|
 |
0d16b69 |
2007-12-29 Jakub Jelinek <jakub@redhat.com>
|
|
|
0d16b69 |
|
|
|
0d16b69 |
* config/i386/sse.md (sse5_pperm, sse5_pperm_pack_v2di_v4si,
|
|
|
0d16b69 |
sse5_pperm_pack_v4si_v8hi, sse5_pperm_pack_v8hi_v16qi,
|
|
|
0d16b69 |
sse5_perm<mode>): Fix constraints.
|
|
|
0d16b69 |
|
|
|
0d16b69 |
* gcc.target/i386/i386.exp (check_effective_target_sse5): Use __v8hi
|
|
|
0d16b69 |
rather than __v2di type.
|
|
|
0d16b69 |
|
|
|
0d16b69 |
--- gcc/config/i386/sse.md.jj 2007-12-29 20:58:15.000000000 +0100
|
|
|
0d16b69 |
+++ gcc/config/i386/sse.md 2007-12-29 21:12:49.000000000 +0100
|
|
|
0d16b69 |
@@ -8350,13 +8350,13 @@
|
|
|
0d16b69 |
[(set_attr "type" "sseiadd1")])
|
|
|
0d16b69 |
|
|
|
0d16b69 |
;; SSE5 permute instructions
|
|
|
0d16b69 |
(define_insn "sse5_pperm"
|
|
|
0d16b69 |
[(set (match_operand:V16QI 0 "register_operand" "=x,x,x,x")
|
|
|
0d16b69 |
- (unspec:V16QI [(match_operand:V16QI 1 "nonimmediate_operand" "0,0,xm,xm")
|
|
|
0d16b69 |
- (match_operand:V16QI 2 "nonimmediate_operand" "x,xm,0,x")
|
|
|
0d16b69 |
- (match_operand:V16QI 3 "nonimmediate_operand" "xm,x,x,0")]
|
|
|
0d16b69 |
+ (unspec:V16QI [(match_operand:V16QI 1 "nonimmediate_operand" "0,0,x,xm")
|
|
|
0d16b69 |
+ (match_operand:V16QI 2 "nonimmediate_operand" "x,xm,xm,x")
|
|
|
0d16b69 |
+ (match_operand:V16QI 3 "nonimmediate_operand" "xm,x,0,0")]
|
|
|
0d16b69 |
UNSPEC_SSE5_PERMUTE))]
|
|
|
0d16b69 |
"TARGET_SSE5 && ix86_sse5_valid_op_p (operands, insn, 4, true, 1)"
|
|
|
0d16b69 |
"pperm\t{%3, %2, %1, %0|%0, %1, %2, %3}"
|
|
|
0d16b69 |
[(set_attr "type" "sse4arg")
|
|
|
0d16b69 |
(set_attr "mode" "TI")])
|
|
|
0d16b69 |
@@ -8453,52 +8453,52 @@
|
|
|
0d16b69 |
;; SSE5 pack instructions that combine two vectors into a smaller vector
|
|
|
0d16b69 |
(define_insn "sse5_pperm_pack_v2di_v4si"
|
|
|
0d16b69 |
[(set (match_operand:V4SI 0 "register_operand" "=x,x,x,x")
|
|
|
0d16b69 |
(vec_concat:V4SI
|
|
|
0d16b69 |
(truncate:V2SI
|
|
|
0d16b69 |
- (match_operand:V2DI 1 "nonimmediate_operand" "0,0,xm,xm"))
|
|
|
0d16b69 |
+ (match_operand:V2DI 1 "nonimmediate_operand" "0,0,x,xm"))
|
|
|
0d16b69 |
(truncate:V2SI
|
|
|
0d16b69 |
- (match_operand:V2DI 2 "nonimmediate_operand" "x,xm,0,x"))))
|
|
|
0d16b69 |
- (use (match_operand:V16QI 3 "nonimmediate_operand" "xm,x,x,0"))]
|
|
|
0d16b69 |
+ (match_operand:V2DI 2 "nonimmediate_operand" "x,xm,xm,x"))))
|
|
|
0d16b69 |
+ (use (match_operand:V16QI 3 "nonimmediate_operand" "xm,x,0,0"))]
|
|
|
0d16b69 |
"TARGET_SSE5 && ix86_sse5_valid_op_p (operands, insn, 4, true, 1)"
|
|
|
0d16b69 |
"pperm\t{%3, %2, %1, %0|%0, %1, %2, %3}"
|
|
|
0d16b69 |
[(set_attr "type" "sse4arg")
|
|
|
0d16b69 |
(set_attr "mode" "TI")])
|
|
|
0d16b69 |
|
|
|
0d16b69 |
(define_insn "sse5_pperm_pack_v4si_v8hi"
|
|
|
0d16b69 |
[(set (match_operand:V8HI 0 "register_operand" "=x,x,x,x")
|
|
|
0d16b69 |
(vec_concat:V8HI
|
|
|
0d16b69 |
(truncate:V4HI
|
|
|
0d16b69 |
- (match_operand:V4SI 1 "nonimmediate_operand" "0,0,xm,xm"))
|
|
|
0d16b69 |
+ (match_operand:V4SI 1 "nonimmediate_operand" "0,0,x,xm"))
|
|
|
0d16b69 |
(truncate:V4HI
|
|
|
0d16b69 |
- (match_operand:V4SI 2 "nonimmediate_operand" "x,xm,0,x"))))
|
|
|
0d16b69 |
- (use (match_operand:V16QI 3 "nonimmediate_operand" "xm,x,x,0"))]
|
|
|
0d16b69 |
+ (match_operand:V4SI 2 "nonimmediate_operand" "x,xm,xm,x"))))
|
|
|
0d16b69 |
+ (use (match_operand:V16QI 3 "nonimmediate_operand" "xm,x,0,0"))]
|
|
|
0d16b69 |
"TARGET_SSE5 && ix86_sse5_valid_op_p (operands, insn, 4, true, 1)"
|
|
|
0d16b69 |
"pperm\t{%3, %2, %1, %0|%0, %1, %2, %3}"
|
|
|
0d16b69 |
[(set_attr "type" "sse4arg")
|
|
|
0d16b69 |
(set_attr "mode" "TI")])
|
|
|
0d16b69 |
|
|
|
0d16b69 |
(define_insn "sse5_pperm_pack_v8hi_v16qi"
|
|
|
0d16b69 |
[(set (match_operand:V16QI 0 "register_operand" "=x,x,x,x")
|
|
|
0d16b69 |
(vec_concat:V16QI
|
|
|
0d16b69 |
(truncate:V8QI
|
|
|
0d16b69 |
- (match_operand:V8HI 1 "nonimmediate_operand" "0,0,xm,xm"))
|
|
|
0d16b69 |
+ (match_operand:V8HI 1 "nonimmediate_operand" "0,0,x,xm"))
|
|
|
0d16b69 |
(truncate:V8QI
|
|
|
0d16b69 |
- (match_operand:V8HI 2 "nonimmediate_operand" "x,xm,0,x"))))
|
|
|
0d16b69 |
- (use (match_operand:V16QI 3 "nonimmediate_operand" "xm,x,x,0"))]
|
|
|
0d16b69 |
+ (match_operand:V8HI 2 "nonimmediate_operand" "x,xm,xm,x"))))
|
|
|
0d16b69 |
+ (use (match_operand:V16QI 3 "nonimmediate_operand" "xm,x,0,0"))]
|
|
|
0d16b69 |
"TARGET_SSE5 && ix86_sse5_valid_op_p (operands, insn, 4, true, 1)"
|
|
|
0d16b69 |
"pperm\t{%3, %2, %1, %0|%0, %1, %2, %3}"
|
|
|
0d16b69 |
[(set_attr "type" "sse4arg")
|
|
|
0d16b69 |
(set_attr "mode" "TI")])
|
|
|
0d16b69 |
|
|
|
0d16b69 |
;; Floating point permutation (permps, permpd)
|
|
|
0d16b69 |
(define_insn "sse5_perm<mode>"
|
|
|
0d16b69 |
[(set (match_operand:SSEMODEF2P 0 "register_operand" "=x,x,x,x")
|
|
|
0d16b69 |
(unspec:SSEMODEF2P
|
|
|
0d16b69 |
- [(match_operand:SSEMODEF2P 1 "nonimmediate_operand" "0,0,xm,xm")
|
|
|
0d16b69 |
- (match_operand:SSEMODEF2P 2 "nonimmediate_operand" "x,xm,0,x")
|
|
|
0d16b69 |
- (match_operand:V16QI 3 "nonimmediate_operand" "xm,x,x,0")]
|
|
|
0d16b69 |
+ [(match_operand:SSEMODEF2P 1 "nonimmediate_operand" "0,0,x,xm")
|
|
|
0d16b69 |
+ (match_operand:SSEMODEF2P 2 "nonimmediate_operand" "x,xm,xm,x")
|
|
|
0d16b69 |
+ (match_operand:V16QI 3 "nonimmediate_operand" "xm,x,0,0")]
|
|
|
0d16b69 |
UNSPEC_SSE5_PERMUTE))]
|
|
|
0d16b69 |
"TARGET_SSE5 && ix86_sse5_valid_op_p (operands, insn, 4, true, 1)"
|
|
|
0d16b69 |
"perm<ssemodesuffixf4>\t{%3, %2, %1, %0|%0, %1, %2, %3}"
|
|
|
0d16b69 |
[(set_attr "type" "sse4arg")
|
|
|
0d16b69 |
(set_attr "mode" "<MODE>")])
|
|
|
0d16b69 |
--- gcc/testsuite/gcc.target/i386/i386.exp.jj 2007-09-14 11:54:26.000000000 +0200
|
|
|
0d16b69 |
+++ gcc/testsuite/gcc.target/i386/i386.exp 2007-12-29 21:42:08.000000000 +0100
|
|
|
0d16b69 |
@@ -68,13 +68,13 @@ proc check_effective_target_sse4a { } {
|
|
|
0d16b69 |
proc check_effective_target_sse5 { } {
|
|
|
0d16b69 |
return [check_no_compiler_messages sse5 object {
|
|
|
0d16b69 |
typedef long long __m128i __attribute__ ((__vector_size__ (16)));
|
|
|
0d16b69 |
- typedef long long __v2di __attribute__ ((__vector_size__ (16)));
|
|
|
0d16b69 |
+ typedef short __v8hi __attribute__ ((__vector_size__ (16)));
|
|
|
0d16b69 |
|
|
|
0d16b69 |
__m128i _mm_maccs_epi16(__m128i __A, __m128i __B, __m128i __C)
|
|
|
0d16b69 |
{
|
|
|
0d16b69 |
- return (__m128i) __builtin_ia32_pmacssww ((__v2di)__A,
|
|
|
0d16b69 |
- (__v2di)__B,
|
|
|
0d16b69 |
- (__v2di)__C);
|
|
|
0d16b69 |
+ return (__m128i) __builtin_ia32_pmacssww ((__v8hi)__A,
|
|
|
0d16b69 |
+ (__v8hi)__B,
|
|
|
0d16b69 |
+ (__v8hi)__C);
|
|
|
0d16b69 |
}
|
|
|
0d16b69 |
} "-O2 -msse5" ]
|
|
|
0d16b69 |
}
|