[PATCH] alignaddr and edge VIS intrinsic refinements
David Miller
davem@davemloft.net
Tue Sep 20 18:42:00 GMT 2011
We provided the big-endian 'alignaddr' but not the little-endian
version 'alignaddrl', add it.
The edge operations are meant to operate on addresses and return
a very small integer bitmask so use Pmode and return SImode.
Update documentation and intrinsic header, as needed.
Committed to trunk.
gcc/
* config/sparc/sparc.md (UNSPEC_ALIGNADDRL): New unspec.
(aligneddrl<P:mode>_vis): New pattern.
(edge8_vis, edge8l_vis, edge16_vis, edge16l_vis, edge32_vis,
edge32l_vis): Adjust to take Pmode arguments, and return SImode.
* config/sparc/sparc.c (sparc_vis_init_builtins): Handle new
alignaddrl insn, and adjust edge operations for updated types.
* config/sparc/visintrin.h: Likewise.
* doc/extend.texi: Make typing in VIS documentation match reality.
diff --git a/gcc/config/sparc/sparc.c b/gcc/config/sparc/sparc.c
index 6ccc97b..68b5961 100644
--- a/gcc/config/sparc/sparc.c
+++ b/gcc/config/sparc/sparc.c
@@ -9149,6 +9149,9 @@ sparc_vis_init_builtins (void)
tree ptr_ftype_ptr_di = build_function_type_list (ptr_type_node,
ptr_type_node,
intDI_type_node, 0);
+ tree si_ftype_ptr_ptr = build_function_type_list (intSI_type_node,
+ ptr_type_node,
+ ptr_type_node, 0);
/* Packing and expanding vectors. */
def_builtin ("__builtin_vis_fpack16", CODE_FOR_fpack16_vis, v4qi_ftype_v4hi);
@@ -9186,29 +9189,55 @@ sparc_vis_init_builtins (void)
def_builtin ("__builtin_vis_faligndatadi", CODE_FOR_faligndatadi_vis,
di_ftype_di_di);
if (TARGET_ARCH64)
- def_builtin ("__builtin_vis_alignaddr", CODE_FOR_alignaddrdi_vis,
- ptr_ftype_ptr_di);
+ {
+ def_builtin ("__builtin_vis_alignaddr", CODE_FOR_alignaddrdi_vis,
+ ptr_ftype_ptr_di);
+ def_builtin ("__builtin_vis_alignaddrl", CODE_FOR_alignaddrldi_vis,
+ ptr_ftype_ptr_di);
+ }
else
- def_builtin ("__builtin_vis_alignaddr", CODE_FOR_alignaddrsi_vis,
- ptr_ftype_ptr_si);
+ {
+ def_builtin ("__builtin_vis_alignaddr", CODE_FOR_alignaddrsi_vis,
+ ptr_ftype_ptr_si);
+ def_builtin ("__builtin_vis_alignaddrl", CODE_FOR_alignaddrlsi_vis,
+ ptr_ftype_ptr_si);
+ }
/* Pixel distance. */
def_builtin ("__builtin_vis_pdist", CODE_FOR_pdist_vis,
di_ftype_v8qi_v8qi_di);
/* Edge handling. */
- def_builtin ("__builtin_vis_edge8", CODE_FOR_edge8_vis,
- di_ftype_di_di);
- def_builtin ("__builtin_vis_edge8l", CODE_FOR_edge8l_vis,
- di_ftype_di_di);
- def_builtin ("__builtin_vis_edge16", CODE_FOR_edge16_vis,
- di_ftype_di_di);
- def_builtin ("__builtin_vis_edge16l", CODE_FOR_edge16l_vis,
- di_ftype_di_di);
- def_builtin ("__builtin_vis_edge32", CODE_FOR_edge32_vis,
- di_ftype_di_di);
- def_builtin ("__builtin_vis_edge32l", CODE_FOR_edge32l_vis,
- di_ftype_di_di);
+ if (TARGET_ARCH64)
+ {
+ def_builtin ("__builtin_vis_edge8", CODE_FOR_edge8di_vis,
+ si_ftype_ptr_ptr);
+ def_builtin ("__builtin_vis_edge8l", CODE_FOR_edge8ldi_vis,
+ si_ftype_ptr_ptr);
+ def_builtin ("__builtin_vis_edge16", CODE_FOR_edge16di_vis,
+ si_ftype_ptr_ptr);
+ def_builtin ("__builtin_vis_edge16l", CODE_FOR_edge16ldi_vis,
+ si_ftype_ptr_ptr);
+ def_builtin ("__builtin_vis_edge32", CODE_FOR_edge32di_vis,
+ si_ftype_ptr_ptr);
+ def_builtin ("__builtin_vis_edge32l", CODE_FOR_edge32ldi_vis,
+ si_ftype_ptr_ptr);
+ }
+ else
+ {
+ def_builtin ("__builtin_vis_edge8", CODE_FOR_edge8si_vis,
+ si_ftype_ptr_ptr);
+ def_builtin ("__builtin_vis_edge8l", CODE_FOR_edge8lsi_vis,
+ si_ftype_ptr_ptr);
+ def_builtin ("__builtin_vis_edge16", CODE_FOR_edge16si_vis,
+ si_ftype_ptr_ptr);
+ def_builtin ("__builtin_vis_edge16l", CODE_FOR_edge16lsi_vis,
+ si_ftype_ptr_ptr);
+ def_builtin ("__builtin_vis_edge32", CODE_FOR_edge32si_vis,
+ si_ftype_ptr_ptr);
+ def_builtin ("__builtin_vis_edge32l", CODE_FOR_edge32lsi_vis,
+ si_ftype_ptr_ptr);
+ }
}
/* Handle TARGET_EXPAND_BUILTIN target hook.
diff --git a/gcc/config/sparc/sparc.md b/gcc/config/sparc/sparc.md
index 483dea1..1fb59cc 100644
--- a/gcc/config/sparc/sparc.md
+++ b/gcc/config/sparc/sparc.md
@@ -66,6 +66,7 @@
(UNSPEC_EDGE16L 54)
(UNSPEC_EDGE32 55)
(UNSPEC_EDGE32L 56)
+ (UNSPEC_ALIGNADDRL 57)
(UNSPEC_SP_SET 60)
(UNSPEC_SP_TEST 61)
@@ -7798,6 +7799,14 @@
"TARGET_VIS"
"alignaddr\t%r1, %r2, %0")
+(define_insn "alignaddrl<P:mode>_vis"
+ [(set (match_operand:P 0 "register_operand" "=r")
+ (unspec:P [(match_operand:P 1 "register_or_zero_operand" "rJ")
+ (match_operand:P 2 "register_or_zero_operand" "rJ")]
+ UNSPEC_ALIGNADDRL))]
+ "TARGET_VIS"
+ "alignaddrl\t%r1, %r2, %0")
+
(define_insn "pdist_vis"
[(set (match_operand:DI 0 "register_operand" "=e")
(unspec:DI [(match_operand:V8QI 1 "register_operand" "e")
@@ -7811,68 +7820,68 @@
;; Edge instructions produce condition codes equivalent to a 'subcc'
;; with the same operands.
-(define_insn "edge8_vis"
- [(set (reg:CCX_NOOV 100)
- (compare:CCX_NOOV (minus:DI (match_operand:DI 1 "register_operand" "rJ")
- (match_operand:DI 2 "register_operand" "rJ"))
- (const_int 0)))
- (set (match_operand:DI 0 "register_operand" "=r")
- (unspec:DI [(match_dup 1) (match_dup 2)] UNSPEC_EDGE8))]
+(define_insn "edge8<P:mode>_vis"
+ [(set (reg:CC_NOOV 100)
+ (compare:CC_NOOV (minus:P (match_operand:P 1 "register_operand" "rJ")
+ (match_operand:P 2 "register_operand" "rJ"))
+ (const_int 0)))
+ (set (match_operand:SI 0 "register_operand" "=r")
+ (unspec:SI [(match_dup 1) (match_dup 2)] UNSPEC_EDGE8))]
"TARGET_VIS"
"edge8\t%r1, %r2, %0"
[(set_attr "type" "edge")])
-(define_insn "edge8l_vis"
- [(set (reg:CCX_NOOV 100)
- (compare:CCX_NOOV (minus:DI (match_operand:DI 1 "register_operand" "rJ")
- (match_operand:DI 2 "register_operand" "rJ"))
- (const_int 0)))
- (set (match_operand:DI 0 "register_operand" "=r")
- (unspec:DI [(match_dup 1) (match_dup 2)] UNSPEC_EDGE8L))]
+(define_insn "edge8l<P:mode>_vis"
+ [(set (reg:CC_NOOV 100)
+ (compare:CC_NOOV (minus:P (match_operand:P 1 "register_operand" "rJ")
+ (match_operand:P 2 "register_operand" "rJ"))
+ (const_int 0)))
+ (set (match_operand:SI 0 "register_operand" "=r")
+ (unspec:SI [(match_dup 1) (match_dup 2)] UNSPEC_EDGE8L))]
"TARGET_VIS"
"edge8l\t%r1, %r2, %0"
[(set_attr "type" "edge")])
-(define_insn "edge16_vis"
- [(set (reg:CCX_NOOV 100)
- (compare:CCX_NOOV (minus:DI (match_operand:DI 1 "register_operand" "rJ")
- (match_operand:DI 2 "register_operand" "rJ"))
- (const_int 0)))
- (set (match_operand:DI 0 "register_operand" "=r")
- (unspec:DI [(match_dup 1) (match_dup 2)] UNSPEC_EDGE16))]
+(define_insn "edge16<P:mode>_vis"
+ [(set (reg:CC_NOOV 100)
+ (compare:CC_NOOV (minus:P (match_operand:P 1 "register_operand" "rJ")
+ (match_operand:P 2 "register_operand" "rJ"))
+ (const_int 0)))
+ (set (match_operand:SI 0 "register_operand" "=r")
+ (unspec:SI [(match_dup 1) (match_dup 2)] UNSPEC_EDGE16))]
"TARGET_VIS"
"edge16\t%r1, %r2, %0"
[(set_attr "type" "edge")])
-(define_insn "edge16l_vis"
- [(set (reg:CCX_NOOV 100)
- (compare:CCX_NOOV (minus:DI (match_operand:DI 1 "register_operand" "rJ")
- (match_operand:DI 2 "register_operand" "rJ"))
- (const_int 0)))
- (set (match_operand:DI 0 "register_operand" "=r")
- (unspec:DI [(match_dup 1) (match_dup 2)] UNSPEC_EDGE16L))]
+(define_insn "edge16l<P:mode>_vis"
+ [(set (reg:CC_NOOV 100)
+ (compare:CC_NOOV (minus:P (match_operand:P 1 "register_operand" "rJ")
+ (match_operand:P 2 "register_operand" "rJ"))
+ (const_int 0)))
+ (set (match_operand:SI 0 "register_operand" "=r")
+ (unspec:SI [(match_dup 1) (match_dup 2)] UNSPEC_EDGE16L))]
"TARGET_VIS"
"edge16l\t%r1, %r2, %0"
[(set_attr "type" "edge")])
-(define_insn "edge32_vis"
- [(set (reg:CCX_NOOV 100)
- (compare:CCX_NOOV (minus:DI (match_operand:DI 1 "register_operand" "rJ")
- (match_operand:DI 2 "register_operand" "rJ"))
- (const_int 0)))
- (set (match_operand:DI 0 "register_operand" "=r")
- (unspec:DI [(match_dup 1) (match_dup 2)] UNSPEC_EDGE32))]
+(define_insn "edge32<P:mode>_vis"
+ [(set (reg:CC_NOOV 100)
+ (compare:CC_NOOV (minus:P (match_operand:P 1 "register_operand" "rJ")
+ (match_operand:P 2 "register_operand" "rJ"))
+ (const_int 0)))
+ (set (match_operand:SI 0 "register_operand" "=r")
+ (unspec:SI [(match_dup 1) (match_dup 2)] UNSPEC_EDGE32))]
"TARGET_VIS"
"edge32\t%r1, %r2, %0"
[(set_attr "type" "edge")])
-(define_insn "edge32l_vis"
- [(set (reg:CCX_NOOV 100)
- (compare:CCX_NOOV (minus:DI (match_operand:DI 1 "register_operand" "rJ")
- (match_operand:DI 2 "register_operand" "rJ"))
- (const_int 0)))
- (set (match_operand:DI 0 "register_operand" "=r")
- (unspec:DI [(match_dup 1) (match_dup 2)] UNSPEC_EDGE32L))]
+(define_insn "edge32l<P:mode>_vis"
+ [(set (reg:CC_NOOV 100)
+ (compare:CC_NOOV (minus:P (match_operand:P 1 "register_operand" "rJ")
+ (match_operand:P 2 "register_operand" "rJ"))
+ (const_int 0)))
+ (set (match_operand:SI 0 "register_operand" "=r")
+ (unspec:SI [(match_dup 1) (match_dup 2)] UNSPEC_EDGE32L))]
"TARGET_VIS"
"edge32l\t%r1, %r2, %0"
[(set_attr "type" "edge")])
diff --git a/gcc/config/sparc/visintrin.h b/gcc/config/sparc/visintrin.h
index 25d7bab..1b31451 100644
--- a/gcc/config/sparc/visintrin.h
+++ b/gcc/config/sparc/visintrin.h
@@ -38,6 +38,13 @@ __vis_alignaddr (void *__A, long __B)
return __builtin_vis_alignaddr (__A, __B);
}
+extern __inline void *
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+__vis_alignaddrl (void *__A, long __B)
+{
+ return __builtin_vis_alignaddrl (__A, __B);
+}
+
extern __inline __i64
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
__vis_faligndatadi (__i64 __A, __i64 __B)
@@ -157,44 +164,44 @@ __vis_pdist (__v8qi __A, __v8qi __B, __i64 __C)
return __builtin_vis_pdist (__A, __B, __C);
}
-extern __inline __i64
+extern __inline int
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
-__vis_edge8 (__i64 __A, __i64 __B)
+__vis_edge8 (void *__A, void *__B)
{
return __builtin_vis_edge8 (__A, __B);
}
-extern __inline __i64
+extern __inline int
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
-__vis_edge8l (__i64 __A, __i64 __B)
+__vis_edge8l (void *__A, void *__B)
{
return __builtin_vis_edge8l (__A, __B);
}
-extern __inline __i64
+extern __inline int
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
-__vis_edge16 (__i64 __A, __i64 __B)
+__vis_edge16 (void *__A, void *__B)
{
return __builtin_vis_edge16 (__A, __B);
}
-extern __inline __i64
+extern __inline int
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
-__vis_edge16l (__i64 __A, __i64 __B)
+__vis_edge16l (void *__A, void *__B)
{
return __builtin_vis_edge16l (__A, __B);
}
-extern __inline __i64
+extern __inline int
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
-__vis_edge32 (__i64 __A, __i64 __B)
+__vis_edge32 (void *__A, void *__B)
{
return __builtin_vis_edge32 (__A, __B);
}
-extern __inline __i64
+extern __inline int
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
-__vis_edge32l (__i64 __A, __i64 __B)
+__vis_edge32l (void *__A, void *__B)
{
return __builtin_vis_edge32l (__A, __B);
}
diff --git a/gcc/doc/extend.texi b/gcc/doc/extend.texi
index 75f4874..3e6e05e 100644
--- a/gcc/doc/extend.texi
+++ b/gcc/doc/extend.texi
@@ -12932,10 +12932,11 @@ switch, the VIS extension is exposed as the following built-in functions:
typedef int v2si __attribute__ ((vector_size (8)));
typedef short v4hi __attribute__ ((vector_size (8)));
typedef short v2hi __attribute__ ((vector_size (4)));
-typedef char v8qi __attribute__ ((vector_size (8)));
-typedef char v4qi __attribute__ ((vector_size (4)));
+typedef unsigned char v8qi __attribute__ ((vector_size (8)));
+typedef unsigned char v4qi __attribute__ ((vector_size (4)));
void * __builtin_vis_alignaddr (void *, long);
+void * __builtin_vis_alignaddrl (void *, long);
int64_t __builtin_vis_faligndatadi (int64_t, int64_t);
v2si __builtin_vis_faligndatav2si (v2si, v2si);
v4hi __builtin_vis_faligndatav4hi (v4si, v4si);
@@ -12944,26 +12945,26 @@ v8qi __builtin_vis_faligndatav8qi (v8qi, v8qi);
v4hi __builtin_vis_fexpand (v4qi);
v4hi __builtin_vis_fmul8x16 (v4qi, v4hi);
-v4hi __builtin_vis_fmul8x16au (v4qi, v4hi);
-v4hi __builtin_vis_fmul8x16al (v4qi, v4hi);
+v4hi __builtin_vis_fmul8x16au (v4qi, v2hi);
+v4hi __builtin_vis_fmul8x16al (v4qi, v2hi);
v4hi __builtin_vis_fmul8sux16 (v8qi, v4hi);
v4hi __builtin_vis_fmul8ulx16 (v8qi, v4hi);
v2si __builtin_vis_fmuld8sux16 (v4qi, v2hi);
v2si __builtin_vis_fmuld8ulx16 (v4qi, v2hi);
v4qi __builtin_vis_fpack16 (v4hi);
-v8qi __builtin_vis_fpack32 (v2si, v2si);
+v8qi __builtin_vis_fpack32 (v2si, v8qi);
v2hi __builtin_vis_fpackfix (v2si);
v8qi __builtin_vis_fpmerge (v4qi, v4qi);
int64_t __builtin_vis_pdist (v8qi, v8qi, int64_t);
-int64_t __builtin_vis_edge8 (int64_t, int64_t);
-int64_t __builtin_vis_edge8l (int64_t, int64_t);
-int64_t __builtin_vis_edge16 (int64_t, int64_t);
-int64_t __builtin_vis_edge16l (int64_t, int64_t);
-int64_t __builtin_vis_edge32 (int64_t, int64_t);
-int64_t __builtin_vis_edge32l (int64_t, int64_t);
+int __builtin_vis_edge8 (void *, void *);
+int __builtin_vis_edge8l (void *, void *);
+int __builtin_vis_edge16 (void *, void *);
+int __builtin_vis_edge16l (void *, void *);
+int __builtin_vis_edge32 (void *, void *);
+int __builtin_vis_edge32l (void *, void *);
@end smallexample
@node SPU Built-in Functions
More information about the Gcc-patches
mailing list