This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
[PATCH][AArch64] Vector shift by 64 fix
- From: Alex Velenko <Alex dot Velenko at arm dot com>
- To: "gcc-patches at gcc dot gnu dot org" <gcc-patches at gcc dot gnu dot org>
- Cc: Marcus dot Shawcroft at arm dot com
- Date: Mon, 06 Jan 2014 11:52:41 +0000
- Subject: [PATCH][AArch64] Vector shift by 64 fix
- Authentication-results: sourceware.org; auth=none
Hi,
This patch fixes vector shift by 64 behavior to meet reference
manual expectations. Testcase included to check that expectations
are now met. No regressions found.
Is patch OK?
Thanks,
Alex
2014-01-06 Alex Velenko <Alex.Velenko@arm.com>
gcc/
* config/aarch64/aarch64-simd-builtins.def (ashr): DI mode removed.
(ashr_simd): New builtin handling DI mode.
* config/aarch64/aarch64-simd.md (aarch64_ashr_simddi): New pattern.
(aarch64_sshr_simddi): New match pattern.
* config/aarch64/arm_neon.h (vshr_n_s32): Builtin call modified.
(vshrd_n_s64): Likewise.
* config/aarch64/predicates.md (aarch64_shift_imm64_di): New predicate.
gcc/testsuite/
* gcc.target/aarch64/sshr64_1.c: New testcase.
diff --git a/gcc/config/aarch64/aarch64-simd-builtins.def b/gcc/config/aarch64/aarch64-simd-builtins.def
index 1dc3c1fe33fdb8148d2ff9c7198e4d85d5dac5d7..1e88661fd2f0f756ce1427681c843fc0783ab6a2 100644
--- a/gcc/config/aarch64/aarch64-simd-builtins.def
+++ b/gcc/config/aarch64/aarch64-simd-builtins.def
@@ -189,7 +189,8 @@
BUILTIN_VSDQ_I_DI (BINOP, srshl, 0)
BUILTIN_VSDQ_I_DI (BINOP, urshl, 0)
- BUILTIN_VSDQ_I_DI (SHIFTIMM, ashr, 3)
+ BUILTIN_VDQ_I (SHIFTIMM, ashr, 3)
+ VAR1 (SHIFTIMM, ashr_simd, 0, di)
BUILTIN_VSDQ_I_DI (SHIFTIMM, lshr, 3)
/* Implemented by aarch64_<sur>shr_n<mode>. */
BUILTIN_VSDQ_I_DI (SHIFTIMM, srshr_n, 0)
diff --git a/gcc/config/aarch64/aarch64-simd.md b/gcc/config/aarch64/aarch64-simd.md
index 158b3dca6da12322de0af80d35f593039d716de6..839186a5e3e3363973186d68aeed6fbaf7f0dfea 100644
--- a/gcc/config/aarch64/aarch64-simd.md
+++ b/gcc/config/aarch64/aarch64-simd.md
@@ -668,6 +668,32 @@
DONE;
})
+;; DI vector shift
+(define_expand "aarch64_ashr_simddi"
+ [(match_operand:DI 0 "register_operand" "=w")
+ (match_operand:DI 1 "register_operand" "w")
+ (match_operand:QI 2 "aarch64_shift_imm64_di" "")]
+ "TARGET_SIMD"
+ {
+ if (INTVAL (operands[2]) == 64)
+ emit_insn (gen_aarch64_sshr_simddi (operands[0], operands[1]));
+ else
+ emit_insn (gen_ashrdi3 (operands[0], operands[1], operands[2]));
+ DONE;
+ }
+)
+
+;; SIMD shift by 64. This pattern is a special case as standard pattern does
+;; not handle NEON shifts by 64.
+(define_insn "aarch64_sshr_simddi"
+ [(set (match_operand:DI 0 "register_operand" "=w")
+ (unspec:DI
+ [(match_operand:DI 1 "register_operand" "w")] UNSPEC_SSHR64))]
+ "TARGET_SIMD"
+ "sshr\t%d0, %d1, 64"
+ [(set_attr "type" "neon_shift_imm")]
+)
+
(define_expand "vlshr<mode>3"
[(match_operand:VQ_S 0 "register_operand" "")
(match_operand:VQ_S 1 "register_operand" "")
diff --git a/gcc/config/aarch64/aarch64.md b/gcc/config/aarch64/aarch64.md
index 8b3dbd7550e8e9037de1a1384276bee28d21cb3d..130a11c0231c32440573276fd78e62b6f019d302 100644
--- a/gcc/config/aarch64/aarch64.md
+++ b/gcc/config/aarch64/aarch64.md
@@ -92,6 +92,7 @@
UNSPEC_SISD_SSHL
UNSPEC_SISD_USHL
UNSPEC_SSHL_2S
+ UNSPEC_SSHR64
UNSPEC_ST2
UNSPEC_ST3
UNSPEC_ST4
diff --git a/gcc/config/aarch64/arm_neon.h b/gcc/config/aarch64/arm_neon.h
index 03549bd7a27cccb14ed8cdce91cbd4e4278c273f..64012775b3fa7d174af1472f73aadf4174d0d291 100644
--- a/gcc/config/aarch64/arm_neon.h
+++ b/gcc/config/aarch64/arm_neon.h
@@ -23235,7 +23235,7 @@ vshr_n_s32 (int32x2_t __a, const int __b)
__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
vshr_n_s64 (int64x1_t __a, const int __b)
{
- return (int64x1_t) __builtin_aarch64_ashrdi (__a, __b);
+ return (int64x1_t) __builtin_aarch64_ashr_simddi (__a, __b);
}
__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
@@ -23313,7 +23313,7 @@ vshrq_n_u64 (uint64x2_t __a, const int __b)
__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
vshrd_n_s64 (int64x1_t __a, const int __b)
{
- return (int64x1_t) __builtin_aarch64_ashrdi (__a, __b);
+ return (int64x1_t) __builtin_aarch64_ashr_simddi (__a, __b);
}
__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
diff --git a/gcc/config/aarch64/predicates.md b/gcc/config/aarch64/predicates.md
index dbc90826665d19a6ac6131918efb2c8a32bd1f04..9538107a5c148408f5c6e8e37aeef92aa5be0856 100644
--- a/gcc/config/aarch64/predicates.md
+++ b/gcc/config/aarch64/predicates.md
@@ -86,6 +86,10 @@
(and (match_code "const_int")
(match_test "(unsigned HOST_WIDE_INT) INTVAL (op) < 64")))
+(define_predicate "aarch64_shift_imm64_di"
+ (and (match_code "const_int")
+ (match_test "(unsigned HOST_WIDE_INT) INTVAL (op) <= 64")))
+
(define_predicate "aarch64_reg_or_shift_imm_si"
(ior (match_operand 0 "register_operand")
(match_operand 0 "aarch64_shift_imm_si")))
diff --git a/gcc/testsuite/gcc.target/aarch64/sshr64_1.c b/gcc/testsuite/gcc.target/aarch64/sshr64_1.c
new file mode 100644
index 0000000000000000000000000000000000000000..89c6096ad3934d1c42fac2c8fba6eba6170762da
--- /dev/null
+++ b/gcc/testsuite/gcc.target/aarch64/sshr64_1.c
@@ -0,0 +1,115 @@
+/* Test SIMD shift works correctly. */
+/* { dg-do run } */
+/* { dg-options "-O3 --save-temps" } */
+
+#include "arm_neon.h"
+
+extern void abort (void);
+
+int __attribute__ ((noinline))
+test_sshr64 ()
+{
+ int64x1_t arg;
+ int64x1_t result;
+ int64_t got;
+ int64_t exp;
+ arg = vcreate_s64 (0x0000000080000000);
+ result = vshr_n_s64 (arg, 64);
+ got = vget_lane_s64 (result, 0);
+ exp = 0;
+ /* Expect: "result" = 0000000000000000. */
+ if (exp != got)
+ return 1;
+ return 0;
+}
+
+int __attribute__ ((noinline))
+test_sshr64_neg ()
+{
+ int64x1_t arg;
+ int64x1_t result;
+ int64_t got;
+ int64_t exp;
+ arg = vcreate_s64 (0xffffffff80000000);
+ result = vshr_n_s64 (arg, 64);
+ got = vget_lane_s64 (result, 0);
+ exp = 0xffffffffffffffff;
+ /* Expect: "result" = -1. */
+ if (exp != got)
+ return 1;
+ return 0;
+}
+
+int
+__attribute__ ((noinline))
+test_other ()
+{
+ int64x1_t arg;
+ int64x1_t result;
+ int64_t got;
+ int64_t exp;
+ arg = vcreate_s64 (0x0000000080000000);
+ result = vshr_n_s64 (arg, 4);
+ got = vget_lane_s64 (result, 0);
+ exp = 0x0000000008000000;
+ /* Expect: "result" = 0x0000000008000000. */
+ if (exp != got)
+ return 1;
+ return 0;
+}
+
+int __attribute__ ((noinline))
+test_other_neg ()
+{
+ int64x1_t arg;
+ int64x1_t result;
+ int64_t got;
+ int64_t exp;
+ arg = vcreate_s64 (0xffffffff80000000);
+ result = vshr_n_s64 (arg, 4);
+ got = vget_lane_s64 (result, 0);
+ exp = 0xfffffffff8000000;
+ /* Expect: "result" = 0xfffffffff8000000. */
+ if (exp != got)
+ return 1;
+ return 0;
+}
+
+int __attribute__ ((noinline))
+test_no_sshr0 ()
+{
+ int64x1_t arg;
+ int64x1_t result;
+ int64_t got;
+ int64_t exp;
+ arg = vcreate_s64 (0x0000000080000000);
+ result = vshr_n_s64 (arg, 0);
+ got = vget_lane_s64 (result, 0);
+ exp = 0x0000000080000000;
+ /* Expect: "result" = 0x0000000080000000. */
+ if (exp != got)
+ return 1;
+ return 0;
+}
+
+/* { dg-final { scan-assembler-not "sshr\\td\[0-9\]+, d\[0-9\]+, 0" } } */
+int
+main ()
+{
+ if (test_sshr64 ())
+ abort ();
+ if (test_other ())
+ abort ();
+
+ if (test_sshr64_neg ())
+ abort ();
+ if (test_other_neg ())
+ abort ();
+
+ if (test_no_sshr0 ())
+ abort ();
+
+ return 0;
+}
+
+/* { dg-final { cleanup-saved-temps } } */