* config/i386/i386.h (SSE_VEC_FLOAT_MODE_P): Remove.
(AVX_FLOAT_MODE_P): Ditto.
(AVX128_VEC_FLOAT_MODE_P): Ditto.
(AVX256_VEC_FLOAT_MODE_P): Ditto.
(AVX_VEC_FLOAT_MODE_P): Ditto.
* config/i386/i386.md (UNSPEC_MASKLOAD): Remove.
(UNSPEC_MASKSTORE): Ditto.
* config/i386/sse.md (<sse>_movmsk<ssemodesuffix><avxmodesuffix>):
Merge from <sse>_movmsk<ssemodesuffix> and
avx_movmsk<ssemodesuffix>256. Use VF mode iterator.
(*sse2_maskmovdqu): Merge with *sse2_maskmovdqu_rex64. Use P mode
iterator.
(avx_maskload<ssemodesuffix><avxmodesuffix>): New expander.
(avx_maskstore<ssemodesuffix><avxmodesuffix>): Ditto.
(*avx_maskmov<ssemodesuffix><avxmodesuffix>): New insn.
testsuite/ChangeLog:
* gcc.target/i386/sse2-maskmovdqu.c: New test.
* gcc.target/i386/avx-vmaskmovdqu.c: Ditto.
From-SVN: r172669
+2011-04-18 Uros Bizjak <ubizjak@gmail.com>
+
+ * config/i386/i386.h (SSE_VEC_FLOAT_MODE_P): Remove.
+ (AVX_FLOAT_MODE_P): Ditto.
+ (AVX128_VEC_FLOAT_MODE_P): Ditto.
+ (AVX256_VEC_FLOAT_MODE_P): Ditto.
+ (AVX_VEC_FLOAT_MODE_P): Ditto.
+ * config/i386/i386.md (UNSPEC_MASKLOAD): Remove.
+ (UNSPEC_MASKSTORE): Ditto.
+ * config/i386/sse.md (<sse>_movmsk<ssemodesuffix><avxmodesuffix>):
+ Merge from <sse>_movmsk<ssemodesuffix> and
+ avx_movmsk<ssemodesuffix>256. Use VF mode iterator.
+ (*sse2_maskmovdqu): Merge with *sse2_maskmovdqu_rex64. Use P mode
+ iterator.
+ (avx_maskload<ssemodesuffix><avxmodesuffix>): New expander.
+ (avx_maskstore<ssemodesuffix><avxmodesuffix>): Ditto.
+ (*avx_maskmov<ssemodesuffix><avxmodesuffix>): New insn.
+
2011-04-18 Jan Hubicka <jh@suse.cz>
* ipa-inline.c (inline_small_functions): Fix pasto in previous patch.
(inline_small_functions): Move program size estimates here;
actually process whole queue even when unit growth has been
met. (to properly compute inline_failed reasons and for the
- case unit size decrease.) Revisit comments on recursive
- inlining.
+ case unit size decrease.) Revisit comments on recursive inlining.
(ipa_inline): Remove unit summary code; first inline hot calls
of functions called once, cold calls next.
(order, nnodes): Remove unused variables.
want_inline_small_function_p.
(cgraph_decide_recursive_inlining): Rename to...
(recursive_inlining): Use can_inline_edge_p and
- want_inline_self_recursive_call_p; simplify and
- remove no longer valid FIXME.
+ want_inline_self_recursive_call_p; simplify and remove no longer
+ valid FIXME.
(cgraph_set_inline_failed): Remove.
(add_new_edges_to_heap): Use can_inline_edge_p and
want_inline_small_function_p.
(cgraph_decide_inlining_of_small_functions): Rename to ...
(inline_small_functions): ... this one; cleanup; use
- can/want predicates; cleanup debug ouput; work edges
- till fibheap is exhausted and do not stop once unit
- growth is reached; remove later loop processing remaining
- edges.
+ can/want predicates; cleanup debug ouput; work edges till fibheap
+ is exhausted and do not stop once unit growth is reached; remove
+ later loop processing remaining edges.
(cgraph_flatten): Rename to ...
(flatten_function): ... this one; use can_inline_edge_p
and can_early_inline_edge_p predicates.
inlining functions called once; simplify the pass.
(cgraph_perform_always_inlining): Rename to ...
(inline_always_inline_functions): ... this one; use
- DECL_DISREGARD_INLINE_LIMITS; use can_inline_edge_p
- predicate
+ DECL_DISREGARD_INLINE_LIMITS; use can_inline_edge_p predicate.
(cgraph_decide_inlining_incrementally): Rename to ...
(early_inline_small_functions): ... this one; simplify
using new predicates; cleanup; make dumps prettier.
(initialize_inline_failed): Move here from cgraph.c.
* tree-sra.c: Include ipa-inline.h.
(ipa_sra_preliminary_function_checks): Update.
- * Makefile.in: (cgraph.o, cgraphbuild.o): Add dependency on
- ipa-inline.h
+ * Makefile.in (cgraph.o, cgraphbuild.o): Add dependency on
+ ipa-inline.h.
2011-04-16 Uros Bizjak <ubizjak@gmail.com>
Eric Weddington <eric.weddington@atmel.com>
Georg-Johann Lay <avr@gjlay.de>
- * config/avr/avr.c: ("insn-codes.h", "optabs.h", "langhooks.h"):
+ * config/avr/avr.c ("insn-codes.h", "optabs.h", "langhooks.h"):
New Includes
(avr_init_builtins, avr_expand_builtin,
avr_expand_delay_cycles, avr_expand_unop_builtin,
2010-08-13 Vladimir Makarov <vmakarov@redhat.com>
- * ira-build.c: (ira_create_object): Remove initialization of
+ * ira-build.c (ira_create_object): Remove initialization of
OBJECT_PROFITABLE_HARD_REGS. Initialize OBJECT_ADD_DATA.
(ira_create_allocno): Remove initialization of
ALLOCNO_MEM_OPTIMIZED_DEST, ALLOCNO_MEM_OPTIMIZED_DEST_P,
ira_object_conflict_iter_cond.
(FOR_EACH_OBJECT_CONFLICT): Don't use ira_object_conflict_iter_next.
- * ira-live.c: (process_single_reg_class_operands): Call
+ * ira-live.c (process_single_reg_class_operands): Call
ira_init_register_move_cost_if_necessary. Use
ira_register_move_cost instead of ira_get_register_move_cost.
2011-03-14 Andreas Tobler <andreast@fgznet.ch>
- * config/rs6000/freebsd.h: (RELOCATABLE_NEEDS_FIXUP): Define in
+ * config/rs6000/freebsd.h (RELOCATABLE_NEEDS_FIXUP): Define in
terms of target_flags_explicit. Adjust copyright year.
* config.gcc: Add FreeBSD PowerPC soft-float libgcc bits.
2011-01-07 Jan Hubicka <jh@suse.cz>
- * doc/invoke.texi: (-flto, -fuse-linker-plugin): Update defaults
+ * doc/invoke.texi (-flto, -fuse-linker-plugin): Update defaults
and no longer claim that gold is required for linker plugin.
* configure: Regenerate.
* gcc.c (PLUGIN_COND): New macro.
#define SSE_FLOAT_MODE_P(MODE) \
((TARGET_SSE && (MODE) == SFmode) || (TARGET_SSE2 && (MODE) == DFmode))
-#define SSE_VEC_FLOAT_MODE_P(MODE) \
- ((TARGET_SSE && (MODE) == V4SFmode) || (TARGET_SSE2 && (MODE) == V2DFmode))
-
-#define AVX_FLOAT_MODE_P(MODE) \
- (TARGET_AVX && ((MODE) == SFmode || (MODE) == DFmode))
-
-#define AVX128_VEC_FLOAT_MODE_P(MODE) \
- (TARGET_AVX && ((MODE) == V4SFmode || (MODE) == V2DFmode))
-
-#define AVX256_VEC_FLOAT_MODE_P(MODE) \
- (TARGET_AVX && ((MODE) == V8SFmode || (MODE) == V4DFmode))
-
-#define AVX_VEC_FLOAT_MODE_P(MODE) \
- (TARGET_AVX && ((MODE) == V4SFmode || (MODE) == V2DFmode \
- || (MODE) == V8SFmode || (MODE) == V4DFmode))
-
#define FMA4_VEC_FLOAT_MODE_P(MODE) \
(TARGET_FMA4 && ((MODE) == V4SFmode || (MODE) == V2DFmode \
|| (MODE) == V8SFmode || (MODE) == V4DFmode))
UNSPEC_VPERMIL
UNSPEC_VPERMIL2
UNSPEC_VPERMIL2F128
- UNSPEC_MASKLOAD
- UNSPEC_MASKSTORE
UNSPEC_CAST
UNSPEC_VTESTP
UNSPEC_VCVTPH2PS
(set_attr "prefix" "orig,vex")
(set_attr "mode" "TI")])
-(define_insn "avx_movmsk<ssemodesuffix>256"
+(define_insn "<sse>_movmsk<ssemodesuffix><avxmodesuffix>"
[(set (match_operand:SI 0 "register_operand" "=r")
(unspec:SI
- [(match_operand:AVX256MODEF2P 1 "register_operand" "x")]
+ [(match_operand:VF 1 "register_operand" "x")]
UNSPEC_MOVMSK))]
- "AVX256_VEC_FLOAT_MODE_P (<MODE>mode)"
- "vmovmsk<ssemodesuffix>\t{%1, %0|%0, %1}"
- [(set_attr "type" "ssecvt")
- (set_attr "prefix" "vex")
- (set_attr "mode" "<MODE>")])
-
-(define_insn "<sse>_movmsk<ssemodesuffix>"
- [(set (match_operand:SI 0 "register_operand" "=r")
- (unspec:SI
- [(match_operand:SSEMODEF2P 1 "register_operand" "x")]
- UNSPEC_MOVMSK))]
- "SSE_VEC_FLOAT_MODE_P (<MODE>mode)"
+ ""
"%vmovmsk<ssemodesuffix>\t{%1, %0|%0, %1}"
[(set_attr "type" "ssemov")
(set_attr "prefix" "maybe_vex")
"TARGET_SSE2")
(define_insn "*sse2_maskmovdqu"
- [(set (mem:V16QI (match_operand:SI 0 "register_operand" "D"))
- (unspec:V16QI [(match_operand:V16QI 1 "register_operand" "x")
- (match_operand:V16QI 2 "register_operand" "x")
- (mem:V16QI (match_dup 0))]
- UNSPEC_MASKMOV))]
- "TARGET_SSE2 && !TARGET_64BIT"
- ;; @@@ check ordering of operands in intel/nonintel syntax
- "%vmaskmovdqu\t{%2, %1|%1, %2}"
- [(set_attr "type" "ssemov")
- (set_attr "prefix_data16" "1")
- ;; The implicit %rdi operand confuses default length_vex computation.
- (set_attr "length_vex" "3")
- (set_attr "prefix" "maybe_vex")
- (set_attr "mode" "TI")])
-
-(define_insn "*sse2_maskmovdqu_rex64"
- [(set (mem:V16QI (match_operand:DI 0 "register_operand" "D"))
+ [(set (mem:V16QI (match_operand:P 0 "register_operand" "D"))
(unspec:V16QI [(match_operand:V16QI 1 "register_operand" "x")
(match_operand:V16QI 2 "register_operand" "x")
(mem:V16QI (match_dup 0))]
UNSPEC_MASKMOV))]
- "TARGET_SSE2 && TARGET_64BIT"
- ;; @@@ check ordering of operands in intel/nonintel syntax
+ "TARGET_SSE2"
"%vmaskmovdqu\t{%2, %1|%1, %2}"
[(set_attr "type" "ssemov")
(set_attr "prefix_data16" "1")
;; The implicit %rdi operand confuses default length_vex computation.
(set (attr "length_vex")
- (symbol_ref ("REGNO (operands[2]) >= FIRST_REX_SSE_REG ? 3 + 1 : 2 + 1")))
+ (symbol_ref ("3 + REX_SSE_REGNO_P (REGNO (operands[2]))")))
(set_attr "prefix" "maybe_vex")
(set_attr "mode" "TI")])
(set_attr "prefix" "vex")
(set_attr "mode" "V8SF")])
-(define_insn "avx_maskload<ssemodesuffix><avxmodesuffix>"
- [(set (match_operand:AVXMODEF2P 0 "register_operand" "=x")
- (unspec:AVXMODEF2P
- [(match_operand:AVXMODEF2P 1 "memory_operand" "m")
- (match_operand:<avxpermvecmode> 2 "register_operand" "x")
+(define_expand "avx_maskload<ssemodesuffix><avxmodesuffix>"
+ [(set (match_operand:VF 0 "register_operand" "")
+ (unspec:VF
+ [(match_operand:<avxpermvecmode> 2 "register_operand" "")
+ (match_operand:VF 1 "memory_operand" "")
(match_dup 0)]
- UNSPEC_MASKLOAD))]
- "TARGET_AVX"
- "vmaskmov<ssemodesuffix>\t{%1, %2, %0|%0, %2, %1}"
- [(set_attr "type" "sselog1")
- (set_attr "prefix_extra" "1")
- (set_attr "prefix" "vex")
- (set_attr "mode" "<MODE>")])
+ UNSPEC_MASKMOV))]
+ "TARGET_AVX")
-(define_insn "avx_maskstore<ssemodesuffix><avxmodesuffix>"
- [(set (match_operand:AVXMODEF2P 0 "memory_operand" "=m")
- (unspec:AVXMODEF2P
- [(match_operand:<avxpermvecmode> 1 "register_operand" "x")
- (match_operand:AVXMODEF2P 2 "register_operand" "x")
+(define_expand "avx_maskstore<ssemodesuffix><avxmodesuffix>"
+ [(set (match_operand:VF 0 "memory_operand" "")
+ (unspec:VF
+ [(match_operand:<avxpermvecmode> 1 "register_operand" "")
+ (match_operand:VF 2 "register_operand" "")
(match_dup 0)]
- UNSPEC_MASKSTORE))]
- "TARGET_AVX"
+ UNSPEC_MASKMOV))]
+ "TARGET_AVX")
+
+(define_insn "*avx_maskmov<ssemodesuffix><avxmodesuffix>"
+ [(set (match_operand:VF 0 "nonimmediate_operand" "=x,m")
+ (unspec:VF
+ [(match_operand:<avxpermvecmode> 1 "register_operand" "x,x")
+ (match_operand:VF 2 "nonimmediate_operand" "m,x")
+ (match_dup 0)]
+ UNSPEC_MASKMOV))]
+ "TARGET_AVX
+ && (REG_P (operands[0]) == MEM_P (operands[2]))"
"vmaskmov<ssemodesuffix>\t{%2, %1, %0|%0, %1, %2}"
[(set_attr "type" "sselog1")
(set_attr "prefix_extra" "1")
+2011-04-18 Uros Bizjak <ubizjak@gmail.com>
+
+ * gcc.target/i386/sse2-maskmovdqu.c: New test.
+ * gcc.target/i386/avx-vmaskmovdqu.c: Ditto.
+
2011-04-18 Tobias Burnus <burnus@net-b.de>
PR fortran/18918
--- /dev/null
+/* { dg-do run } */
+/* { dg-require-effective-target avx } */
+/* { dg-options "-O2 -mavx" } */
+
+#define CHECK_H "avx-check.h"
+#define TEST avx_test
+
+#include "sse2-maskmovdqu.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-require-effective-target sse2 } */
+/* { dg-options "-O2 -msse2" } */
+
+#ifndef CHECK_H
+#define CHECK_H "sse2-check.h"
+#endif
+
+#ifndef TEST
+#define TEST sse2_test
+#endif
+
+#include CHECK_H
+
+#include <emmintrin.h>
+
+#ifndef MASK
+#define MASK 0x7986
+#endif
+
+#define mask_v(pos) (((MASK & (0x1 << (pos))) >> (pos)) << 7)
+
+void static
+TEST (void)
+{
+ __m128i src, mask;
+ char s[16] = { 1,-2,3,-4,5,-6,7,-8,9,-10,11,-12,13,-14,15,-16 };
+ char m[16];
+
+ char u[20] = { 0 };
+ int i;
+
+ for (i = 0; i < 16; i++)
+ m[i] = mask_v (i);
+
+ src = _mm_loadu_si128 ((__m128i *)s);
+ mask = _mm_loadu_si128 ((__m128i *)m);
+
+ _mm_maskmoveu_si128 (src, mask, u+3);
+
+ for (i = 0; i < 16; i++)
+ if (u[i+3] != (m[i] ? s[i] : 0))
+ abort ();
+}