OpenE2K/gcc
2
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

i386.c (builtin_description): Add __builtin_ia32_paddq and __builtin_ia32_psubq.

Browse Source 
* i386.c (builtin_description): Add __builtin_ia32_paddq and
	__builtin_ia32_psubq. Fix __builtin_ia32_paddq128
	 and __builtin_ia32_psubq128.
	* i386.h (IX86_BUILTIN_PADDQ, IX86_BUILTIN_PSUBQ): New.
	* i386.md (addv*, mmx_ior*, mmx_xoe*, mmx_and*): Add missing '%'.
	(mmx_adddi3, mmx_subdi3): New.
	* mmintrin.h (_mm_add_si64, _mm_sub_si64): New.
	* xmmintrin.h (_mm_movepi64_pi64): New.
	(_mm_add_epi64, _mm_sub_epi64): fix.
	(_mm_mul_pu16): Rename to...
	(_mm_mul_su32): ... this one.

	* builtins.c (expand_builtin_expect):  Do not predict
	flag_guess_branch_prob is not set.
	* c-semantics.c (expand_stmt): Likewise.
	* predict.c (predict_insn): Likewise.
	* stmt.c (expand_continue_loop): Likewise.
	* toplev.c (rest_of_compilation): Do not call
	note_prediction_to_br_prob and note_prediction_to_br_prob
	when not optimizing.

From-SVN: r63263
This commit is contained in:
Jan Hubicka 2003-02-22 01:32:09 +01:00 committed by Jan Hubicka
parent 6a6d417ebd
commit d50672efa7
11 changed files with 112 additions and 32 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

23
gcc/ChangeLog
View File

@ -1,3 +1,26 @@
Sat Feb 22 00:48:22 CET 2003 Jan Hubicka <jh@suse.cz>
* i386.c (builtin_description): Add __builtin_ia32_paddq and
__builtin_ia32_psubq. Fix __builtin_ia32_paddq128
and __builtin_ia32_psubq128.
* i386.h (IX86_BUILTIN_PADDQ, IX86_BUILTIN_PSUBQ): New.
* i386.md (addv*, mmx_ior*, mmx_xoe*, mmx_and*): Add missing '%'.
(mmx_adddi3, mmx_subdi3): New.
* mmintrin.h (_mm_add_si64, _mm_sub_si64): New.
* xmmintrin.h (_mm_movepi64_pi64): New.
(_mm_add_epi64, _mm_sub_epi64): fix.
(_mm_mul_pu16): Rename to...
(_mm_mul_su32): ... this one.
* builtins.c (expand_builtin_expect): Do not predict
flag_guess_branch_prob is not set.
* c-semantics.c (expand_stmt): Likewise.
* predict.c (predict_insn): Likewise.
* stmt.c (expand_continue_loop): Likewise.
* toplev.c (rest_of_compilation): Do not call
note_prediction_to_br_prob and note_prediction_to_br_prob
when not optimizing.
Fri Feb 21 23:10:13 CET 2003 Jan Hubicka <jh@suse.cz>
* cfgrtl.c (commit_edge_insertions): Call

2
gcc/builtins.c
View File

@ -3840,7 +3840,7 @@ expand_builtin_expect (arglist, target)
target = expand_expr (exp, target, VOIDmode, EXPAND_NORMAL);
/* Don't bother with expected value notes for integral constants. */
if (GET_CODE (target) != CONST_INT)
if (flag_guess_branch_prob && GET_CODE (target) != CONST_INT)
{
/* We do need to force this into a register so that we can be
moderately sure to be able to correctly interpret the branch

3
gcc/c-semantics.c
View File

@ -864,7 +864,8 @@ expand_stmt (t)
case GOTO_STMT:
/* Emit information for branch prediction. */
if (!GOTO_FAKE_P (t)
&& TREE_CODE (GOTO_DESTINATION (t)) == LABEL_DECL)
&& TREE_CODE (GOTO_DESTINATION (t)) == LABEL_DECL
&& flag_guess_branch_prob)
{
rtx note = emit_note (NULL, NOTE_INSN_PREDICTION);

6
gcc/config/i386/i386.c
View File

@ -12787,9 +12787,11 @@ static const struct builtin_description bdesc_2arg[] =
{ MASK_MMX, CODE_FOR_addv8qi3, "__builtin_ia32_paddb", IX86_BUILTIN_PADDB, 0, 0 },
{ MASK_MMX, CODE_FOR_addv4hi3, "__builtin_ia32_paddw", IX86_BUILTIN_PADDW, 0, 0 },
{ MASK_MMX, CODE_FOR_addv2si3, "__builtin_ia32_paddd", IX86_BUILTIN_PADDD, 0, 0 },
{ MASK_MMX, CODE_FOR_mmx_adddi3, "__builtin_ia32_paddq", IX86_BUILTIN_PADDQ, 0, 0 },
{ MASK_MMX, CODE_FOR_subv8qi3, "__builtin_ia32_psubb", IX86_BUILTIN_PSUBB, 0, 0 },
{ MASK_MMX, CODE_FOR_subv4hi3, "__builtin_ia32_psubw", IX86_BUILTIN_PSUBW, 0, 0 },
{ MASK_MMX, CODE_FOR_subv2si3, "__builtin_ia32_psubd", IX86_BUILTIN_PSUBD, 0, 0 },
{ MASK_MMX, CODE_FOR_mmx_subdi3, "__builtin_ia32_psubq", IX86_BUILTIN_PSUBQ, 0, 0 },
{ MASK_MMX, CODE_FOR_ssaddv8qi3, "__builtin_ia32_paddsb", IX86_BUILTIN_PADDSB, 0, 0 },
{ MASK_MMX, CODE_FOR_ssaddv4hi3, "__builtin_ia32_paddsw", IX86_BUILTIN_PADDSW, 0, 0 },
@ -12910,11 +12912,11 @@ static const struct builtin_description bdesc_2arg[] =
{ MASK_SSE2, CODE_FOR_addv16qi3, "__builtin_ia32_paddb128", IX86_BUILTIN_PADDB128, 0, 0 },
{ MASK_SSE2, CODE_FOR_addv8hi3, "__builtin_ia32_paddw128", IX86_BUILTIN_PADDW128, 0, 0 },
{ MASK_SSE2, CODE_FOR_addv4si3, "__builtin_ia32_paddd128", IX86_BUILTIN_PADDD128, 0, 0 },
{ MASK_SSE2, CODE_FOR_addv4si3, "__builtin_ia32_paddq128", IX86_BUILTIN_PADDQ128, 0, 0 },
{ MASK_SSE2, CODE_FOR_addv2di3, "__builtin_ia32_paddq128", IX86_BUILTIN_PADDQ128, 0, 0 },
{ MASK_SSE2, CODE_FOR_subv16qi3, "__builtin_ia32_psubb128", IX86_BUILTIN_PSUBB128, 0, 0 },
{ MASK_SSE2, CODE_FOR_subv8hi3, "__builtin_ia32_psubw128", IX86_BUILTIN_PSUBW128, 0, 0 },
{ MASK_SSE2, CODE_FOR_subv4si3, "__builtin_ia32_psubd128", IX86_BUILTIN_PSUBD128, 0, 0 },
{ MASK_SSE2, CODE_FOR_subv4si3, "__builtin_ia32_psubq128", IX86_BUILTIN_PSUBQ128, 0, 0 },
{ MASK_SSE2, CODE_FOR_subv2di3, "__builtin_ia32_psubq128", IX86_BUILTIN_PSUBQ128, 0, 0 },
{ MASK_MMX, CODE_FOR_ssaddv16qi3, "__builtin_ia32_paddsb128", IX86_BUILTIN_PADDSB128, 0, 0 },
{ MASK_MMX, CODE_FOR_ssaddv8hi3, "__builtin_ia32_paddsw128", IX86_BUILTIN_PADDSW128, 0, 0 },

2
gcc/config/i386/i386.h
View File

@ -2163,6 +2163,7 @@ enum ix86_builtins
IX86_BUILTIN_PADDB,
IX86_BUILTIN_PADDW,
IX86_BUILTIN_PADDD,
IX86_BUILTIN_PADDQ,
IX86_BUILTIN_PADDSB,
IX86_BUILTIN_PADDSW,
IX86_BUILTIN_PADDUSB,
@ -2170,6 +2171,7 @@ enum ix86_builtins
IX86_BUILTIN_PSUBB,
IX86_BUILTIN_PSUBW,
IX86_BUILTIN_PSUBD,
IX86_BUILTIN_PSUBQ,
IX86_BUILTIN_PSUBSB,
IX86_BUILTIN_PSUBSW,
IX86_BUILTIN_PSUBUSB,

58
gcc/config/i386/i386.md
View File

@ -20242,7 +20242,7 @@
(define_insn "addv8qi3"
[(set (match_operand:V8QI 0 "register_operand" "=y")
(plus:V8QI (match_operand:V8QI 1 "register_operand" "0")
(plus:V8QI (match_operand:V8QI 1 "register_operand" "%0")
(match_operand:V8QI 2 "nonimmediate_operand" "ym")))]
"TARGET_MMX"
"paddb\t{%2, %0|%0, %2}"
@ -20251,7 +20251,7 @@
(define_insn "addv4hi3"
[(set (match_operand:V4HI 0 "register_operand" "=y")
(plus:V4HI (match_operand:V4HI 1 "register_operand" "0")
(plus:V4HI (match_operand:V4HI 1 "register_operand" "%0")
(match_operand:V4HI 2 "nonimmediate_operand" "ym")))]
"TARGET_MMX"
"paddw\t{%2, %0|%0, %2}"
@ -20260,16 +20260,27 @@
(define_insn "addv2si3"
[(set (match_operand:V2SI 0 "register_operand" "=y")
(plus:V2SI (match_operand:V2SI 1 "register_operand" "0")
(plus:V2SI (match_operand:V2SI 1 "register_operand" "%0")
(match_operand:V2SI 2 "nonimmediate_operand" "ym")))]
"TARGET_MMX"
"paddd\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 "register_operand" "%0")
(match_operand:DI 2 "nonimmediate_operand" "ym"))]
UNSPEC_NOP))]
"TARGET_MMX"
"paddq\t{%2, %0|%0, %2}"
[(set_attr "type" "mmxadd")
(set_attr "mode" "DI")])
(define_insn "ssaddv8qi3"
[(set (match_operand:V8QI 0 "register_operand" "=y")
(ss_plus:V8QI (match_operand:V8QI 1 "register_operand" "0")
(ss_plus:V8QI (match_operand:V8QI 1 "register_operand" "%0")
(match_operand:V8QI 2 "nonimmediate_operand" "ym")))]
"TARGET_MMX"
"paddsb\t{%2, %0|%0, %2}"
@ -20278,7 +20289,7 @@
(define_insn "ssaddv4hi3"
[(set (match_operand:V4HI 0 "register_operand" "=y")
(ss_plus:V4HI (match_operand:V4HI 1 "register_operand" "0")
(ss_plus:V4HI (match_operand:V4HI 1 "register_operand" "%0")
(match_operand:V4HI 2 "nonimmediate_operand" "ym")))]
"TARGET_MMX"
"paddsw\t{%2, %0|%0, %2}"
@ -20287,7 +20298,7 @@
(define_insn "usaddv8qi3"
[(set (match_operand:V8QI 0 "register_operand" "=y")
(us_plus:V8QI (match_operand:V8QI 1 "register_operand" "0")
(us_plus:V8QI (match_operand:V8QI 1 "register_operand" "%0")
(match_operand:V8QI 2 "nonimmediate_operand" "ym")))]
"TARGET_MMX"
"paddusb\t{%2, %0|%0, %2}"
@ -20296,7 +20307,7 @@
(define_insn "usaddv4hi3"
[(set (match_operand:V4HI 0 "register_operand" "=y")
(us_plus:V4HI (match_operand:V4HI 1 "register_operand" "0")
(us_plus:V4HI (match_operand:V4HI 1 "register_operand" "%0")
(match_operand:V4HI 2 "nonimmediate_operand" "ym")))]
"TARGET_MMX"
"paddusw\t{%2, %0|%0, %2}"
@ -20330,6 +20341,17 @@
[(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_MMX"
"psubq\t{%2, %0|%0, %2}"
[(set_attr "type" "mmxadd")
(set_attr "mode" "DI")])
(define_insn "sssubv8qi3"
[(set (match_operand:V8QI 0 "register_operand" "=y")
(ss_minus:V8QI (match_operand:V8QI 1 "register_operand" "0")
@ -20433,7 +20455,7 @@
(define_insn "mmx_iordi3"
[(set (match_operand:DI 0 "register_operand" "=y")
(unspec:DI
[(ior:DI (match_operand:DI 1 "register_operand" "0")
[(ior:DI (match_operand:DI 1 "register_operand" "%0")
(match_operand:DI 2 "nonimmediate_operand" "ym"))]
UNSPEC_NOP))]
"TARGET_MMX"
@ -20444,7 +20466,7 @@
(define_insn "mmx_xordi3"
[(set (match_operand:DI 0 "register_operand" "=y")
(unspec:DI
[(xor:DI (match_operand:DI 1 "register_operand" "0")
[(xor:DI (match_operand:DI 1 "register_operand" "%0")
(match_operand:DI 2 "nonimmediate_operand" "ym"))]
UNSPEC_NOP))]
"TARGET_MMX"
@ -20467,7 +20489,7 @@
(define_insn "mmx_anddi3"
[(set (match_operand:DI 0 "register_operand" "=y")
(unspec:DI
[(and:DI (match_operand:DI 1 "register_operand" "0")
[(and:DI (match_operand:DI 1 "register_operand" "%0")
(match_operand:DI 2 "nonimmediate_operand" "ym"))]
UNSPEC_NOP))]
"TARGET_MMX"
@ -21894,7 +21916,7 @@
(define_insn "addv16qi3"
[(set (match_operand:V16QI 0 "register_operand" "=x")
(plus:V16QI (match_operand:V16QI 1 "register_operand" "0")
(plus:V16QI (match_operand:V16QI 1 "register_operand" "%0")
(match_operand:V16QI 2 "nonimmediate_operand" "xm")))]
"TARGET_SSE2"
"paddb\t{%2, %0|%0, %2}"
@ -21903,7 +21925,7 @@
(define_insn "addv8hi3"
[(set (match_operand:V8HI 0 "register_operand" "=x")
(plus:V8HI (match_operand:V8HI 1 "register_operand" "0")
(plus:V8HI (match_operand:V8HI 1 "register_operand" "%0")
(match_operand:V8HI 2 "nonimmediate_operand" "xm")))]
"TARGET_SSE2"
"paddw\t{%2, %0|%0, %2}"
@ -21912,7 +21934,7 @@
(define_insn "addv4si3"
[(set (match_operand:V4SI 0 "register_operand" "=x")
(plus:V4SI (match_operand:V4SI 1 "register_operand" "0")
(plus:V4SI (match_operand:V4SI 1 "register_operand" "%0")
(match_operand:V4SI 2 "nonimmediate_operand" "xm")))]
"TARGET_SSE2"
"paddd\t{%2, %0|%0, %2}"
@ -21921,7 +21943,7 @@
(define_insn "addv2di3"
[(set (match_operand:V2DI 0 "register_operand" "=x")
(plus:V2DI (match_operand:V2DI 1 "register_operand" "0")
(plus:V2DI (match_operand:V2DI 1 "register_operand" "%0")
(match_operand:V2DI 2 "nonimmediate_operand" "xm")))]
"TARGET_SSE2"
"paddq\t{%2, %0|%0, %2}"
@ -21930,7 +21952,7 @@
(define_insn "ssaddv16qi3"
[(set (match_operand:V16QI 0 "register_operand" "=x")
(ss_plus:V16QI (match_operand:V16QI 1 "register_operand" "0")
(ss_plus:V16QI (match_operand:V16QI 1 "register_operand" "%0")
(match_operand:V16QI 2 "nonimmediate_operand" "xm")))]
"TARGET_SSE2"
"paddsb\t{%2, %0|%0, %2}"
@ -21939,7 +21961,7 @@
(define_insn "ssaddv8hi3"
[(set (match_operand:V8HI 0 "register_operand" "=x")
(ss_plus:V8HI (match_operand:V8HI 1 "register_operand" "0")
(ss_plus:V8HI (match_operand:V8HI 1 "register_operand" "%0")
(match_operand:V8HI 2 "nonimmediate_operand" "xm")))]
"TARGET_SSE2"
"paddsw\t{%2, %0|%0, %2}"
@ -21948,7 +21970,7 @@
(define_insn "usaddv16qi3"
[(set (match_operand:V16QI 0 "register_operand" "=x")
(us_plus:V16QI (match_operand:V16QI 1 "register_operand" "0")
(us_plus:V16QI (match_operand:V16QI 1 "register_operand" "%0")
(match_operand:V16QI 2 "nonimmediate_operand" "xm")))]
"TARGET_SSE2"
"paddusb\t{%2, %0|%0, %2}"
@ -21957,7 +21979,7 @@
(define_insn "usaddv8hi3"
[(set (match_operand:V8HI 0 "register_operand" "=x")
(us_plus:V8HI (match_operand:V8HI 1 "register_operand" "0")
(us_plus:V8HI (match_operand:V8HI 1 "register_operand" "%0")
(match_operand:V8HI 2 "nonimmediate_operand" "xm")))]
"TARGET_SSE2"
"paddusw\t{%2, %0|%0, %2}"

14
gcc/config/i386/mmintrin.h
View File

@ -160,6 +160,13 @@ _mm_add_pi32 (__m64 __m1, __m64 __m2)
return (__m64) __builtin_ia32_paddd ((__v2si)__m1, (__v2si)__m2);
}
/* Add the 64-bit values in M1 to the 64-bit values in M2. */
static __inline __m64
_mm_add_si64 (__m64 __m1, __m64 __m2)
{
return (__m64) __builtin_ia32_paddq ((long long)__m1, (long long)__m2);
}
/* Add the 8-bit values in M1 to the 8-bit values in M2 using signed
saturated arithmetic. */
static __inline __m64
@ -213,6 +220,13 @@ _mm_sub_pi32 (__m64 __m1, __m64 __m2)
return (__m64) __builtin_ia32_psubd ((__v2si)__m1, (__v2si)__m2);
}
/* Add the 64-bit values in M1 to the 64-bit values in M2. */
static __inline __m64
_mm_sub_si64 (__m64 __m1, __m64 __m2)
{
return (__m64) __builtin_ia32_psubq ((long long)__m1, (long long)__m2);
}
/* Subtract the 8-bit values in M2 from the 8-bit values in M1 using signed
saturating arithmetic. */
static __inline __m64

12
gcc/config/i386/xmmintrin.h
View File

@ -1619,6 +1619,12 @@ _mm_storel_epi64 (__m128i *__P, __m128i __B)
*(long long *)__P = __builtin_ia32_movdq2q ((__v2di)__B);
}
static __inline __m64
_mm_movepi64_pi64 (__m128i __B)
{
return (__m64) __builtin_ia32_movdq2q ((__v2di)__B);
}
static __inline __m128i
_mm_move_epi64 (__m128i __A)
{
@ -2048,7 +2054,7 @@ _mm_add_epi32 (__m128i __A, __m128i __B)
static __inline __m128i
_mm_add_epi64 (__m128i __A, __m128i __B)
{
return (__m128i)__builtin_ia32_paddq128 ((__v4si)__A, (__v4si)__B);
return (__m128i)__builtin_ia32_paddq128 ((__v2di)__A, (__v2di)__B);
}
static __inline __m128i
@ -2096,7 +2102,7 @@ _mm_sub_epi32 (__m128i __A, __m128i __B)
static __inline __m128i
_mm_sub_epi64 (__m128i __A, __m128i __B)
{
return (__m128i)__builtin_ia32_psubq128 ((__v4si)__A, (__v4si)__B);
return (__m128i)__builtin_ia32_psubq128 ((__v2di)__A, (__v2di)__B);
}
static __inline __m128i
@ -2142,7 +2148,7 @@ _mm_mullo_epi16 (__m128i __A, __m128i __B)
}
static __inline __m64
_mm_mul_pu16 (__m64 __A, __m64 __B)
_mm_mul_su32 (__m64 __A, __m64 __B)
{
return (__m64)__builtin_ia32_pmuludq ((__v2si)__A, (__v2si)__B);
}

2
gcc/predict.c
View File

@ -186,6 +186,8 @@ predict_insn (insn, predictor, probability)
{
if (!any_condjump_p (insn))
abort ();
if (!flag_guess_branch_prob)
return;
REG_NOTES (insn)
= gen_rtx_EXPR_LIST (REG_BR_PRED,

10
gcc/stmt.c
View File

@ -2781,8 +2781,11 @@ expand_continue_loop (whichloop)
/* Emit information for branch prediction. */
rtx note;
note = emit_note (NULL, NOTE_INSN_PREDICTION);
NOTE_PREDICTION (note) = NOTE_PREDICT (PRED_CONTINUE, IS_TAKEN);
if (flag_guess_branch_prob)
{
note = emit_note (NULL, NOTE_INSN_PREDICTION);
NOTE_PREDICTION (note) = NOTE_PREDICT (PRED_CONTINUE, IS_TAKEN);
}
clear_last_expr ();
if (whichloop == 0)
whichloop = loop_stack;
@ -2974,7 +2977,8 @@ expand_value_return (val)
rtx return_reg;
enum br_predictor pred;
if ((pred = return_prediction (val)) != PRED_NO_PREDICTION)
if (flag_guess_branch_prob
&& (pred = return_prediction (val)) != PRED_NO_PREDICTION)
{
/* Emit information for branch prediction. */
rtx note;

12
gcc/toplev.c
View File

@ -2605,9 +2605,12 @@ rest_of_compilation (decl)
delete_unreachable_blocks ();
/* Turn NOTE_INSN_PREDICTIONs into branch predictions. */
timevar_push (TV_BRANCH_PROB);
note_prediction_to_br_prob ();
timevar_pop (TV_BRANCH_PROB);
if (flag_guess_branch_prob)
{
timevar_push (TV_BRANCH_PROB);
note_prediction_to_br_prob ();
timevar_pop (TV_BRANCH_PROB);
}
/* We may have potential sibling or tail recursion sites. Select one
(of possibly multiple) methods of performing the call. */
@ -2688,7 +2691,8 @@ rest_of_compilation (decl)
timevar_push (TV_JUMP);
/* Turn NOTE_INSN_EXPECTED_VALUE into REG_BR_PROB. Do this
before jump optimization switches branch directions. */
expected_value_to_br_prob ();
if (flag_guess_branch_prob)
expected_value_to_br_prob ();
reg_scan (insns, max_reg_num (), 0);
rebuild_jump_labels (insns);