psraw ((__v4hi)__m, (long long)__count);
+ return (__m64) __builtin_ia32_psraw ((__v4hi)__m, (__v4hi)__count);
}
static __inline __m64 __attribute__((__always_inline__, __artificial__))
@@ -607,7 +608,7 @@ _m_psraw (__m64 __m, __m64 __count)
static __inline __m64 __attribute__((__always_inline__, __artificial__))
_mm_srai_pi16 (__m64 __m, int __count)
{
- return (__m64) __builtin_ia32_psraw ((__v4hi)__m, __count);
+ return (__m64) __builtin_ia32_psrawi ((__v4hi)__m, __count);
}
static __inline __m64 __attribute__((__always_inline__, __artificial__))
@@ -620,7 +621,7 @@ _m_psrawi (__m64 __m, int __count)
static __inline __m64 __attribute__((__always_inline__, __artificial__))
_mm_sra_pi32 (__m64 __m, __m64 __count)
{
- return (__m64) __builtin_ia32_psrad ((__v2si)__m, (long long)__count);
+ return (__m64) __builtin_ia32_psrad ((__v2si)__m, (__v2si)__count);
}
static __inline __m64 __attribute__((__always_inline__, __artificial__))
@@ -632,7 +633,7 @@ _m_psrad (__m64 __m, __m64 __count)
static __inline __m64 __attribute__((__always_inline__, __artificial__))
_mm_srai_pi32 (__m64 __m, int __count)
{
- return (__m64) __builtin_ia32_psrad ((__v2si)__m, __count);
+ return (__m64) __builtin_ia32_psradi ((__v2si)__m, __count);
}
static __inline __m64 __attribute__((__always_inline__, __artificial__))
@@ -645,7 +646,7 @@ _m_psradi (__m64 __m, int __count)
static __inline __m64 __attribute__((__always_inline__, __artificial__))
_mm_srl_pi16 (__m64 __m, __m64 __count)
{
- return (__m64) __builtin_ia32_psrlw ((__v4hi)__m, (long long)__count);
+ return (__m64) __builtin_ia32_psrlw ((__v4hi)__m, (__v4hi)__count);
}
static __inline __m64 __attribute__((__always_inline__, __artificial__))
@@ -657,7 +658,7 @@ _m_psrlw (__m64 __m, __m64 __count)
static __inline __m64 __attribute__((__always_inline__, __artificial__))
_mm_srli_pi16 (__m64 __m, int __count)
{
- return (__m64) __builtin_ia32_psrlw ((__v4hi)__m, __count);
+ return (__m64) __builtin_ia32_psrlwi ((__v4hi)__m, __count);
}
static __inline __m64 __attribute__((__always_inline__, __artificial__))
@@ -670,7 +671,7 @@ _m_psrlwi (__m64 __m, int __count)
static __inline __m64 __attribute__((__always_inline__, __artificial__))
_mm_srl_pi32 (__m64 __m, __m64 __count)
{
- return (__m64) __builtin_ia32_psrld ((__v2si)__m, (long long)__count);
+ return (__m64) __builtin_ia32_psrld ((__v2si)__m, (__v2si)__count);
}
static __inline __m64 __attribute__((__always_inline__, __artificial__))
@@ -682,7 +683,7 @@ _m_psrld (__m64 __m, __m64 __count)
static __inline __m64 __attribute__((__always_inline__, __artificial__))
_mm_srli_pi32 (__m64 __m, int __count)
{
- return (__m64) __builtin_ia32_psrld ((__v2si)__m, __count);
+ return (__m64) __builtin_ia32_psrldi ((__v2si)__m, __count);
}
static __inline __m64 __attribute__((__always_inline__, __artificial__))
@@ -695,7 +696,7 @@ _m_psrldi (__m64 __m, int __count)
static __inline __m64 __attribute__((__always_inline__, __artificial__))
_mm_srl_si64 (__m64 __m, __m64 __count)
{
- return (__m64) __builtin_ia32_psrlq ((long long)__m, (long long)__count);
+ return (__m64) __builtin_ia32_psrlq ((__v1di)__m, (__v1di)__count);
}
static __inline __m64 __attribute__((__always_inline__, __artificial__))
@@ -707,7 +708,7 @@ _m_psrlq (__m64 __m, __m64 __count)
static __inline __m64 __attribute__((__always_inline__, __artificial__))
_mm_srli_si64 (__m64 __m, int __count)
{
- return (__m64) __builtin_ia32_psrlq ((long long)__m, (long long)__count);
+ return (__m64) __builtin_ia32_psrlqi ((__v1di)__m, __count);
}
static __inline __m64 __attribute__((__always_inline__, __artificial__))
Index: config/i386/i386-modes.def
===================================================================
--- config/i386/i386-modes.def (revision 133009)
+++ config/i386/i386-modes.def (working copy)
@@ -79,6 +79,7 @@ VECTOR_MODES (INT, 8); /* V
VECTOR_MODES (INT, 16); /* V16QI V8HI V4SI V2DI */
VECTOR_MODES (FLOAT, 8); /* V4HF V2SF */
VECTOR_MODES (FLOAT, 16); /* V8HF V4SF V2DF */
+VECTOR_MODE (INT, DI, 1); /* V1DI */
VECTOR_MODE (INT, QI, 2); /* V2QI */
VECTOR_MODE (INT, DI, 4); /* V4DI */
VECTOR_MODE (INT, SI, 8); /* V8SI */
Index: config/i386/i386.c
===================================================================
--- config/i386/i386.c (revision 133009)
+++ config/i386/i386.c (working copy)
@@ -3838,7 +3838,7 @@ classify_argument (enum machine_mode mod
}
/* for V1xx modes, just use the base mode */
- if (VECTOR_MODE_P (mode)
+ if (VECTOR_MODE_P (mode) && mode != V1DImode
&& GET_MODE_SIZE (GET_MODE_INNER (mode)) == bytes)
mode = GET_MODE_INNER (mode);
@@ -3910,6 +3910,7 @@ classify_argument (enum machine_mode mod
classes[0] = X86_64_SSE_CLASS;
classes[1] = X86_64_SSEUP_CLASS;
return 2;
+ case V1DImode:
case V2SFmode:
case V2SImode:
case V4HImode:
@@ -4211,6 +4212,7 @@ function_arg_advance_32 (CUMULATIVE_ARGS
case V4HImode:
case V2SImode:
case V2SFmode:
+ case V1DImode:
if (!type || !AGGREGATE_TYPE_P (type))
{
cum->mmx_words += words;
@@ -4374,6 +4376,7 @@ function_arg_32 (CUMULATIVE_ARGS *cum, e
case V4HImode:
case V2SImode:
case V2SFmode:
+ case V1DImode:
if (!type || !AGGREGATE_TYPE_P (type))
{
if (!TARGET_MMX && !warnedmmx && cum->warn_mmx)
@@ -17955,11 +17958,11 @@ static const struct builtin_description
{ OPTION_MASK_ISA_MMX, CODE_FOR_mmx_addv8qi3, "__builtin_ia32_paddb", IX86_BUILTIN_PADDB, UNKNOWN, 0 },
{ OPTION_MASK_ISA_MMX, CODE_FOR_mmx_addv4hi3, "__builtin_ia32_paddw", IX86_BUILTIN_PADDW, UNKNOWN, 0 },
{ OPTION_MASK_ISA_MMX, CODE_FOR_mmx_addv2si3, "__builtin_ia32_paddd", IX86_BUILTIN_PADDD, UNKNOWN, 0 },
- { OPTION_MASK_ISA_SSE2, CODE_FOR_mmx_adddi3, "__builtin_ia32_paddq", IX86_BUILTIN_PADDQ, UNKNOWN, 0 },
+ { OPTION_MASK_ISA_SSE2, CODE_FOR_mmx_addv1di3, "__builtin_ia32_paddq", IX86_BUILTIN_PADDQ, UNKNOWN, 0 },
{ OPTION_MASK_ISA_MMX, CODE_FOR_mmx_subv8qi3, "__builtin_ia32_psubb", IX86_BUILTIN_PSUBB, UNKNOWN, 0 },
{ OPTION_MASK_ISA_MMX, CODE_FOR_mmx_subv4hi3, "__builtin_ia32_psubw", IX86_BUILTIN_PSUBW, UNKNOWN, 0 },
{ OPTION_MASK_ISA_MMX, CODE_FOR_mmx_subv2si3, "__builtin_ia32_psubd", IX86_BUILTIN_PSUBD, UNKNOWN, 0 },
- { OPTION_MASK_ISA_SSE2, CODE_FOR_mmx_subdi3, "__builtin_ia32_psubq", IX86_BUILTIN_PSUBQ, UNKNOWN, 0 },
+ { OPTION_MASK_ISA_SSE2, CODE_FOR_mmx_subv1di3, "__builtin_ia32_psubq", IX86_BUILTIN_PSUBQ, UNKNOWN, 0 },
{ OPTION_MASK_ISA_MMX, CODE_FOR_mmx_ssaddv8qi3, "__builtin_ia32_paddsb", IX86_BUILTIN_PADDSB, UNKNOWN, 0 },
{ OPTION_MASK_ISA_MMX, CODE_FOR_mmx_ssaddv4hi3, "__builtin_ia32_paddsw", IX86_BUILTIN_PADDSW, UNKNOWN, 0 },
@@ -18010,25 +18013,6 @@ static const struct builtin_description
{ OPTION_MASK_ISA_SSE, CODE_FOR_sse_cvtsi2ss, 0, IX86_BUILTIN_CVTSI2SS, UNKNOWN, 0 },
{ OPTION_MASK_ISA_SSE | OPTION_MASK_ISA_64BIT, CODE_FOR_sse_cvtsi2ssq, 0, IX86_BUILTIN_CVTSI642SS, UNKNOWN, 0 },
- { OPTION_MASK_ISA_MMX, CODE_FOR_mmx_ashlv4hi3, 0, IX86_BUILTIN_PSLLW, UNKNOWN, 0 },
- { OPTION_MASK_ISA_MMX, CODE_FOR_mmx_ashlv4hi3, 0, IX86_BUILTIN_PSLLWI, UNKNOWN, 0 },
- { OPTION_MASK_ISA_MMX, CODE_FOR_mmx_ashlv2si3, 0, IX86_BUILTIN_PSLLD, UNKNOWN, 0 },
- { OPTION_MASK_ISA_MMX, CODE_FOR_mmx_ashlv2si3, 0, IX86_BUILTIN_PSLLDI, UNKNOWN, 0 },
- { OPTION_MASK_ISA_MMX, CODE_FOR_mmx_ashldi3, 0, IX86_BUILTIN_PSLLQ, UNKNOWN, 0 },
- { OPTION_MASK_ISA_MMX, CODE_FOR_mmx_ashldi3, 0, IX86_BUILTIN_PSLLQI, UNKNOWN, 0 },
-
- { OPTION_MASK_ISA_MMX, CODE_FOR_mmx_lshrv4hi3, 0, IX86_BUILTIN_PSRLW, UNKNOWN, 0 },
- { OPTION_MASK_ISA_MMX, CODE_FOR_mmx_lshrv4hi3, 0, IX86_BUILTIN_PSRLWI, UNKNOWN, 0 },
- { OPTION_MASK_ISA_MMX, CODE_FOR_mmx_lshrv2si3, 0, IX86_BUILTIN_PSRLD, UNKNOWN, 0 },
- { OPTION_MASK_ISA_MMX, CODE_FOR_mmx_lshrv2si3, 0, IX86_BUILTIN_PSRLDI, UNKNOWN, 0 },
- { OPTION_MASK_ISA_MMX, CODE_FOR_mmx_lshrdi3, 0, IX86_BUILTIN_PSRLQ, UNKNOWN, 0 },
- { OPTION_MASK_ISA_MMX, CODE_FOR_mmx_lshrdi3, 0, IX86_BUILTIN_PSRLQI, UNKNOWN, 0 },
-
- { OPTION_MASK_ISA_MMX, CODE_FOR_mmx_ashrv4hi3, 0, IX86_BUILTIN_PSRAW, UNKNOWN, 0 },
- { OPTION_MASK_ISA_MMX, CODE_FOR_mmx_ashrv4hi3, 0, IX86_BUILTIN_PSRAWI, UNKNOWN, 0 },
- { OPTION_MASK_ISA_MMX, CODE_FOR_mmx_ashrv2si3, 0, IX86_BUILTIN_PSRAD, UNKNOWN, 0 },
- { OPTION_MASK_ISA_MMX, CODE_FOR_mmx_ashrv2si3, 0, IX86_BUILTIN_PSRADI, UNKNOWN, 0 },
-
{ OPTION_MASK_ISA_SSE | OPTION_MASK_ISA_3DNOW_A, CODE_FOR_mmx_psadbw, 0, IX86_BUILTIN_PSADBW, UNKNOWN, 0 },
{ OPTION_MASK_ISA_MMX, CODE_FOR_mmx_pmaddwd, 0, IX86_BUILTIN_PMADDWD, UNKNOWN, 0 },
@@ -18140,17 +18124,6 @@ static const struct builtin_description
{ OPTION_MASK_ISA_SSE2, CODE_FOR_sse2_umulsidi3, 0, IX86_BUILTIN_PMULUDQ, UNKNOWN, 0 },
{ OPTION_MASK_ISA_SSE2, CODE_FOR_sse2_umulv2siv2di3, 0, IX86_BUILTIN_PMULUDQ128, UNKNOWN, 0 },
- { OPTION_MASK_ISA_SSE2, CODE_FOR_ashlv8hi3, 0, IX86_BUILTIN_PSLLWI128, UNKNOWN, 0 },
- { OPTION_MASK_ISA_SSE2, CODE_FOR_ashlv4si3, 0, IX86_BUILTIN_PSLLDI128, UNKNOWN, 0 },
- { OPTION_MASK_ISA_SSE2, CODE_FOR_ashlv2di3, 0, IX86_BUILTIN_PSLLQI128, UNKNOWN, 0 },
-
- { OPTION_MASK_ISA_SSE2, CODE_FOR_lshrv8hi3, 0, IX86_BUILTIN_PSRLWI128, UNKNOWN, 0 },
- { OPTION_MASK_ISA_SSE2, CODE_FOR_lshrv4si3, 0, IX86_BUILTIN_PSRLDI128, UNKNOWN, 0 },
- { OPTION_MASK_ISA_SSE2, CODE_FOR_lshrv2di3, 0, IX86_BUILTIN_PSRLQI128, UNKNOWN, 0 },
-
- { OPTION_MASK_ISA_SSE2, CODE_FOR_ashrv8hi3, 0, IX86_BUILTIN_PSRAWI128, UNKNOWN, 0 },
- { OPTION_MASK_ISA_SSE2, CODE_FOR_ashrv4si3, 0, IX86_BUILTIN_PSRADI128, UNKNOWN, 0 },
-
{ OPTION_MASK_ISA_SSE2, CODE_FOR_sse2_pmaddwd, 0, IX86_BUILTIN_PMADDWD128, UNKNOWN, 0 },
{ OPTION_MASK_ISA_SSE2, CODE_FOR_sse2_cvtsi2sd, 0, IX86_BUILTIN_CVTSI2SD, UNKNOWN, 0 },
@@ -18212,6 +18185,7 @@ static const struct builtin_description
static const struct builtin_description bdesc_1arg[] =
{
+ /* SSE */
{ 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 },
@@ -18228,6 +18202,7 @@ static const struct builtin_description
{ OPTION_MASK_ISA_SSE, CODE_FOR_sse_cvttss2si, 0, IX86_BUILTIN_CVTTSS2SI, UNKNOWN, 0 },
{ OPTION_MASK_ISA_SSE | OPTION_MASK_ISA_64BIT, CODE_FOR_sse_cvttss2siq, 0, IX86_BUILTIN_CVTTSS2SI64, UNKNOWN, 0 },
+ /* SSE2 */
{ OPTION_MASK_ISA_SSE2, CODE_FOR_sse2_pmovmskb, 0, IX86_BUILTIN_PMOVMSKB128, UNKNOWN, 0 },
{ OPTION_MASK_ISA_SSE2, CODE_FOR_sse2_movmskpd, 0, IX86_BUILTIN_MOVMSKPD, UNKNOWN, 0 },
@@ -18585,6 +18560,8 @@ ix86_init_mmx_sse_builtins (void)
tree V16QI_type_node = build_vector_type_for_mode (char_type_node, V16QImode);
tree V2SI_type_node = build_vector_type_for_mode (intSI_type_node, V2SImode);
+ tree V1DI_type_node
+ = build_vector_type_for_mode (long_long_integer_type_node, V1DImode);
tree V2SF_type_node = build_vector_type_for_mode (float_type_node, V2SFmode);
tree V2DI_type_node
= build_vector_type_for_mode (long_long_integer_type_node, V2DImode);
@@ -18649,14 +18626,13 @@ ix86_init_mmx_sse_builtins (void)
tree v4hi_ftype_v4hi_int
= build_function_type_list (V4HI_type_node,
V4HI_type_node, integer_type_node, NULL_TREE);
- tree v4hi_ftype_v4hi_di
- = build_function_type_list (V4HI_type_node,
- V4HI_type_node, long_long_unsigned_type_node,
- NULL_TREE);
- tree v2si_ftype_v2si_di
+ tree v2si_ftype_v2si_int
= build_function_type_list (V2SI_type_node,
- V2SI_type_node, long_long_unsigned_type_node,
- NULL_TREE);
+ V2SI_type_node, integer_type_node, NULL_TREE);
+ tree v1di_ftype_v1di_int
+ = build_function_type_list (V1DI_type_node,
+ V1DI_type_node, integer_type_node, NULL_TREE);
+
tree void_ftype_void
= build_function_type (void_type_node, void_list_node);
tree void_ftype_unsigned
@@ -18723,10 +18699,9 @@ ix86_init_mmx_sse_builtins (void)
tree v2si_ftype_v2si_v2si
= build_function_type_list (V2SI_type_node,
V2SI_type_node, V2SI_type_node, NULL_TREE);
- tree di_ftype_di_di
- = build_function_type_list (long_long_unsigned_type_node,
- long_long_unsigned_type_node,
- long_long_unsigned_type_node, NULL_TREE);
+ tree v1di_ftype_v1di_v1di
+ = build_function_type_list (V1DI_type_node,
+ V1DI_type_node, V1DI_type_node, NULL_TREE);
tree di_ftype_di_di_int
= build_function_type_list (long_long_unsigned_type_node,
@@ -19182,8 +19157,8 @@ ix86_init_mmx_sse_builtins (void)
case V2SImode:
type = v2si_ftype_v2si_v2si;
break;
- case DImode:
- type = di_ftype_di_di;
+ case V1DImode:
+ type = v1di_ftype_v1di_v1di;
break;
default:
@@ -19275,16 +19250,25 @@ ix86_init_mmx_sse_builtins (void)
/* Add the remaining MMX insns with somewhat more complicated types. */
def_builtin (OPTION_MASK_ISA_MMX, "__builtin_ia32_emms", void_ftype_void, IX86_BUILTIN_EMMS);
- def_builtin_const (OPTION_MASK_ISA_MMX, "__builtin_ia32_psllw", v4hi_ftype_v4hi_di, IX86_BUILTIN_PSLLW);
- def_builtin_const (OPTION_MASK_ISA_MMX, "__builtin_ia32_pslld", v2si_ftype_v2si_di, IX86_BUILTIN_PSLLD);
- def_builtin_const (OPTION_MASK_ISA_MMX, "__builtin_ia32_psllq", di_ftype_di_di, IX86_BUILTIN_PSLLQ);
-
- def_builtin_const (OPTION_MASK_ISA_MMX, "__builtin_ia32_psrlw", v4hi_ftype_v4hi_di, IX86_BUILTIN_PSRLW);
- def_builtin_const (OPTION_MASK_ISA_MMX, "__builtin_ia32_psrld", v2si_ftype_v2si_di, IX86_BUILTIN_PSRLD);
- def_builtin_const (OPTION_MASK_ISA_MMX, "__builtin_ia32_psrlq", di_ftype_di_di, IX86_BUILTIN_PSRLQ);
- def_builtin_const (OPTION_MASK_ISA_MMX, "__builtin_ia32_psraw", v4hi_ftype_v4hi_di, IX86_BUILTIN_PSRAW);
- def_builtin_const (OPTION_MASK_ISA_MMX, "__builtin_ia32_psrad", v2si_ftype_v2si_di, IX86_BUILTIN_PSRAD);
+ def_builtin_const (OPTION_MASK_ISA_MMX, "__builtin_ia32_psllwi", v4hi_ftype_v4hi_int, IX86_BUILTIN_PSLLWI);
+ def_builtin_const (OPTION_MASK_ISA_MMX, "__builtin_ia32_pslldi", v2si_ftype_v2si_int, IX86_BUILTIN_PSLLDI);
+ def_builtin_const (OPTION_MASK_ISA_MMX, "__builtin_ia32_psllqi", v1di_ftype_v1di_int, IX86_BUILTIN_PSLLQI);
+ def_builtin_const (OPTION_MASK_ISA_MMX, "__builtin_ia32_psllw", v4hi_ftype_v4hi_v4hi, IX86_BUILTIN_PSLLW);
+ def_builtin_const (OPTION_MASK_ISA_MMX, "__builtin_ia32_pslld", v2si_ftype_v2si_v2si, IX86_BUILTIN_PSLLD);
+ def_builtin_const (OPTION_MASK_ISA_MMX, "__builtin_ia32_psllq", v1di_ftype_v1di_v1di, IX86_BUILTIN_PSLLQ);
+
+ def_builtin_const (OPTION_MASK_ISA_MMX, "__builtin_ia32_psrlwi", v4hi_ftype_v4hi_int, IX86_BUILTIN_PSRLWI);
+ def_builtin_const (OPTION_MASK_ISA_MMX, "__builtin_ia32_psrldi", v2si_ftype_v2si_int, IX86_BUILTIN_PSRLDI);
+ def_builtin_const (OPTION_MASK_ISA_MMX, "__builtin_ia32_psrlqi", v1di_ftype_v1di_int, IX86_BUILTIN_PSRLQI);
+ def_builtin_const (OPTION_MASK_ISA_MMX, "__builtin_ia32_psrlw", v4hi_ftype_v4hi_v4hi, IX86_BUILTIN_PSRLW);
+ def_builtin_const (OPTION_MASK_ISA_MMX, "__builtin_ia32_psrld", v2si_ftype_v2si_v2si, IX86_BUILTIN_PSRLD);
+ def_builtin_const (OPTION_MASK_ISA_MMX, "__builtin_ia32_psrlq", v1di_ftype_v1di_v1di, IX86_BUILTIN_PSRLQ);
+
+ def_builtin_const (OPTION_MASK_ISA_MMX, "__builtin_ia32_psrawi", v4hi_ftype_v4hi_int, IX86_BUILTIN_PSRAWI);
+ def_builtin_const (OPTION_MASK_ISA_MMX, "__builtin_ia32_psradi", v2si_ftype_v2si_int, IX86_BUILTIN_PSRADI);
+ def_builtin_const (OPTION_MASK_ISA_MMX, "__builtin_ia32_psraw", v4hi_ftype_v4hi_v4hi, IX86_BUILTIN_PSRAW);
+ def_builtin_const (OPTION_MASK_ISA_MMX, "__builtin_ia32_psrad", v2si_ftype_v2si_v2si, IX86_BUILTIN_PSRAD);
def_builtin_const (OPTION_MASK_ISA_SSE | OPTION_MASK_ISA_3DNOW_A, "__builtin_ia32_pshufw", v4hi_ftype_v4hi_int, IX86_BUILTIN_PSHUFW);
def_builtin_const (OPTION_MASK_ISA_MMX, "__builtin_ia32_pmaddwd", v2si_ftype_v4hi_v4hi, IX86_BUILTIN_PMADDWD);
@@ -20829,6 +20813,39 @@ ix86_expand_builtin (tree exp, rtx targe
emit_insn (pat);
return target;
+ case IX86_BUILTIN_PSLLW:
+ case IX86_BUILTIN_PSLLWI:
+ icode = CODE_FOR_mmx_ashlv4hi3;
+ goto do_pshift;
+ case IX86_BUILTIN_PSLLD:
+ case IX86_BUILTIN_PSLLDI:
+ icode = CODE_FOR_mmx_ashlv2si3;
+ goto do_pshift;
+ case IX86_BUILTIN_PSLLQ:
+ case IX86_BUILTIN_PSLLQI:
+ icode = CODE_FOR_mmx_ashlv1di3;
+ goto do_pshift;
+ case IX86_BUILTIN_PSRAW:
+ case IX86_BUILTIN_PSRAWI:
+ icode = CODE_FOR_mmx_ashrv4hi3;
+ goto do_pshift;
+ case IX86_BUILTIN_PSRAD:
+ case IX86_BUILTIN_PSRADI:
+ icode = CODE_FOR_mmx_ashrv2si3;
+ goto do_pshift;
+ case IX86_BUILTIN_PSRLW:
+ case IX86_BUILTIN_PSRLWI:
+ icode = CODE_FOR_mmx_lshrv4hi3;
+ goto do_pshift;
+ case IX86_BUILTIN_PSRLD:
+ case IX86_BUILTIN_PSRLDI:
+ icode = CODE_FOR_mmx_lshrv2si3;
+ goto do_pshift;
+ case IX86_BUILTIN_PSRLQ:
+ case IX86_BUILTIN_PSRLQI:
+ icode = CODE_FOR_mmx_lshrv1di3;
+ goto do_pshift;
+
case IX86_BUILTIN_PSLLW128:
case IX86_BUILTIN_PSLLWI128:
icode = CODE_FOR_ashlv8hi3;
Index: doc/extend.texi
===================================================================
--- doc/extend.texi (revision 133009)
+++ doc/extend.texi (working copy)
@@ -7476,6 +7476,24 @@ v2si __builtin_ia32_punpckldq (v2si, v2s
v8qi __builtin_ia32_packsswb (v4hi, v4hi)
v4hi __builtin_ia32_packssdw (v2si, v2si)
v8qi __builtin_ia32_packuswb (v4hi, v4hi)
+
+v8hi __builtin_ia32_psllw (v4hi, v4hi)
+v4si __builtin_ia32_pslld (v2si, v2si)
+v2di __builtin_ia32_psllq (v1di, v1di)
+v8hi __builtin_ia32_psrlw (v8hi, v2si)
+v4si __builtin_ia32_psrld (v4si, v2si)
+v2di __builtin_ia32_psrlq (v1di, v1di)
+v8hi __builtin_ia32_psraw (v4hi, v4hi)
+v4si __builtin_ia32_psrad (v2si, v2si)
+v8hi __builtin_ia32_psllwi (v4hi, int)
+v4si __builtin_ia32_pslldi (v2si, int)
+v2di __builtin_ia32_psllqi (v1di, int)
+v8hi __builtin_ia32_psrlwi (v4hi, int)
+v4si __builtin_ia32_psrldi (v2si, int)
+v2di __builtin_ia32_psrlqi (v1di, int)
+v8hi __builtin_ia32_psrawi (v4hi, int)
+v4si __builtin_ia32_psradi (v2si, int)
+
@end smallexample
The following built-in functions are made available either with
^ permalink raw reply [flat|nested] 5+ messages in thread
* Re: [PATCH, i386]: Committed: Fix PR target/22152
2008-03-08 7:01 [PATCH, i386]: Committed: Fix PR target/22152 Uros Bizjak
@ 2008-03-08 9:41 ` Paolo Bonzini
2008-03-13 17:50 ` Uros Bizjak
2008-03-08 10:21 ` Richard Guenther
1 sibling, 1 reply; 5+ messages in thread
From: Paolo Bonzini @ 2008-03-08 9:41 UTC (permalink / raw)
To: Uros Bizjak; +Cc: GCC Patches
> +v2di __builtin_ia32_psrlqi (v1di, int)
Something wrong in the return types?
Paolo
^ permalink raw reply [flat|nested] 5+ messages in thread
* Re: [PATCH, i386]: Committed: Fix PR target/22152
2008-03-08 9:41 ` Paolo Bonzini
@ 2008-03-13 17:50 ` Uros Bizjak
0 siblings, 0 replies; 5+ messages in thread
From: Uros Bizjak @ 2008-03-13 17:50 UTC (permalink / raw)
To: Paolo Bonzini; +Cc: GCC Patches
[-- Attachment #1: Type: text/plain, Size: 591 bytes --]
Paolo Bonzini wrote:
>> +v2di __builtin_ia32_psrlqi (v1di, int)
>
> Something wrong in the return types?
Fixed by a follow-up doc fix to match documents with reality:
2008-03-13 Uros Bizjak <ubizjak@gmail.com>
* doc/extend.texi (X86 Built-in Functions) [__builtin_ia32_psll?,
__builtin_ia32_psrl?, __builtin_ia32_psra?, __builtin_ia32_psll?i,
__builtin_ia32_psrl?i, __builtin_ia32_psra?i,
__builtin_ia32_psll?128,
__builtin_ia32_psrl?128, __builtin_ia32_psra?128]: Fix the mode of
input arguments and the mode of return value.
Thanks,
Uros.
[-- Attachment #2: p.diff.txt --]
[-- Type: text/plain, Size: 2525 bytes --]
Index: doc/extend.texi
===================================================================
--- doc/extend.texi (revision 133170)
+++ doc/extend.texi (working copy)
@@ -7477,22 +7477,22 @@
v4hi __builtin_ia32_packssdw (v2si, v2si)
v8qi __builtin_ia32_packuswb (v4hi, v4hi)
-v8hi __builtin_ia32_psllw (v4hi, v4hi)
-v4si __builtin_ia32_pslld (v2si, v2si)
-v2di __builtin_ia32_psllq (v1di, v1di)
-v8hi __builtin_ia32_psrlw (v8hi, v2si)
-v4si __builtin_ia32_psrld (v4si, v2si)
-v2di __builtin_ia32_psrlq (v1di, v1di)
-v8hi __builtin_ia32_psraw (v4hi, v4hi)
-v4si __builtin_ia32_psrad (v2si, v2si)
-v8hi __builtin_ia32_psllwi (v4hi, int)
-v4si __builtin_ia32_pslldi (v2si, int)
-v2di __builtin_ia32_psllqi (v1di, int)
-v8hi __builtin_ia32_psrlwi (v4hi, int)
-v4si __builtin_ia32_psrldi (v2si, int)
-v2di __builtin_ia32_psrlqi (v1di, int)
-v8hi __builtin_ia32_psrawi (v4hi, int)
-v4si __builtin_ia32_psradi (v2si, int)
+v4hi __builtin_ia32_psllw (v4hi, v4hi)
+v2si __builtin_ia32_pslld (v2si, v2si)
+v1di __builtin_ia32_psllq (v1di, v1di)
+v4hi __builtin_ia32_psrlw (v4hi, v4hi)
+v2si __builtin_ia32_psrld (v2si, v2si)
+v1di __builtin_ia32_psrlq (v1di, v1di)
+v4hi __builtin_ia32_psraw (v4hi, v4hi)
+v2si __builtin_ia32_psrad (v2si, v2si)
+v4hi __builtin_ia32_psllwi (v4hi, int)
+v2si __builtin_ia32_pslldi (v2si, int)
+v1di __builtin_ia32_psllqi (v1di, int)
+v4hi __builtin_ia32_psrlwi (v4hi, int)
+v2si __builtin_ia32_psrldi (v2si, int)
+v1di __builtin_ia32_psrlqi (v1di, int)
+v4hi __builtin_ia32_psrawi (v4hi, int)
+v2si __builtin_ia32_psradi (v2si, int)
@end smallexample
@@ -7755,14 +7755,14 @@
void __builtin_ia32_storedqu (char *, v16qi)
unsigned long long __builtin_ia32_pmuludq (v2si, v2si)
v2di __builtin_ia32_pmuludq128 (v4si, v4si)
-v8hi __builtin_ia32_psllw128 (v8hi, v2di)
-v4si __builtin_ia32_pslld128 (v4si, v2di)
-v2di __builtin_ia32_psllq128 (v4si, v2di)
-v8hi __builtin_ia32_psrlw128 (v8hi, v2di)
-v4si __builtin_ia32_psrld128 (v4si, v2di)
+v8hi __builtin_ia32_psllw128 (v8hi, v8hi)
+v4si __builtin_ia32_pslld128 (v4si, v4si)
+v2di __builtin_ia32_psllq128 (v2di, v2di)
+v8hi __builtin_ia32_psrlw128 (v8hi, v8hi)
+v4si __builtin_ia32_psrld128 (v4si, v4si)
v2di __builtin_ia32_psrlq128 (v2di, v2di)
-v8hi __builtin_ia32_psraw128 (v8hi, v2di)
-v4si __builtin_ia32_psrad128 (v4si, v2di)
+v8hi __builtin_ia32_psraw128 (v8hi, v8hi)
+v4si __builtin_ia32_psrad128 (v4si, v4si)
v2di __builtin_ia32_pslldqi128 (v2di, int)
v8hi __builtin_ia32_psllwi128 (v8hi, int)
v4si __builtin_ia32_pslldi128 (v4si, int)
^ permalink raw reply [flat|nested] 5+ messages in thread
* Re: [PATCH, i386]: Committed: Fix PR target/22152
2008-03-08 7:01 [PATCH, i386]: Committed: Fix PR target/22152 Uros Bizjak
2008-03-08 9:41 ` Paolo Bonzini
@ 2008-03-08 10:21 ` Richard Guenther
2008-03-08 12:44 ` Uros Bizjak
1 sibling, 1 reply; 5+ messages in thread
From: Richard Guenther @ 2008-03-08 10:21 UTC (permalink / raw)
To: Uros Bizjak; +Cc: GCC Patches
On Fri, Mar 7, 2008 at 7:13 PM, Uros Bizjak <ubizjak@gmail.com> wrote:
> Hello!
>
> Attached patch substantially improves handling of MMX DImode values. It
> turns out, that in order to reliably separate "native" and MMX DImode
> values, a new mode is needed. Since we are using vector registers, the
> natural choice is V1DImode, vector mode with one DImode element. After
> this change, V1DI mode is considered to be native MMX mode (native MMX
> modes were successfully separated out from non-vector modes by the
> patch, committed a couple of weeks ago).
>
> Patch rewrites all MMX DImode patterns into V1DImode, also updating
> mmintrin.h for the new/changed builtins on the way. The patch also
> rewrites MMX shift patterns in the same way as SSE shift patterns were
> changed some time ago.
>
> The benefits of the patch can be seen from the original testcase from PR
> target/22152. For a slightly changed testcase (to avoid uninitialized
> variables), non-patched gcc creates barely recognizable code:
I realize this may be hard, but with all the many tweaking patches for SSE, MMX,
etc. how do we make sure to not regress in cases we fixed earlier? So, may I
suggest you try to enter at least "something" into the testsuite? For example
scan-assembler-not ".L6.*ebp.*.L6" (no stack operations between the .L6 loop
entry and the backedge)? Maybe a little bit fragile, but at least
some confidence
would be there (and some testcases to eventually manually inspect) that
we won't regress again?
Thanks,
Richard.
^ permalink raw reply [flat|nested] 5+ messages in thread
* Re: [PATCH, i386]: Committed: Fix PR target/22152
2008-03-08 10:21 ` Richard Guenther
@ 2008-03-08 12:44 ` Uros Bizjak
0 siblings, 0 replies; 5+ messages in thread
From: Uros Bizjak @ 2008-03-08 12:44 UTC (permalink / raw)
To: Richard Guenther; +Cc: GCC Patches
[-- Attachment #1: Type: text/plain, Size: 3433 bytes --]
Richard Guenther wrote:
> I realize this may be hard, but with all the many tweaking patches for SSE, MMX,
> etc. how do we make sure to not regress in cases we fixed earlier? So, may I
> suggest you try to enter at least "something" into the testsuite? For example
> scan-assembler-not ".L6.*ebp.*.L6" (no stack operations between the .L6 loop
> entry and the backedge)? Maybe a little bit fragile, but at least
> some confidence
> would be there (and some testcases to eventually manually inspect) that
> we won't regress again?
>
Heh, I _did_ say that "... The testcase will be committed in a separate
commit, as I have to clean it a bit."
Attached to this message, please find a couple of testcases, derived
from PR 22152:
- pr22152.c is a compile time test that checks if (long long) value
stays inside MMX register. Due to the cast to (long long) in the
mmintrin.h, the value was dragged to and from the memory (-O2 -m32 -msse2):
.L3:
movl (%ebx,%eax,8), %esi
movl 4(%ebx,%eax,8), %edi
movl %esi, -24(%ebp)
movl %edi, -20(%ebp)
movq -24(%ebp), %mm0
paddq (%ecx,%eax,8), %mm0
addl $1, %eax
cmpl %eax, %edx
movq %mm0, -24(%ebp)
movq -24(%ebp), %mm0
ja .L3
The situation is now much better:
.L3:
movq (%ebx,%eax,8), %mm0
paddq (%ecx,%eax,8), %mm0
addl $1, %eax
cmpl %eax, %edx
ja .L3
- sse2-mmx.c
This is a runtime test, based on the large testcase from the PR. The
test should add two huge numbers together using MMX stuff, but
unfortunately, it doesn't work correctly (carry propagation logic is
fatally flawed). Attached test fixes this logic, so it can be used to
increase the runtime coverage of SSE2 based MMX operations.
FWIW, the loop from the testcase is now:
.L3:
movq (%esi,%eax,8), %mm2 #* a, _a.37
movq (%ebx,%eax,8), %mm3 #* b, _b
movq %mm2, %mm0 # D.2452, tmp94
paddq %mm3, %mm0 # D.2451, tmp94
movq %mm2, %mm1 # _a.37, D.2452
movq %mm3, %mm4 # _b, D.2451
paddq %mm5, %mm0 # carry, tmp94
psrlq $1, %mm1 #, D.2452
movq %mm0, (%ecx,%eax,8) # tmp94,* result
movq %mm2, %mm0 # _a.37, tmp96
pxor %mm3, %mm0 # _b, tmp96
pand %mm5, %mm0 # carry, tmp96
pand %mm3, %mm2 # _b, _a.37
por %mm0, %mm2 # tmp96, _a.37
psrlq $1, %mm4 #, D.2451
pand %mm6, %mm2 # one.38, _a.37
paddq %mm4, %mm1 # D.2451, D.2452
paddq %mm2, %mm1 # _a.37, D.2452
addl $1, %eax #, i
psrlq $63, %mm1 #, D.2452
cmpl %eax, %edx # i, count
movq %mm1, %mm5 # D.2452, carry
ja .L3 #,
.L2:
Other than that, previous changes to MMX patterns are covered by
pr22076.c, pr34256.c. In addition, all vecinit-N.c tests check that no
MMX register is used in vector initialization code (we had some problems
with this in the past).
2008-03-08 Uros Bizjak <ubizjak@gmail.com>
PR target/22152
* gcc.target/i386/pr22152.c: New test.
* gcc.target/i386/sse2-mmx.c: Ditto.
These new tests were checked on x86_64-linux-gnu {,-m32} and are
committed to mainline.
Uros.
[-- Attachment #2: p.diff.txt --]
[-- Type: text/plain, Size: 2429 bytes --]
Index: gcc.target/i386/sse2-mmx.c
===================================================================
--- gcc.target/i386/sse2-mmx.c (revision 0)
+++ gcc.target/i386/sse2-mmx.c (revision 0)
@@ -0,0 +1,75 @@
+/* { dg-do run } */
+/* { dg-options "-O2 -msse2" } */
+
+#include "sse2-check.h"
+
+#include <mmintrin.h>
+
+#define N 4
+
+unsigned long long a[N], b[N], result[N];
+
+unsigned long long check[N] =
+ { 0x101010101010100full,
+ 0x1010101010101010ull,
+ 0x1010101010101010ull,
+ 0x1010101010101010ull };
+
+__m64
+unsigned_add3 (const __m64 * a, const __m64 * b,
+ __m64 * result, unsigned int count)
+{
+ __m64 _a, _b, one, sum, carry, onesCarry;
+
+ unsigned int i;
+
+ one = _mm_cmpeq_pi8 (_a, _a);
+ one = _mm_sub_si64 (_mm_xor_si64 (one, one), one);
+
+ carry = _mm_xor_si64 (one, one);
+
+ for (i = 0; i < count; i++)
+ {
+ _a = a[i];
+ _b = b[i];
+
+ sum = _mm_add_si64 (_a, _b);
+ sum = _mm_add_si64 (sum, carry);
+
+ result[i] = sum;
+
+ onesCarry = _mm_and_si64 (_mm_xor_si64 (_a, _b), carry);
+ onesCarry = _mm_or_si64 (_mm_and_si64 (_a, _b), onesCarry);
+ onesCarry = _mm_and_si64 (onesCarry, one);
+
+ _a = _mm_srli_si64 (_a, 1);
+ _b = _mm_srli_si64 (_b, 1);
+
+ carry = _mm_add_si64 (_mm_add_si64 (_a, _b), onesCarry);
+ carry = _mm_srli_si64 (carry, 63);
+ }
+
+ _mm_empty ();
+ return carry;
+}
+
+void __attribute__((noinline))
+sse2_test (void)
+{
+ unsigned long long carry;
+ int i;
+
+ /* Really long numbers. */
+ a[3] = a[2] = a[1] = a[0] = 0xd3d3d3d3d3d3d3d3ull;
+ b[3] = b[2] = b[1] = b[0] = 0x3c3c3c3c3c3c3c3cull;
+
+ carry = (unsigned long long) unsigned_add3
+ ((__m64 *)a, (__m64 *)b, (__m64 *)result, N);
+
+ if (carry != 1)
+ abort ();
+
+ for (i = 0; i < N; i++)
+ if (result [i] != check[i])
+ abort ();
+}
Index: gcc.target/i386/pr22152.c
===================================================================
--- gcc.target/i386/pr22152.c (revision 0)
+++ gcc.target/i386/pr22152.c (revision 0)
@@ -0,0 +1,18 @@
+/* { dg-do compile } */
+/* { dg-options "-O2 -msse2" } */
+
+#include <mmintrin.h>
+
+__m64
+unsigned_add3 (const __m64 * a, const __m64 * b, unsigned long count)
+{
+ __m64 sum;
+ unsigned int i;
+
+ for (i = 1; i < count; i++)
+ sum = _mm_add_si64 (a[i], b[i]);
+
+ return sum;
+}
+
+/* { dg-final { scan-assembler-times "movq\[ \\t\]+.*%mm" 1 } } */
^ permalink raw reply [flat|nested] 5+ messages in thread
end of thread, other threads:[~2008-03-13 17:50 UTC | newest]
Thread overview: 5+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2008-03-08 7:01 [PATCH, i386]: Committed: Fix PR target/22152 Uros Bizjak
2008-03-08 9:41 ` Paolo Bonzini
2008-03-13 17:50 ` Uros Bizjak
2008-03-08 10:21 ` Richard Guenther
2008-03-08 12:44 ` Uros Bizjak
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).