* [PATCH 0/3] [GCC] arm: vld1q_types_xN ACLE intrinsics
@ 2023-10-06 9:49 Ezra.Sitorus
2023-10-06 9:49 ` [PATCH 1/3] [GCC] arm: vld1q_types_x2 " Ezra.Sitorus
` (3 more replies)
0 siblings, 4 replies; 6+ messages in thread
From: Ezra.Sitorus @ 2023-10-06 9:49 UTC (permalink / raw)
To: gcc-patches; +Cc: richard.earnshaw, kyrylo.tkachov
Add xN variants of vld1q_types intrinsic.
^ permalink raw reply [flat|nested] 6+ messages in thread
* [PATCH 1/3] [GCC] arm: vld1q_types_x2 ACLE intrinsics
2023-10-06 9:49 [PATCH 0/3] [GCC] arm: vld1q_types_xN ACLE intrinsics Ezra.Sitorus
@ 2023-10-06 9:49 ` Ezra.Sitorus
2023-10-24 10:20 ` [PING] " Ezra Sitorus
2023-10-06 9:49 ` [PATCH 2/3] [GCC] arm: vld1q_types_x3 " Ezra.Sitorus
` (2 subsequent siblings)
3 siblings, 1 reply; 6+ messages in thread
From: Ezra.Sitorus @ 2023-10-06 9:49 UTC (permalink / raw)
To: gcc-patches; +Cc: richard.earnshaw, kyrylo.tkachov
From: Ezra Sitorus <ezra.sitorus@arm.com>
This patch is part of a series of patches implementing the _xN variants of the vld1q intrinsic for arm32.
This patch adds the _x2 variants of the vld1q intrinsic. Tests use xN so that the latter variants (_x3, _x4) could be added.
ACLE documents are at https://developer.arm.com/documentation/ihi0053/latest/
ISA documents are at https://developer.arm.com/documentation/ddi0487/latest/
gcc/ChangeLog:
* config/arm/arm_neon.h
(vld1q_u8_x2, vld1q_u16_x2, vld1q_u32_x2, vld1q_u64_x2): New.
(vld1q_s8_x2, vld1q_s16_x2, vld1q_s32_x2, vld1q_s64_x2): New.
(vld1q_f16_x2, vld1q_f32_x2): New.
(vld1q_p8_x2, vld1q_p16_x2, vld1q_p64_x2): New.
(vld1q_bf16_x2): New.
* config/arm/arm_neon_builtins.def (vld1_x2): New entries.
* config/arm/neon.md (vld1_x2<mode>): New.
gcc/testsuite/ChangeLog:
* gcc.target/arm/simd/vld1q_base_xN_1.c: Add new test.
* gcc.target/arm/simd/vld1q_bf16_xN_1.c: Add new test.
* gcc.target/arm/simd/vld1q_fp16_xN_1.c: Add new test.
* gcc.target/arm/simd/vld1q_p64_xN_1.c: Add new test.
---
gcc/config/arm/arm_neon.h | 128 ++++++++++++++++++
gcc/config/arm/arm_neon_builtins.def | 1 +
gcc/config/arm/neon.md | 10 ++
.../gcc.target/arm/simd/vld1q_base_xN_1.c | 67 +++++++++
.../gcc.target/arm/simd/vld1q_bf16_xN_1.c | 13 ++
.../gcc.target/arm/simd/vld1q_fp16_xN_1.c | 14 ++
.../gcc.target/arm/simd/vld1q_p64_xN_1.c | 14 ++
7 files changed, 247 insertions(+)
create mode 100644 gcc/testsuite/gcc.target/arm/simd/vld1q_base_xN_1.c
create mode 100644 gcc/testsuite/gcc.target/arm/simd/vld1q_bf16_xN_1.c
create mode 100644 gcc/testsuite/gcc.target/arm/simd/vld1q_fp16_xN_1.c
create mode 100644 gcc/testsuite/gcc.target/arm/simd/vld1q_p64_xN_1.c
diff --git a/gcc/config/arm/arm_neon.h b/gcc/config/arm/arm_neon.h
index cdfdb44259a..3eb41c6bdc8 100644
--- a/gcc/config/arm/arm_neon.h
+++ b/gcc/config/arm/arm_neon.h
@@ -10403,6 +10403,15 @@ vld1q_p64 (const poly64_t * __a)
return (poly64x2_t)__builtin_neon_vld1v2di ((const __builtin_neon_di *) __a);
}
+__extension__ extern __inline poly64x2x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vld1q_p64_x2 (const poly64_t * __a)
+{
+ union { poly64x2x2_t __i; __builtin_neon_oi __o; } __rv;
+ __rv.__o = __builtin_neon_vld1_x2v2di ((const __builtin_neon_di *) __a);
+ return __rv.__i;
+}
+
#pragma GCC pop_options
__extension__ extern __inline int8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
@@ -10432,6 +10441,42 @@ vld1q_s64 (const int64_t * __a)
return (int64x2_t)__builtin_neon_vld1v2di ((const __builtin_neon_di *) __a);
}
+__extension__ extern __inline int8x16x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vld1q_s8_x2 (const int8_t * __a)
+{
+ union { int8x16x2_t __i; __builtin_neon_oi __o; } __rv;
+ __rv.__o = __builtin_neon_vld1_x2v16qi ((const __builtin_neon_qi *) __a);
+ return __rv.__i;
+}
+
+__extension__ extern __inline int16x8x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vld1q_s16_x2 (const int16_t * __a)
+{
+ union { int16x8x2_t __i; __builtin_neon_oi __o; } __rv;
+ __rv.__o = __builtin_neon_vld1_x2v8hi ((const __builtin_neon_hi *) __a);
+ return __rv.__i;
+}
+
+__extension__ extern __inline int32x4x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vld1q_s32_x2 (const int32_t * __a)
+{
+ union { int32x4x2_t __i; __builtin_neon_oi __o; } __rv;
+ __rv.__o = __builtin_neon_vld1_x2v4si ((const __builtin_neon_si *) __a);
+ return __rv.__i;
+}
+
+__extension__ extern __inline int64x2x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vld1q_s64_x2 (const int64_t * __a)
+{
+ union { int64x2x2_t __i; __builtin_neon_oi __o; } __rv;
+ __rv.__o = __builtin_neon_vld1_x2v2di ((const __builtin_neon_di *) __a);
+ return __rv.__i;
+}
+
#if defined (__ARM_FP16_FORMAT_IEEE) || defined (__ARM_FP16_FORMAT_ALTERNATIVE)
__extension__ extern __inline float16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
@@ -10448,6 +10493,26 @@ vld1q_f32 (const float32_t * __a)
return (float32x4_t)__builtin_neon_vld1v4sf ((const __builtin_neon_sf *) __a);
}
+#if defined (__ARM_FP16_FORMAT_IEEE) || defined (__ARM_FP16_FORMAT_ALTERNATIVE)
+__extension__ extern __inline float16x8x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vld1q_f16_x2 (const float16_t * __a)
+{
+ union { float16x8x2_t __i; __builtin_neon_oi __o; } __rv;
+ __rv.__o = __builtin_neon_vld1_x2v8hf (__a);
+ return __rv.__i;
+}
+#endif
+
+__extension__ extern __inline float32x4x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vld1q_f32_x2 (const float32_t * __a)
+{
+ union { float32x4x2_t __i; __builtin_neon_oi __o; } __rv;
+ __rv.__o = __builtin_neon_vld1_x2v4sf ((const __builtin_neon_sf *) __a);
+ return __rv.__i;
+}
+
__extension__ extern __inline uint8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vld1q_u8 (const uint8_t * __a)
@@ -10476,6 +10541,42 @@ vld1q_u64 (const uint64_t * __a)
return (uint64x2_t)__builtin_neon_vld1v2di ((const __builtin_neon_di *) __a);
}
+__extension__ extern __inline uint8x16x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vld1q_u8_x2 (const uint8_t * __a)
+{
+ union { uint8x16x2_t __i; __builtin_neon_oi __o; } __rv;
+ __rv.__o = __builtin_neon_vld1_x2v16qi ((const __builtin_neon_qi *) __a);
+ return __rv.__i;
+}
+
+__extension__ extern __inline uint16x8x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vld1q_u16_x2 (const uint16_t * __a)
+{
+ union { uint16x8x2_t __i; __builtin_neon_oi __o; } __rv;
+ __rv.__o = __builtin_neon_vld1_x2v8hi ((const __builtin_neon_hi *) __a);
+ return __rv.__i;
+}
+
+__extension__ extern __inline uint32x4x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vld1q_u32_x2 (const uint32_t * __a)
+{
+ union { uint32x4x2_t __i; __builtin_neon_oi __o; } __rv;
+ __rv.__o = __builtin_neon_vld1_x2v4si ((const __builtin_neon_si *) __a);
+ return __rv.__i;
+}
+
+__extension__ extern __inline uint64x2x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vld1q_u64_x2 (const uint64_t * __a)
+{
+ union { uint64x2x2_t __i; __builtin_neon_oi __o; } __rv;
+ __rv.__o = __builtin_neon_vld1_x2v2di ((const __builtin_neon_di *) __a);
+ return __rv.__i;
+}
+
__extension__ extern __inline poly8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vld1q_p8 (const poly8_t * __a)
@@ -10490,6 +10591,24 @@ vld1q_p16 (const poly16_t * __a)
return (poly16x8_t)__builtin_neon_vld1v8hi ((const __builtin_neon_hi *) __a);
}
+__extension__ extern __inline poly8x16x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vld1q_p8_x2 (const poly8_t * __a)
+{
+ union { poly8x16x2_t __i; __builtin_neon_oi __o; } __rv;
+ __rv.__o = __builtin_neon_vld1_x2v16qi ((const __builtin_neon_qi *) __a);
+ return __rv.__i;
+}
+
+__extension__ extern __inline poly16x8x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vld1q_p16_x2 (const poly16_t * __a)
+{
+ union { poly16x8x2_t __i; __builtin_neon_oi __o; } __rv;
+ __rv.__o = __builtin_neon_vld1_x2v8hi ((const __builtin_neon_hi *) __a);
+ return __rv.__i;
+}
+
__extension__ extern __inline int8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vld1_lane_s8 (const int8_t * __a, int8x8_t __b, const int __c)
@@ -19782,6 +19901,15 @@ vld1q_bf16 (const bfloat16_t * __ptr)
return __builtin_neon_vld1v8bf (__ptr);
}
+__extension__ extern __inline bfloat16x8x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vld1q_bf16_x2 (const bfloat16_t * __ptr)
+{
+ union { bfloat16x8x2_t __i; __builtin_neon_oi __o; } __rv;
+ __rv.__o = __builtin_neon_vld1_x2v8bf ((const __builtin_neon_bf *) __ptr);
+ return __rv.__i;
+}
+
__extension__ extern __inline bfloat16x4x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vld2_bf16 (bfloat16_t const * __ptr)
diff --git a/gcc/config/arm/arm_neon_builtins.def b/gcc/config/arm/arm_neon_builtins.def
index 94b15238123..5fadd255c18 100644
--- a/gcc/config/arm/arm_neon_builtins.def
+++ b/gcc/config/arm/arm_neon_builtins.def
@@ -298,6 +298,7 @@ VAR1 (TERNOP, vtbx1, v8qi)
VAR1 (TERNOP, vtbx2, v8qi)
VAR1 (TERNOP, vtbx3, v8qi)
VAR1 (TERNOP, vtbx4, v8qi)
+VAR7 (LOAD1, vld1_x2, v16qi, v8hi, v4si, v2di, v8hf, v4sf, v8bf)
VAR13 (LOAD1, vld1,
v8qi, v4hi, v4hf, v2si, v2sf, v16qi, v8hi, v8hf, v4si, v4sf, v2di,
v4bf, v8bf)
diff --git a/gcc/config/arm/neon.md b/gcc/config/arm/neon.md
index d213369ffc3..55049ea549f 100644
--- a/gcc/config/arm/neon.md
+++ b/gcc/config/arm/neon.md
@@ -4957,6 +4957,16 @@ if (BYTES_BIG_ENDIAN)
[(set_attr "type" "neon_load1_1reg<q>")]
)
+(define_insn "neon_vld1_x2<mode>"
+ [(set (match_operand:OI 0 "s_register_operand" "=w")
+ (unspec:OI [(match_operand:OI 1 "neon_struct_operand" "Um")
+ (unspec:VQXBF [(const_int 0)] UNSPEC_VSTRUCTDUMMY)]
+ UNSPEC_VLD1))]
+ "TARGET_NEON"
+ "vld1.<V_sz_elem>\t%h0, %A1"
+ [(set_attr "type" "neon_load1_2reg<q>")]
+)
+
;; The lane numbers in the RTL are in GCC lane order, having been flipped
;; in arm_expand_neon_args. The lane numbers are restored to architectural
;; lane order here.
diff --git a/gcc/testsuite/gcc.target/arm/simd/vld1q_base_xN_1.c b/gcc/testsuite/gcc.target/arm/simd/vld1q_base_xN_1.c
new file mode 100644
index 00000000000..1d31777afdf
--- /dev/null
+++ b/gcc/testsuite/gcc.target/arm/simd/vld1q_base_xN_1.c
@@ -0,0 +1,67 @@
+/* { dg-do assemble } */
+/* { dg-require-effective-target arm_neon_ok } */
+/* { dg-options "-save-temps -O2" } */
+/* { dg-add-options arm_neon } */
+
+#include "arm_neon.h"
+
+uint8x16x2_t test_vld1q_u8_x2 (uint8_t * a)
+{
+ return vld1q_u8_x2 (a);
+}
+
+uint16x8x2_t test_vld1q_u16_x2 (uint16_t * a)
+{
+ return vld1q_u16_x2 (a);
+}
+
+uint32x4x2_t test_vld1q_u32_x2 (uint32_t * a)
+{
+ return vld1q_u32_x2 (a);
+}
+
+uint64x2x2_t test_vld1q_u64_x2 (uint64_t * a)
+{
+ return vld1q_u64_x2 (a);
+}
+
+int8x16x2_t test_vld1q_s8_x2 (int8_t * a)
+{
+ return vld1q_s8_x2 (a);
+}
+
+int16x8x2_t test_vld1q_s16_x2 (int16_t * a)
+{
+ return vld1q_s16_x2 (a);
+}
+
+int32x4x2_t test_vld1q_s32_x2 (int32_t * a)
+{
+ return vld1q_s32_x2 (a);
+}
+
+int64x2x2_t test_vld1q_s64_x2 (int64_t * a)
+{
+ return vld1q_s64_x2 (a);
+}
+
+float32x4x2_t test_vld1q_f32_x2 (float32_t * a)
+{
+ return vld1q_f32_x2 (a);
+}
+
+poly8x16x2_t test_vld1q_p8_x2 (poly8_t * a)
+{
+ return vld1q_p8_x2 (a);
+}
+
+poly16x8x2_t test_vld1q_p16_x2 (poly16_t * a)
+{
+ return vld1q_p16_x2 (a);
+}
+
+/* { dg-final { scan-assembler-times {vld1.8\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+\]\n} 3 } } */
+/* { dg-final { scan-assembler-times {vld1.16\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+\]\n} 3 } } */
+/* { dg-final { scan-assembler-times {vld1.32\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+\]\n} 3 } } */
+/* { dg-final { scan-assembler-times {vld1.64\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+:64\]\n} 2 } } */
+
diff --git a/gcc/testsuite/gcc.target/arm/simd/vld1q_bf16_xN_1.c b/gcc/testsuite/gcc.target/arm/simd/vld1q_bf16_xN_1.c
new file mode 100644
index 00000000000..5f6fc98640e
--- /dev/null
+++ b/gcc/testsuite/gcc.target/arm/simd/vld1q_bf16_xN_1.c
@@ -0,0 +1,13 @@
+/* { dg-do assemble } */
+/* { dg-require-effective-target arm_v8_2a_bf16_neon_ok } */
+/* { dg-options "-save-temps -O2" } */
+/* { dg-add-options arm_v8_2a_bf16_neon } */
+
+#include "arm_neon.h"
+
+bfloat16x8x2_t test_vld1q_bf16_x2 (bfloat16_t * a)
+{
+ return vld1q_bf16_x2 (a);
+}
+
+/* { dg-final { scan-assembler-times {vld1.16\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+\]\n} 1 } } */
diff --git a/gcc/testsuite/gcc.target/arm/simd/vld1q_fp16_xN_1.c b/gcc/testsuite/gcc.target/arm/simd/vld1q_fp16_xN_1.c
new file mode 100644
index 00000000000..aecf491a4de
--- /dev/null
+++ b/gcc/testsuite/gcc.target/arm/simd/vld1q_fp16_xN_1.c
@@ -0,0 +1,14 @@
+/* { dg-do assemble } */
+/* { dg-require-effective-target arm_neon_fp16_ok } */
+/* { dg-options "-save-temps -O2" } */
+/* { dg-add-options arm_neon_fp16 } */
+
+#include "arm_neon.h"
+
+float16x8x2_t test_vld1q_f16_x2 (float16_t * a)
+{
+ return vld1q_f16_x2 (a);
+}
+
+/* { dg-final { scan-assembler-times {vld1.16\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+\]\n} 1 } } */
+
diff --git a/gcc/testsuite/gcc.target/arm/simd/vld1q_p64_xN_1.c b/gcc/testsuite/gcc.target/arm/simd/vld1q_p64_xN_1.c
new file mode 100644
index 00000000000..04ceb5e4a24
--- /dev/null
+++ b/gcc/testsuite/gcc.target/arm/simd/vld1q_p64_xN_1.c
@@ -0,0 +1,14 @@
+/* { dg-do assemble } */
+/* { dg-require-effective-target arm_crypto_ok } */
+/* { dg-options "-save-temps -O2" } */
+/* { dg-add-options arm_crypto } */
+
+#include "arm_neon.h"
+
+poly64x2x2_t test_vld1q_p64_x2 (poly64_t * a)
+{
+ return vld1q_p64_x2 (a);
+}
+
+/* { dg-final { scan-assembler-times {vld1.64\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+:64\]\n} 1 } } */
+
--
2.25.1
^ permalink raw reply [flat|nested] 6+ messages in thread
* [PATCH 2/3] [GCC] arm: vld1q_types_x3 ACLE intrinsics
2023-10-06 9:49 [PATCH 0/3] [GCC] arm: vld1q_types_xN ACLE intrinsics Ezra.Sitorus
2023-10-06 9:49 ` [PATCH 1/3] [GCC] arm: vld1q_types_x2 " Ezra.Sitorus
@ 2023-10-06 9:49 ` Ezra.Sitorus
2023-10-06 9:49 ` [PATCH 3/3] [GCC] arm: vld1q_types_x4 " Ezra.Sitorus
2023-11-27 15:06 ` [PATCH 0/3] [GCC] arm: vld1q_types_xN " Richard Earnshaw
3 siblings, 0 replies; 6+ messages in thread
From: Ezra.Sitorus @ 2023-10-06 9:49 UTC (permalink / raw)
To: gcc-patches; +Cc: richard.earnshaw, kyrylo.tkachov
From: Ezra Sitorus <ezra.sitorus@arm.com>
This patch is part of a series of patches implementing the _xN variants of the vld1q intrinsic for arm32.
This patch adds the _x3 variants of the vld1q intrinsic. This depends on the the _x2 patch.
ACLE documents are at https://developer.arm.com/documentation/ihi0053/latest/
ISA documents are at https://developer.arm.com/documentation/ddi0487/latest/
gcc/ChangeLog:
* config/arm/arm_neon.h
(vld1q_u8_x3, vld1q_u16_x3, vld1q_u32_x3, vld1q_u64_x3): New.
(vld1q_s8_x3, vld1q_s16_x3, vld1q_s32_x3, vld1q_s64_x3): New.
(vld1q_f16_x3, vld1q_f32_x3): New.
(vld1q_p8_x3, vld1q_p16_x3, vld1q_p64_x3): New.
(vld1q_bf16_x3): New.
* config/arm/arm_neon_builtins.def (vld1_x3): New entries.
* config/arm/neon.md (vld1_x3<mode>): New.
gcc/testsuite/ChangeLog:
* gcc.target/arm/simd/vld1q_base_xN_1.c: Add new tests.
* gcc.target/arm/simd/vld1q_bf16_xN_1.c: Add new tests.
* gcc.target/arm/simd/vld1q_fp16_xN_1.c: Add new tests.
* gcc.target/arm/simd/vld1q_p64_xN_1.c: Add new tests.
---
gcc/config/arm/arm_neon.h | 128 ++++++++++++++++++
gcc/config/arm/arm_neon_builtins.def | 1 +
gcc/config/arm/neon.md | 27 ++++
.../gcc.target/arm/simd/vld1q_base_xN_1.c | 63 ++++++++-
.../gcc.target/arm/simd/vld1q_bf16_xN_1.c | 6 +
.../gcc.target/arm/simd/vld1q_fp16_xN_1.c | 7 +-
.../gcc.target/arm/simd/vld1q_p64_xN_1.c | 7 +-
7 files changed, 236 insertions(+), 3 deletions(-)
diff --git a/gcc/config/arm/arm_neon.h b/gcc/config/arm/arm_neon.h
index 3eb41c6bdc8..557873ac028 100644
--- a/gcc/config/arm/arm_neon.h
+++ b/gcc/config/arm/arm_neon.h
@@ -10412,6 +10412,15 @@ vld1q_p64_x2 (const poly64_t * __a)
return __rv.__i;
}
+__extension__ extern __inline poly64x2x3_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vld1q_p64_x3 (const poly64_t * __a)
+{
+ union { poly64x2x3_t __i; __builtin_neon_ci __o; } __rv;
+ __rv.__o = __builtin_neon_vld1_x3v2di ((const __builtin_neon_di *) __a);
+ return __rv.__i;
+}
+
#pragma GCC pop_options
__extension__ extern __inline int8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
@@ -10477,6 +10486,42 @@ vld1q_s64_x2 (const int64_t * __a)
return __rv.__i;
}
+__extension__ extern __inline int8x16x3_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vld1q_s8_x3 (const uint8_t * __a)
+{
+ union { int8x16x3_t __i; __builtin_neon_ci __o; } __rv;
+ __rv.__o = __builtin_neon_vld1_x3v16qi ((const __builtin_neon_qi *) __a);
+ return __rv.__i;
+}
+
+__extension__ extern __inline int16x8x3_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vld1q_s16_x3 (const uint16_t * __a)
+{
+ union { int16x8x3_t __i; __builtin_neon_ci __o; } __rv;
+ __rv.__o = __builtin_neon_vld1_x3v8hi ((const __builtin_neon_hi *) __a);
+ return __rv.__i;
+}
+
+__extension__ extern __inline int32x4x3_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vld1q_s32_x3 (const int32_t * __a)
+{
+ union { int32x4x3_t __i; __builtin_neon_ci __o; } __rv;
+ __rv.__o = __builtin_neon_vld1_x3v4si ((const __builtin_neon_si *) __a);
+ return __rv.__i;
+}
+
+__extension__ extern __inline int64x2x3_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vld1q_s64_x3 (const int64_t * __a)
+{
+ union { int64x2x3_t __i; __builtin_neon_ci __o; } __rv;
+ __rv.__o = __builtin_neon_vld1_x3v2di ((const __builtin_neon_di *) __a);
+ return __rv.__i;
+}
+
#if defined (__ARM_FP16_FORMAT_IEEE) || defined (__ARM_FP16_FORMAT_ALTERNATIVE)
__extension__ extern __inline float16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
@@ -10513,6 +10558,26 @@ vld1q_f32_x2 (const float32_t * __a)
return __rv.__i;
}
+#if defined (__ARM_FP16_FORMAT_IEEE) || defined (__ARM_FP16_FORMAT_ALTERNATIVE)
+__extension__ extern __inline float16x8x3_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vld1q_f16_x3 (const float16_t * __a)
+{
+ union { float16x8x3_t __i; __builtin_neon_ci __o; } __rv;
+ __rv.__o = __builtin_neon_vld1_x3v8hf (__a);
+ return __rv.__i;
+}
+#endif
+
+__extension__ extern __inline float32x4x3_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vld1q_f32_x3 (const float32_t * __a)
+{
+ union { float32x4x3_t __i; __builtin_neon_ci __o; } __rv;
+ __rv.__o = __builtin_neon_vld1_x3v4sf ((const __builtin_neon_sf *) __a);
+ return __rv.__i;
+}
+
__extension__ extern __inline uint8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vld1q_u8 (const uint8_t * __a)
@@ -10577,6 +10642,42 @@ vld1q_u64_x2 (const uint64_t * __a)
return __rv.__i;
}
+__extension__ extern __inline uint8x16x3_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vld1q_u8_x3 (const uint8_t * __a)
+{
+ union { uint8x16x3_t __i; __builtin_neon_ci __o; } __rv;
+ __rv.__o = __builtin_neon_vld1_x3v16qi ((const __builtin_neon_qi *) __a);
+ return __rv.__i;
+}
+
+__extension__ extern __inline uint16x8x3_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vld1q_u16_x3 (const uint16_t * __a)
+{
+ union { uint16x8x3_t __i; __builtin_neon_ci __o; } __rv;
+ __rv.__o = __builtin_neon_vld1_x3v8hi ((const __builtin_neon_hi *) __a);
+ return __rv.__i;
+}
+
+__extension__ extern __inline uint32x4x3_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vld1q_u32_x3 (const uint32_t * __a)
+{
+ union { uint32x4x3_t __i; __builtin_neon_ci __o; } __rv;
+ __rv.__o = __builtin_neon_vld1_x3v4si ((const __builtin_neon_si *) __a);
+ return __rv.__i;
+}
+
+__extension__ extern __inline uint64x2x3_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vld1q_u64_x3 (const uint64_t * __a)
+{
+ union { uint64x2x3_t __i; __builtin_neon_ci __o; } __rv;
+ __rv.__o = __builtin_neon_vld1_x3v2di ((const __builtin_neon_di *) __a);
+ return __rv.__i;
+}
+
__extension__ extern __inline poly8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vld1q_p8 (const poly8_t * __a)
@@ -10609,6 +10710,24 @@ vld1q_p16_x2 (const poly16_t * __a)
return __rv.__i;
}
+__extension__ extern __inline poly8x16x3_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vld1q_p8_x3 (const poly8_t * __a)
+{
+ union { poly8x16x3_t __i; __builtin_neon_ci __o; } __rv;
+ __rv.__o = __builtin_neon_vld1_x3v16qi ((const __builtin_neon_qi *) __a);
+ return __rv.__i;
+}
+
+__extension__ extern __inline poly16x8x3_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vld1q_p16_x3 (const poly16_t * __a)
+{
+ union { poly16x8x3_t __i; __builtin_neon_ci __o; } __rv;
+ __rv.__o = __builtin_neon_vld1_x3v8hi ((const __builtin_neon_hi *) __a);
+ return __rv.__i;
+}
+
__extension__ extern __inline int8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vld1_lane_s8 (const int8_t * __a, int8x8_t __b, const int __c)
@@ -19910,6 +20029,15 @@ vld1q_bf16_x2 (const bfloat16_t * __ptr)
return __rv.__i;
}
+__extension__ extern __inline bfloat16x8x3_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vld1q_bf16_x3 (const bfloat16_t * __ptr)
+{
+ union { bfloat16x8x3_t __i; __builtin_neon_oi __o; } __rv;
+ __rv.__o = __builtin_neon_vld1_x3v8bf ((const __builtin_neon_bf *) __ptr);
+ return __rv.__i;
+}
+
__extension__ extern __inline bfloat16x4x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vld2_bf16 (bfloat16_t const * __ptr)
diff --git a/gcc/config/arm/arm_neon_builtins.def b/gcc/config/arm/arm_neon_builtins.def
index 5fadd255c18..57783bd0089 100644
--- a/gcc/config/arm/arm_neon_builtins.def
+++ b/gcc/config/arm/arm_neon_builtins.def
@@ -299,6 +299,7 @@ VAR1 (TERNOP, vtbx2, v8qi)
VAR1 (TERNOP, vtbx3, v8qi)
VAR1 (TERNOP, vtbx4, v8qi)
VAR7 (LOAD1, vld1_x2, v16qi, v8hi, v4si, v2di, v8hf, v4sf, v8bf)
+VAR7 (LOAD1, vld1_x3, v16qi, v8hi, v4si, v2di, v8hf, v4sf, v8bf)
VAR13 (LOAD1, vld1,
v8qi, v4hi, v4hf, v2si, v2sf, v16qi, v8hi, v8hf, v4si, v4sf, v2di,
v4bf, v8bf)
diff --git a/gcc/config/arm/neon.md b/gcc/config/arm/neon.md
index 55049ea549f..b37d95f1fa0 100644
--- a/gcc/config/arm/neon.md
+++ b/gcc/config/arm/neon.md
@@ -4967,6 +4967,33 @@ if (BYTES_BIG_ENDIAN)
[(set_attr "type" "neon_load1_2reg<q>")]
)
+(define_insn "neon_vld1_x3<mode>"
+ [(set (match_operand:CI 0 "s_register_operand" "=w")
+ (unspec:CI [(match_operand:EI 1 "neon_struct_operand" "Um")
+ (unspec:VQXBF [(const_int 0)] UNSPEC_VSTRUCTDUMMY)]
+ UNSPEC_VLD3A))]
+ "TARGET_NEON"
+{
+ int regno = REGNO (operands[0]);
+ rtx ops[4];
+ ops[0] = gen_rtx_REG (DImode, regno);
+ ops[1] = gen_rtx_REG (DImode, regno + 2);
+ ops[2] = gen_rtx_REG (DImode, regno + 4);
+ ops[3] = operands[1];
+
+ output_asm_insn ("vld1.<V_sz_elem>\t{%P0, %P1, %P2}, %A3", ops);
+
+ ops[0] = gen_rtx_REG (DImode, regno + 6);
+ ops[1] = gen_rtx_REG (DImode, regno + 8);
+ ops[2] = gen_rtx_REG (DImode, regno + 10);
+ ops[3] = operands[1];
+
+ output_asm_insn ("vld1.<V_sz_elem>\t{%P0, %P1, %P2}, %A3", ops);
+ return "";
+}
+ [(set_attr "type" "neon_load1_3reg<q>")]
+)
+
;; The lane numbers in the RTL are in GCC lane order, having been flipped
;; in arm_expand_neon_args. The lane numbers are restored to architectural
;; lane order here.
diff --git a/gcc/testsuite/gcc.target/arm/simd/vld1q_base_xN_1.c b/gcc/testsuite/gcc.target/arm/simd/vld1q_base_xN_1.c
index 1d31777afdf..bfad282751b 100644
--- a/gcc/testsuite/gcc.target/arm/simd/vld1q_base_xN_1.c
+++ b/gcc/testsuite/gcc.target/arm/simd/vld1q_base_xN_1.c
@@ -60,8 +60,69 @@ poly16x8x2_t test_vld1q_p16_x2 (poly16_t * a)
return vld1q_p16_x2 (a);
}
+uint8x16x3_t test_vld1q_u8_x3 (uint8_t * a)
+{
+ return vld1q_u8_x3 (a);
+}
+
+uint16x8x3_t test_vld1q_u16_x3 (uint16_t * a)
+{
+ return vld1q_u16_x3 (a);
+}
+
+uint32x4x3_t test_vld1q_u32_x3 (uint32_t * a)
+{
+ return vld1q_u32_x3 (a);
+}
+
+uint64x2x3_t test_vld1q_u64_x3 (uint64_t * a)
+{
+ return vld1q_u64_x3 (a);
+}
+
+int8x16x3_t test_vld1q_s8_x3 (int8_t * a)
+{
+ return vld1q_s8_x3 (a);
+}
+
+int16x8x3_t test_vld1q_s16_x3 (int16_t * a)
+{
+ return vld1q_s16_x3 (a);
+}
+
+int32x4x3_t test_vld1q_s32_x3 (int32_t * a)
+{
+ return vld1q_s32_x3 (a);
+}
+
+int64x2x3_t test_vld1q_s64_x3 (int64_t * a)
+{
+ return vld1q_s64_x3 (a);
+}
+
+float32x4x3_t test_vld1q_f32_x3 (float32_t * a)
+{
+ return vld1q_f32_x3 (a);
+}
+
+poly8x16x3_t test_vld1q_p8_x3 (poly8_t * a)
+{
+ return vld1q_p8_x3 (a);
+}
+
+poly16x8x3_t test_vld1q_p16_x3 (poly16_t * a)
+{
+ return vld1q_p16_x3 (a);
+}
+
/* { dg-final { scan-assembler-times {vld1.8\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+\]\n} 3 } } */
+/* { dg-final { scan-assembler-times {vld1.8\t\{d[0-9]+, d[0-9]+, d[0-9]+\}, \[r[0-9]+\]\n} 6 } } */
+
/* { dg-final { scan-assembler-times {vld1.16\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+\]\n} 3 } } */
+/* { dg-final { scan-assembler-times {vld1.16\t\{d[0-9]+, d[0-9]+, d[0-9]+\}, \[r[0-9]+\]\n} 6 } } */
+
/* { dg-final { scan-assembler-times {vld1.32\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+\]\n} 3 } } */
-/* { dg-final { scan-assembler-times {vld1.64\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+:64\]\n} 2 } } */
+/* { dg-final { scan-assembler-times {vld1.32\t\{d[0-9]+, d[0-9]+, d[0-9]+\}, \[r[0-9]+\]\n} 6 } } */
+/* { dg-final { scan-assembler-times {vld1.64\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+:64\]\n} 2 } } */
+/* { dg-final { scan-assembler-times {vld1.64\t\{d[0-9]+, d[0-9]+, d[0-9]+\}, \[r[0-9]+:64\]\n} 4 } } */
diff --git a/gcc/testsuite/gcc.target/arm/simd/vld1q_bf16_xN_1.c b/gcc/testsuite/gcc.target/arm/simd/vld1q_bf16_xN_1.c
index 5f6fc98640e..4138fe951ee 100644
--- a/gcc/testsuite/gcc.target/arm/simd/vld1q_bf16_xN_1.c
+++ b/gcc/testsuite/gcc.target/arm/simd/vld1q_bf16_xN_1.c
@@ -10,4 +10,10 @@ bfloat16x8x2_t test_vld1q_bf16_x2 (bfloat16_t * a)
return vld1q_bf16_x2 (a);
}
+bfloat16x8x3_t test_vld1q_bf16_x3 (bfloat16_t * a)
+{
+ return vld1q_bf16_x3 (a);
+}
+
/* { dg-final { scan-assembler-times {vld1.16\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+\]\n} 1 } } */
+/* { dg-final { scan-assembler-times {vld1.16\t\{d[0-9]+, d[0-9]+, d[0-9]+\}, \[r[0-9]+\]\n} 2 } } */
diff --git a/gcc/testsuite/gcc.target/arm/simd/vld1q_fp16_xN_1.c b/gcc/testsuite/gcc.target/arm/simd/vld1q_fp16_xN_1.c
index aecf491a4de..01640d7cc1f 100644
--- a/gcc/testsuite/gcc.target/arm/simd/vld1q_fp16_xN_1.c
+++ b/gcc/testsuite/gcc.target/arm/simd/vld1q_fp16_xN_1.c
@@ -10,5 +10,10 @@ float16x8x2_t test_vld1q_f16_x2 (float16_t * a)
return vld1q_f16_x2 (a);
}
-/* { dg-final { scan-assembler-times {vld1.16\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+\]\n} 1 } } */
+float16x8x3_t test_vld1q_f16_x3 (float16_t * a)
+{
+ return vld1q_f16_x3 (a);
+}
+/* { dg-final { scan-assembler-times {vld1.16\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+\]\n} 1 } } */
+/* { dg-final { scan-assembler-times {vld1.16\t\{d[0-9]+, d[0-9]+, d[0-9]+\}, \[r[0-9]+\]\n} 2 } } */
diff --git a/gcc/testsuite/gcc.target/arm/simd/vld1q_p64_xN_1.c b/gcc/testsuite/gcc.target/arm/simd/vld1q_p64_xN_1.c
index 04ceb5e4a24..ae2ab36df57 100644
--- a/gcc/testsuite/gcc.target/arm/simd/vld1q_p64_xN_1.c
+++ b/gcc/testsuite/gcc.target/arm/simd/vld1q_p64_xN_1.c
@@ -10,5 +10,10 @@ poly64x2x2_t test_vld1q_p64_x2 (poly64_t * a)
return vld1q_p64_x2 (a);
}
-/* { dg-final { scan-assembler-times {vld1.64\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+:64\]\n} 1 } } */
+poly64x2x3_t test_vld1q_p64_x3 (poly64_t * a)
+{
+ return vld1q_p64_x3 (a);
+}
+/* { dg-final { scan-assembler-times {vld1.64\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+:64\]\n} 1 } } */
+/* { dg-final { scan-assembler-times {vld1.64\t\{d[0-9]+, d[0-9]+, d[0-9]+\}, \[r[0-9]+:64\]\n} 2 } } */
--
2.25.1
^ permalink raw reply [flat|nested] 6+ messages in thread
* [PATCH 3/3] [GCC] arm: vld1q_types_x4 ACLE intrinsics
2023-10-06 9:49 [PATCH 0/3] [GCC] arm: vld1q_types_xN ACLE intrinsics Ezra.Sitorus
2023-10-06 9:49 ` [PATCH 1/3] [GCC] arm: vld1q_types_x2 " Ezra.Sitorus
2023-10-06 9:49 ` [PATCH 2/3] [GCC] arm: vld1q_types_x3 " Ezra.Sitorus
@ 2023-10-06 9:49 ` Ezra.Sitorus
2023-11-27 15:06 ` [PATCH 0/3] [GCC] arm: vld1q_types_xN " Richard Earnshaw
3 siblings, 0 replies; 6+ messages in thread
From: Ezra.Sitorus @ 2023-10-06 9:49 UTC (permalink / raw)
To: gcc-patches; +Cc: richard.earnshaw, kyrylo.tkachov
From: Ezra Sitorus <ezra.sitorus@arm.com>
This patch is part of a series of patches implementing the _xN variants of the vld1q intrinsic for arm32.
This patch adds the _x4 variants of the vld1q intrinsic. This depends on the the _x2 patch.
ACLE documents are at https://developer.arm.com/documentation/ihi0053/latest/
ISA documents are at https://developer.arm.com/documentation/ddi0487/latest/
gcc/ChangeLog:
* config/arm/arm_neon.h
(vld1q_u8_x4, vld1q_u16_x4, vld1q_u32_x4, vld1q_u64_x4): New.
(vld1q_s8_x4, vld1q_s16_x4, vld1q_s32_x4, vld1q_s64_x4): New.
(vld1q_f16_x4, vld1q_f32_x4): New.
(vld1q_p8_x4, vld1q_p16_x4, vld1q_p64_x4): New.
(vld1q_bf16_x4): New.
* config/arm/arm_neon_builtins.def (vld1_x4): New entries.
* config/arm/neon.md (vld1_x4<mode>): New.
gcc/testsuite/ChangeLog:
* gcc.target/arm/simd/vld1q_base_xN_1.c: Add new tests.
* gcc.target/arm/simd/vld1q_bf16_xN_1.c: Add new tests.
* gcc.target/arm/simd/vld1q_fp16_xN_1.c: Add new tests.
* gcc.target/arm/simd/vld1q_p64_xN_1.c: Add new tests.
---
gcc/config/arm/arm_neon.h | 128 ++++++++++++++++++
gcc/config/arm/arm_neon_builtins.def | 1 +
gcc/config/arm/neon.md | 30 ++++
.../gcc.target/arm/simd/vld1q_base_xN_1.c | 59 ++++++++
.../gcc.target/arm/simd/vld1q_bf16_xN_1.c | 6 +
.../gcc.target/arm/simd/vld1q_fp16_xN_1.c | 6 +
.../gcc.target/arm/simd/vld1q_p64_xN_1.c | 6 +
7 files changed, 236 insertions(+)
diff --git a/gcc/config/arm/arm_neon.h b/gcc/config/arm/arm_neon.h
index 557873ac028..c03be9912f8 100644
--- a/gcc/config/arm/arm_neon.h
+++ b/gcc/config/arm/arm_neon.h
@@ -10421,6 +10421,15 @@ vld1q_p64_x3 (const poly64_t * __a)
return __rv.__i;
}
+__extension__ extern __inline poly64x2x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vld1q_p64_x4 (const poly64_t * __a)
+{
+ union { poly64x2x4_t __i; __builtin_neon_xi __o; } __rv;
+ __rv.__o = __builtin_neon_vld1_x4v2di ((const __builtin_neon_di *) __a);
+ return __rv.__i;
+}
+
#pragma GCC pop_options
__extension__ extern __inline int8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
@@ -10522,6 +10531,42 @@ vld1q_s64_x3 (const int64_t * __a)
return __rv.__i;
}
+__extension__ extern __inline int8x16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vld1q_s8_x4 (const uint8_t * __a)
+{
+ union { int8x16x4_t __i; __builtin_neon_xi __o; } __rv;
+ __rv.__o = __builtin_neon_vld1_x4v16qi ((const __builtin_neon_qi *) __a);
+ return __rv.__i;
+}
+
+__extension__ extern __inline int16x8x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vld1q_s16_x4 (const uint16_t * __a)
+{
+ union { int16x8x4_t __i; __builtin_neon_xi __o; } __rv;
+ __rv.__o = __builtin_neon_vld1_x4v8hi ((const __builtin_neon_hi *) __a);
+ return __rv.__i;
+}
+
+__extension__ extern __inline int32x4x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vld1q_s32_x4 (const int32_t * __a)
+{
+ union { int32x4x4_t __i; __builtin_neon_xi __o; } __rv;
+ __rv.__o = __builtin_neon_vld1_x4v4si ((const __builtin_neon_si *) __a);
+ return __rv.__i;
+}
+
+__extension__ extern __inline int64x2x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vld1q_s64_x4 (const int64_t * __a)
+{
+ union { int64x2x4_t __i; __builtin_neon_xi __o; } __rv;
+ __rv.__o = __builtin_neon_vld1_x4v2di ((const __builtin_neon_di *) __a);
+ return __rv.__i;
+}
+
#if defined (__ARM_FP16_FORMAT_IEEE) || defined (__ARM_FP16_FORMAT_ALTERNATIVE)
__extension__ extern __inline float16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
@@ -10578,6 +10623,26 @@ vld1q_f32_x3 (const float32_t * __a)
return __rv.__i;
}
+#if defined (__ARM_FP16_FORMAT_IEEE) || defined (__ARM_FP16_FORMAT_ALTERNATIVE)
+__extension__ extern __inline float16x8x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vld1q_f16_x4 (const float16_t * __a)
+{
+ union { float16x8x4_t __i; __builtin_neon_xi __o; } __rv;
+ __rv.__o = __builtin_neon_vld1_x4v8hf (__a);
+ return __rv.__i;
+}
+#endif
+
+__extension__ extern __inline float32x4x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vld1q_f32_x4 (const float32_t * __a)
+{
+ union { float32x4x4_t __i; __builtin_neon_xi __o; } __rv;
+ __rv.__o = __builtin_neon_vld1_x4v4sf ((const __builtin_neon_sf *) __a);
+ return __rv.__i;
+}
+
__extension__ extern __inline uint8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vld1q_u8 (const uint8_t * __a)
@@ -10678,6 +10743,42 @@ vld1q_u64_x3 (const uint64_t * __a)
return __rv.__i;
}
+__extension__ extern __inline uint8x16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vld1q_u8_x4 (const uint8_t * __a)
+{
+ union { uint8x16x4_t __i; __builtin_neon_xi __o; } __rv;
+ __rv.__o = __builtin_neon_vld1_x4v16qi ((const __builtin_neon_qi *) __a);
+ return __rv.__i;
+}
+
+__extension__ extern __inline uint16x8x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vld1q_u16_x4 (const uint16_t * __a)
+{
+ union { uint16x8x4_t __i; __builtin_neon_xi __o; } __rv;
+ __rv.__o = __builtin_neon_vld1_x4v8hi ((const __builtin_neon_hi *) __a);
+ return __rv.__i;
+}
+
+__extension__ extern __inline uint32x4x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vld1q_u32_x4 (const uint32_t * __a)
+{
+ union { uint32x4x4_t __i; __builtin_neon_xi __o; } __rv;
+ __rv.__o = __builtin_neon_vld1_x4v4si ((const __builtin_neon_si *) __a);
+ return __rv.__i;
+}
+
+__extension__ extern __inline uint64x2x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vld1q_u64_x4 (const uint64_t * __a)
+{
+ union { uint64x2x4_t __i; __builtin_neon_xi __o; } __rv;
+ __rv.__o = __builtin_neon_vld1_x4v2di ((const __builtin_neon_di *) __a);
+ return __rv.__i;
+}
+
__extension__ extern __inline poly8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vld1q_p8 (const poly8_t * __a)
@@ -10728,6 +10829,24 @@ vld1q_p16_x3 (const poly16_t * __a)
return __rv.__i;
}
+__extension__ extern __inline poly8x16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vld1q_p8_x4 (const poly8_t * __a)
+{
+ union { poly8x16x4_t __i; __builtin_neon_xi __o; } __rv;
+ __rv.__o = __builtin_neon_vld1_x4v16qi ((const __builtin_neon_qi *) __a);
+ return __rv.__i;
+}
+
+__extension__ extern __inline poly16x8x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vld1q_p16_x4 (const poly16_t * __a)
+{
+ union { poly16x8x4_t __i; __builtin_neon_xi __o; } __rv;
+ __rv.__o = __builtin_neon_vld1_x4v8hi ((const __builtin_neon_hi *) __a);
+ return __rv.__i;
+}
+
__extension__ extern __inline int8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vld1_lane_s8 (const int8_t * __a, int8x8_t __b, const int __c)
@@ -20038,6 +20157,15 @@ vld1q_bf16_x3 (const bfloat16_t * __ptr)
return __rv.__i;
}
+__extension__ extern __inline bfloat16x8x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vld1q_bf16_x4 (const bfloat16_t * __ptr)
+{
+ union { bfloat16x8x4_t __i; __builtin_neon_xi __o; } __rv;
+ __rv.__o = __builtin_neon_vld1_x4v8bf ((const __builtin_neon_bf *) __ptr);
+ return __rv.__i;
+}
+
__extension__ extern __inline bfloat16x4x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vld2_bf16 (bfloat16_t const * __ptr)
diff --git a/gcc/config/arm/arm_neon_builtins.def b/gcc/config/arm/arm_neon_builtins.def
index 57783bd0089..f4001b298c4 100644
--- a/gcc/config/arm/arm_neon_builtins.def
+++ b/gcc/config/arm/arm_neon_builtins.def
@@ -300,6 +300,7 @@ VAR1 (TERNOP, vtbx3, v8qi)
VAR1 (TERNOP, vtbx4, v8qi)
VAR7 (LOAD1, vld1_x2, v16qi, v8hi, v4si, v2di, v8hf, v4sf, v8bf)
VAR7 (LOAD1, vld1_x3, v16qi, v8hi, v4si, v2di, v8hf, v4sf, v8bf)
+VAR7 (LOAD1, vld1_x4, v16qi, v8hi, v4si, v2di, v8hf, v4sf, v8bf)
VAR13 (LOAD1, vld1,
v8qi, v4hi, v4hf, v2si, v2sf, v16qi, v8hi, v8hf, v4si, v4sf, v2di,
v4bf, v8bf)
diff --git a/gcc/config/arm/neon.md b/gcc/config/arm/neon.md
index b37d95f1fa0..62decab37a2 100644
--- a/gcc/config/arm/neon.md
+++ b/gcc/config/arm/neon.md
@@ -4994,6 +4994,36 @@ if (BYTES_BIG_ENDIAN)
[(set_attr "type" "neon_load1_3reg<q>")]
)
+(define_insn "neon_vld1_x4<mode>"
+ [(set (match_operand:XI 0 "s_register_operand" "=w")
+ (unspec:XI [(match_operand:OI 1 "neon_struct_operand" "Um")
+ (unspec:VQXBF [(const_int 0)] UNSPEC_VSTRUCTDUMMY)]
+ UNSPEC_VLD4A))]
+ "TARGET_NEON"
+{
+ int regno = REGNO (operands[0]);
+ rtx ops[5];
+ ops[0] = gen_rtx_REG (DImode, regno);
+ ops[1] = gen_rtx_REG (DImode, regno + 2);
+ ops[2] = gen_rtx_REG (DImode, regno + 4);
+ ops[3] = gen_rtx_REG (DImode, regno + 6);
+ ops[4] = operands[1];
+
+ output_asm_insn ("vld1.<V_sz_elem>\t{%P0, %P1, %P2, %P3}, %A4", ops);
+
+ ops[0] = gen_rtx_REG (DImode, regno + 8);
+ ops[1] = gen_rtx_REG (DImode, regno + 10);
+ ops[2] = gen_rtx_REG (DImode, regno + 12);
+ ops[3] = gen_rtx_REG (DImode, regno + 14);
+ ops[4] = operands[1];
+
+ output_asm_insn ("vld1.<V_sz_elem>\t{%P0, %P1, %P2, %P3}, %A4", ops);
+
+ return "";
+}
+ [(set_attr "type" "neon_load1_3reg<q>")]
+)
+
;; The lane numbers in the RTL are in GCC lane order, having been flipped
;; in arm_expand_neon_args. The lane numbers are restored to architectural
;; lane order here.
diff --git a/gcc/testsuite/gcc.target/arm/simd/vld1q_base_xN_1.c b/gcc/testsuite/gcc.target/arm/simd/vld1q_base_xN_1.c
index bfad282751b..9aae34cc18e 100644
--- a/gcc/testsuite/gcc.target/arm/simd/vld1q_base_xN_1.c
+++ b/gcc/testsuite/gcc.target/arm/simd/vld1q_base_xN_1.c
@@ -115,14 +115,73 @@ poly16x8x3_t test_vld1q_p16_x3 (poly16_t * a)
return vld1q_p16_x3 (a);
}
+uint8x16x4_t test_vld1q_u8_x4 (uint8_t * a)
+{
+ return vld1q_u8_x4 (a);
+}
+
+uint16x8x4_t test_vld1q_u16_x4 (uint16_t * a)
+{
+ return vld1q_u16_x4 (a);
+}
+
+uint32x4x4_t test_vld1q_u32_x4 (uint32_t * a)
+{
+ return vld1q_u32_x4 (a);
+}
+
+uint64x2x4_t test_vld1q_u64_x4 (uint64_t * a)
+{
+ return vld1q_u64_x4 (a);
+}
+
+int8x16x4_t test_vld1q_s8_x4 (int8_t * a)
+{
+ return vld1q_s8_x4 (a);
+}
+
+int16x8x4_t test_vld1q_s16_x4 (int16_t * a)
+{
+ return vld1q_s16_x4 (a);
+}
+
+int32x4x4_t test_vld1q_s32_x4 (int32_t * a)
+{
+ return vld1q_s32_x4 (a);
+}
+
+int64x2x4_t test_vld1q_s64_x4 (int64_t * a)
+{
+ return vld1q_s64_x4 (a);
+}
+
+float32x4x4_t test_vld1q_f32_x4 (float32_t * a)
+{
+ return vld1q_f32_x4 (a);
+}
+
+poly8x16x4_t test_vld1q_p8_x4 (poly8_t * a)
+{
+ return vld1q_p8_x4 (a);
+}
+
+poly16x8x4_t test_vld1q_p16_x4 (poly16_t * a)
+{
+ return vld1q_p16_x4 (a);
+}
+
/* { dg-final { scan-assembler-times {vld1.8\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+\]\n} 3 } } */
/* { dg-final { scan-assembler-times {vld1.8\t\{d[0-9]+, d[0-9]+, d[0-9]+\}, \[r[0-9]+\]\n} 6 } } */
+/* { dg-final { scan-assembler-times {vld1.8\t\{d[0-9]+, d[0-9]+, d[0-9]+, d[0-9]+\}, \[r[0-9]+\]\n} 6 } } */
/* { dg-final { scan-assembler-times {vld1.16\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+\]\n} 3 } } */
/* { dg-final { scan-assembler-times {vld1.16\t\{d[0-9]+, d[0-9]+, d[0-9]+\}, \[r[0-9]+\]\n} 6 } } */
+/* { dg-final { scan-assembler-times {vld1.16\t\{d[0-9]+, d[0-9]+, d[0-9]+, d[0-9]+\}, \[r[0-9]+\]\n} 6 } } */
/* { dg-final { scan-assembler-times {vld1.32\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+\]\n} 3 } } */
/* { dg-final { scan-assembler-times {vld1.32\t\{d[0-9]+, d[0-9]+, d[0-9]+\}, \[r[0-9]+\]\n} 6 } } */
+/* { dg-final { scan-assembler-times {vld1.32\t\{d[0-9]+, d[0-9]+, d[0-9]+, d[0-9]+\}, \[r[0-9]+\]\n} 6 } } */
/* { dg-final { scan-assembler-times {vld1.64\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+:64\]\n} 2 } } */
/* { dg-final { scan-assembler-times {vld1.64\t\{d[0-9]+, d[0-9]+, d[0-9]+\}, \[r[0-9]+:64\]\n} 4 } } */
+/* { dg-final { scan-assembler-times {vld1.64\t\{d[0-9]+, d[0-9]+, d[0-9]+, d[0-9]+\}, \[r[0-9]+:64\]\n} 4 } } */
diff --git a/gcc/testsuite/gcc.target/arm/simd/vld1q_bf16_xN_1.c b/gcc/testsuite/gcc.target/arm/simd/vld1q_bf16_xN_1.c
index 4138fe951ee..fd86723f146 100644
--- a/gcc/testsuite/gcc.target/arm/simd/vld1q_bf16_xN_1.c
+++ b/gcc/testsuite/gcc.target/arm/simd/vld1q_bf16_xN_1.c
@@ -15,5 +15,11 @@ bfloat16x8x3_t test_vld1q_bf16_x3 (bfloat16_t * a)
return vld1q_bf16_x3 (a);
}
+bfloat16x8x4_t test_vld1q_bf16_x4 (bfloat16_t * a)
+{
+ return vld1q_bf16_x4 (a);
+}
+
/* { dg-final { scan-assembler-times {vld1.16\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+\]\n} 1 } } */
/* { dg-final { scan-assembler-times {vld1.16\t\{d[0-9]+, d[0-9]+, d[0-9]+\}, \[r[0-9]+\]\n} 2 } } */
+/* { dg-final { scan-assembler-times {vld1.16\t\{d[0-9]+, d[0-9]+, d[0-9]+, d[0-9]+\}, \[r[0-9]+\]\n} 2 } } */
diff --git a/gcc/testsuite/gcc.target/arm/simd/vld1q_fp16_xN_1.c b/gcc/testsuite/gcc.target/arm/simd/vld1q_fp16_xN_1.c
index 01640d7cc1f..2de3495f1da 100644
--- a/gcc/testsuite/gcc.target/arm/simd/vld1q_fp16_xN_1.c
+++ b/gcc/testsuite/gcc.target/arm/simd/vld1q_fp16_xN_1.c
@@ -15,5 +15,11 @@ float16x8x3_t test_vld1q_f16_x3 (float16_t * a)
return vld1q_f16_x3 (a);
}
+float16x8x4_t test_vld1q_f16_x4 (float16_t * a)
+{
+ return vld1q_f16_x4 (a);
+}
+
/* { dg-final { scan-assembler-times {vld1.16\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+\]\n} 1 } } */
/* { dg-final { scan-assembler-times {vld1.16\t\{d[0-9]+, d[0-9]+, d[0-9]+\}, \[r[0-9]+\]\n} 2 } } */
+/* { dg-final { scan-assembler-times {vld1.16\t\{d[0-9]+, d[0-9]+, d[0-9]+, d[0-9]+\}, \[r[0-9]+\]\n} 2 } } */
diff --git a/gcc/testsuite/gcc.target/arm/simd/vld1q_p64_xN_1.c b/gcc/testsuite/gcc.target/arm/simd/vld1q_p64_xN_1.c
index ae2ab36df57..521b784e8e8 100644
--- a/gcc/testsuite/gcc.target/arm/simd/vld1q_p64_xN_1.c
+++ b/gcc/testsuite/gcc.target/arm/simd/vld1q_p64_xN_1.c
@@ -15,5 +15,11 @@ poly64x2x3_t test_vld1q_p64_x3 (poly64_t * a)
return vld1q_p64_x3 (a);
}
+poly64x2x4_t test_vld1q_p64_x4 (poly64_t * a)
+{
+ return vld1q_p64_x4 (a);
+}
+
/* { dg-final { scan-assembler-times {vld1.64\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+:64\]\n} 1 } } */
/* { dg-final { scan-assembler-times {vld1.64\t\{d[0-9]+, d[0-9]+, d[0-9]+\}, \[r[0-9]+:64\]\n} 2 } } */
+/* { dg-final { scan-assembler-times {vld1.64\t\{d[0-9]+, d[0-9]+, d[0-9]+, d[0-9]+\}, \[r[0-9]+:64\]\n} 2 } } */
--
2.25.1
^ permalink raw reply [flat|nested] 6+ messages in thread
* [PING] [PATCH 1/3] [GCC] arm: vld1q_types_x2 ACLE intrinsics
2023-10-06 9:49 ` [PATCH 1/3] [GCC] arm: vld1q_types_x2 " Ezra.Sitorus
@ 2023-10-24 10:20 ` Ezra Sitorus
0 siblings, 0 replies; 6+ messages in thread
From: Ezra Sitorus @ 2023-10-24 10:20 UTC (permalink / raw)
To: gcc-patches; +Cc: Richard Earnshaw, Kyrylo Tkachov
Ping
________________________________________
From: Ezra.Sitorus@arm.com <Ezra.Sitorus@arm.com>
Sent: Friday, October 6, 2023 10:49 AM
To: gcc-patches@gcc.gnu.org
Cc: Richard Earnshaw; Kyrylo Tkachov
Subject: [PATCH 1/3] [GCC] arm: vld1q_types_x2 ACLE intrinsics
From: Ezra Sitorus <ezra.sitorus@arm.com>
This patch is part of a series of patches implementing the _xN variants of the vld1q intrinsic for arm32.
This patch adds the _x2 variants of the vld1q intrinsic. Tests use xN so that the latter variants (_x3, _x4) could be added.
ACLE documents are at https://developer.arm.com/documentation/ihi0053/latest/
ISA documents are at https://developer.arm.com/documentation/ddi0487/latest/
gcc/ChangeLog:
* config/arm/arm_neon.h
(vld1q_u8_x2, vld1q_u16_x2, vld1q_u32_x2, vld1q_u64_x2): New.
(vld1q_s8_x2, vld1q_s16_x2, vld1q_s32_x2, vld1q_s64_x2): New.
(vld1q_f16_x2, vld1q_f32_x2): New.
(vld1q_p8_x2, vld1q_p16_x2, vld1q_p64_x2): New.
(vld1q_bf16_x2): New.
* config/arm/arm_neon_builtins.def (vld1_x2): New entries.
* config/arm/neon.md (vld1_x2<mode>): New.
gcc/testsuite/ChangeLog:
* gcc.target/arm/simd/vld1q_base_xN_1.c: Add new test.
* gcc.target/arm/simd/vld1q_bf16_xN_1.c: Add new test.
* gcc.target/arm/simd/vld1q_fp16_xN_1.c: Add new test.
* gcc.target/arm/simd/vld1q_p64_xN_1.c: Add new test.
---
gcc/config/arm/arm_neon.h | 128 ++++++++++++++++++
gcc/config/arm/arm_neon_builtins.def | 1 +
gcc/config/arm/neon.md | 10 ++
.../gcc.target/arm/simd/vld1q_base_xN_1.c | 67 +++++++++
.../gcc.target/arm/simd/vld1q_bf16_xN_1.c | 13 ++
.../gcc.target/arm/simd/vld1q_fp16_xN_1.c | 14 ++
.../gcc.target/arm/simd/vld1q_p64_xN_1.c | 14 ++
7 files changed, 247 insertions(+)
create mode 100644 gcc/testsuite/gcc.target/arm/simd/vld1q_base_xN_1.c
create mode 100644 gcc/testsuite/gcc.target/arm/simd/vld1q_bf16_xN_1.c
create mode 100644 gcc/testsuite/gcc.target/arm/simd/vld1q_fp16_xN_1.c
create mode 100644 gcc/testsuite/gcc.target/arm/simd/vld1q_p64_xN_1.c
diff --git a/gcc/config/arm/arm_neon.h b/gcc/config/arm/arm_neon.h
index cdfdb44259a..3eb41c6bdc8 100644
--- a/gcc/config/arm/arm_neon.h
+++ b/gcc/config/arm/arm_neon.h
@@ -10403,6 +10403,15 @@ vld1q_p64 (const poly64_t * __a)
return (poly64x2_t)__builtin_neon_vld1v2di ((const __builtin_neon_di *) __a);
}
+__extension__ extern __inline poly64x2x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vld1q_p64_x2 (const poly64_t * __a)
+{
+ union { poly64x2x2_t __i; __builtin_neon_oi __o; } __rv;
+ __rv.__o = __builtin_neon_vld1_x2v2di ((const __builtin_neon_di *) __a);
+ return __rv.__i;
+}
+
#pragma GCC pop_options
__extension__ extern __inline int8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
@@ -10432,6 +10441,42 @@ vld1q_s64 (const int64_t * __a)
return (int64x2_t)__builtin_neon_vld1v2di ((const __builtin_neon_di *) __a);
}
+__extension__ extern __inline int8x16x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vld1q_s8_x2 (const int8_t * __a)
+{
+ union { int8x16x2_t __i; __builtin_neon_oi __o; } __rv;
+ __rv.__o = __builtin_neon_vld1_x2v16qi ((const __builtin_neon_qi *) __a);
+ return __rv.__i;
+}
+
+__extension__ extern __inline int16x8x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vld1q_s16_x2 (const int16_t * __a)
+{
+ union { int16x8x2_t __i; __builtin_neon_oi __o; } __rv;
+ __rv.__o = __builtin_neon_vld1_x2v8hi ((const __builtin_neon_hi *) __a);
+ return __rv.__i;
+}
+
+__extension__ extern __inline int32x4x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vld1q_s32_x2 (const int32_t * __a)
+{
+ union { int32x4x2_t __i; __builtin_neon_oi __o; } __rv;
+ __rv.__o = __builtin_neon_vld1_x2v4si ((const __builtin_neon_si *) __a);
+ return __rv.__i;
+}
+
+__extension__ extern __inline int64x2x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vld1q_s64_x2 (const int64_t * __a)
+{
+ union { int64x2x2_t __i; __builtin_neon_oi __o; } __rv;
+ __rv.__o = __builtin_neon_vld1_x2v2di ((const __builtin_neon_di *) __a);
+ return __rv.__i;
+}
+
#if defined (__ARM_FP16_FORMAT_IEEE) || defined (__ARM_FP16_FORMAT_ALTERNATIVE)
__extension__ extern __inline float16x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
@@ -10448,6 +10493,26 @@ vld1q_f32 (const float32_t * __a)
return (float32x4_t)__builtin_neon_vld1v4sf ((const __builtin_neon_sf *) __a);
}
+#if defined (__ARM_FP16_FORMAT_IEEE) || defined (__ARM_FP16_FORMAT_ALTERNATIVE)
+__extension__ extern __inline float16x8x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vld1q_f16_x2 (const float16_t * __a)
+{
+ union { float16x8x2_t __i; __builtin_neon_oi __o; } __rv;
+ __rv.__o = __builtin_neon_vld1_x2v8hf (__a);
+ return __rv.__i;
+}
+#endif
+
+__extension__ extern __inline float32x4x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vld1q_f32_x2 (const float32_t * __a)
+{
+ union { float32x4x2_t __i; __builtin_neon_oi __o; } __rv;
+ __rv.__o = __builtin_neon_vld1_x2v4sf ((const __builtin_neon_sf *) __a);
+ return __rv.__i;
+}
+
__extension__ extern __inline uint8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vld1q_u8 (const uint8_t * __a)
@@ -10476,6 +10541,42 @@ vld1q_u64 (const uint64_t * __a)
return (uint64x2_t)__builtin_neon_vld1v2di ((const __builtin_neon_di *) __a);
}
+__extension__ extern __inline uint8x16x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vld1q_u8_x2 (const uint8_t * __a)
+{
+ union { uint8x16x2_t __i; __builtin_neon_oi __o; } __rv;
+ __rv.__o = __builtin_neon_vld1_x2v16qi ((const __builtin_neon_qi *) __a);
+ return __rv.__i;
+}
+
+__extension__ extern __inline uint16x8x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vld1q_u16_x2 (const uint16_t * __a)
+{
+ union { uint16x8x2_t __i; __builtin_neon_oi __o; } __rv;
+ __rv.__o = __builtin_neon_vld1_x2v8hi ((const __builtin_neon_hi *) __a);
+ return __rv.__i;
+}
+
+__extension__ extern __inline uint32x4x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vld1q_u32_x2 (const uint32_t * __a)
+{
+ union { uint32x4x2_t __i; __builtin_neon_oi __o; } __rv;
+ __rv.__o = __builtin_neon_vld1_x2v4si ((const __builtin_neon_si *) __a);
+ return __rv.__i;
+}
+
+__extension__ extern __inline uint64x2x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vld1q_u64_x2 (const uint64_t * __a)
+{
+ union { uint64x2x2_t __i; __builtin_neon_oi __o; } __rv;
+ __rv.__o = __builtin_neon_vld1_x2v2di ((const __builtin_neon_di *) __a);
+ return __rv.__i;
+}
+
__extension__ extern __inline poly8x16_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vld1q_p8 (const poly8_t * __a)
@@ -10490,6 +10591,24 @@ vld1q_p16 (const poly16_t * __a)
return (poly16x8_t)__builtin_neon_vld1v8hi ((const __builtin_neon_hi *) __a);
}
+__extension__ extern __inline poly8x16x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vld1q_p8_x2 (const poly8_t * __a)
+{
+ union { poly8x16x2_t __i; __builtin_neon_oi __o; } __rv;
+ __rv.__o = __builtin_neon_vld1_x2v16qi ((const __builtin_neon_qi *) __a);
+ return __rv.__i;
+}
+
+__extension__ extern __inline poly16x8x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vld1q_p16_x2 (const poly16_t * __a)
+{
+ union { poly16x8x2_t __i; __builtin_neon_oi __o; } __rv;
+ __rv.__o = __builtin_neon_vld1_x2v8hi ((const __builtin_neon_hi *) __a);
+ return __rv.__i;
+}
+
__extension__ extern __inline int8x8_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vld1_lane_s8 (const int8_t * __a, int8x8_t __b, const int __c)
@@ -19782,6 +19901,15 @@ vld1q_bf16 (const bfloat16_t * __ptr)
return __builtin_neon_vld1v8bf (__ptr);
}
+__extension__ extern __inline bfloat16x8x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vld1q_bf16_x2 (const bfloat16_t * __ptr)
+{
+ union { bfloat16x8x2_t __i; __builtin_neon_oi __o; } __rv;
+ __rv.__o = __builtin_neon_vld1_x2v8bf ((const __builtin_neon_bf *) __ptr);
+ return __rv.__i;
+}
+
__extension__ extern __inline bfloat16x4x2_t
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vld2_bf16 (bfloat16_t const * __ptr)
diff --git a/gcc/config/arm/arm_neon_builtins.def b/gcc/config/arm/arm_neon_builtins.def
index 94b15238123..5fadd255c18 100644
--- a/gcc/config/arm/arm_neon_builtins.def
+++ b/gcc/config/arm/arm_neon_builtins.def
@@ -298,6 +298,7 @@ VAR1 (TERNOP, vtbx1, v8qi)
VAR1 (TERNOP, vtbx2, v8qi)
VAR1 (TERNOP, vtbx3, v8qi)
VAR1 (TERNOP, vtbx4, v8qi)
+VAR7 (LOAD1, vld1_x2, v16qi, v8hi, v4si, v2di, v8hf, v4sf, v8bf)
VAR13 (LOAD1, vld1,
v8qi, v4hi, v4hf, v2si, v2sf, v16qi, v8hi, v8hf, v4si, v4sf, v2di,
v4bf, v8bf)
diff --git a/gcc/config/arm/neon.md b/gcc/config/arm/neon.md
index d213369ffc3..55049ea549f 100644
--- a/gcc/config/arm/neon.md
+++ b/gcc/config/arm/neon.md
@@ -4957,6 +4957,16 @@ if (BYTES_BIG_ENDIAN)
[(set_attr "type" "neon_load1_1reg<q>")]
)
+(define_insn "neon_vld1_x2<mode>"
+ [(set (match_operand:OI 0 "s_register_operand" "=w")
+ (unspec:OI [(match_operand:OI 1 "neon_struct_operand" "Um")
+ (unspec:VQXBF [(const_int 0)] UNSPEC_VSTRUCTDUMMY)]
+ UNSPEC_VLD1))]
+ "TARGET_NEON"
+ "vld1.<V_sz_elem>\t%h0, %A1"
+ [(set_attr "type" "neon_load1_2reg<q>")]
+)
+
;; The lane numbers in the RTL are in GCC lane order, having been flipped
;; in arm_expand_neon_args. The lane numbers are restored to architectural
;; lane order here.
diff --git a/gcc/testsuite/gcc.target/arm/simd/vld1q_base_xN_1.c b/gcc/testsuite/gcc.target/arm/simd/vld1q_base_xN_1.c
new file mode 100644
index 00000000000..1d31777afdf
--- /dev/null
+++ b/gcc/testsuite/gcc.target/arm/simd/vld1q_base_xN_1.c
@@ -0,0 +1,67 @@
+/* { dg-do assemble } */
+/* { dg-require-effective-target arm_neon_ok } */
+/* { dg-options "-save-temps -O2" } */
+/* { dg-add-options arm_neon } */
+
+#include "arm_neon.h"
+
+uint8x16x2_t test_vld1q_u8_x2 (uint8_t * a)
+{
+ return vld1q_u8_x2 (a);
+}
+
+uint16x8x2_t test_vld1q_u16_x2 (uint16_t * a)
+{
+ return vld1q_u16_x2 (a);
+}
+
+uint32x4x2_t test_vld1q_u32_x2 (uint32_t * a)
+{
+ return vld1q_u32_x2 (a);
+}
+
+uint64x2x2_t test_vld1q_u64_x2 (uint64_t * a)
+{
+ return vld1q_u64_x2 (a);
+}
+
+int8x16x2_t test_vld1q_s8_x2 (int8_t * a)
+{
+ return vld1q_s8_x2 (a);
+}
+
+int16x8x2_t test_vld1q_s16_x2 (int16_t * a)
+{
+ return vld1q_s16_x2 (a);
+}
+
+int32x4x2_t test_vld1q_s32_x2 (int32_t * a)
+{
+ return vld1q_s32_x2 (a);
+}
+
+int64x2x2_t test_vld1q_s64_x2 (int64_t * a)
+{
+ return vld1q_s64_x2 (a);
+}
+
+float32x4x2_t test_vld1q_f32_x2 (float32_t * a)
+{
+ return vld1q_f32_x2 (a);
+}
+
+poly8x16x2_t test_vld1q_p8_x2 (poly8_t * a)
+{
+ return vld1q_p8_x2 (a);
+}
+
+poly16x8x2_t test_vld1q_p16_x2 (poly16_t * a)
+{
+ return vld1q_p16_x2 (a);
+}
+
+/* { dg-final { scan-assembler-times {vld1.8\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+\]\n} 3 } } */
+/* { dg-final { scan-assembler-times {vld1.16\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+\]\n} 3 } } */
+/* { dg-final { scan-assembler-times {vld1.32\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+\]\n} 3 } } */
+/* { dg-final { scan-assembler-times {vld1.64\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+:64\]\n} 2 } } */
+
diff --git a/gcc/testsuite/gcc.target/arm/simd/vld1q_bf16_xN_1.c b/gcc/testsuite/gcc.target/arm/simd/vld1q_bf16_xN_1.c
new file mode 100644
index 00000000000..5f6fc98640e
--- /dev/null
+++ b/gcc/testsuite/gcc.target/arm/simd/vld1q_bf16_xN_1.c
@@ -0,0 +1,13 @@
+/* { dg-do assemble } */
+/* { dg-require-effective-target arm_v8_2a_bf16_neon_ok } */
+/* { dg-options "-save-temps -O2" } */
+/* { dg-add-options arm_v8_2a_bf16_neon } */
+
+#include "arm_neon.h"
+
+bfloat16x8x2_t test_vld1q_bf16_x2 (bfloat16_t * a)
+{
+ return vld1q_bf16_x2 (a);
+}
+
+/* { dg-final { scan-assembler-times {vld1.16\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+\]\n} 1 } } */
diff --git a/gcc/testsuite/gcc.target/arm/simd/vld1q_fp16_xN_1.c b/gcc/testsuite/gcc.target/arm/simd/vld1q_fp16_xN_1.c
new file mode 100644
index 00000000000..aecf491a4de
--- /dev/null
+++ b/gcc/testsuite/gcc.target/arm/simd/vld1q_fp16_xN_1.c
@@ -0,0 +1,14 @@
+/* { dg-do assemble } */
+/* { dg-require-effective-target arm_neon_fp16_ok } */
+/* { dg-options "-save-temps -O2" } */
+/* { dg-add-options arm_neon_fp16 } */
+
+#include "arm_neon.h"
+
+float16x8x2_t test_vld1q_f16_x2 (float16_t * a)
+{
+ return vld1q_f16_x2 (a);
+}
+
+/* { dg-final { scan-assembler-times {vld1.16\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+\]\n} 1 } } */
+
diff --git a/gcc/testsuite/gcc.target/arm/simd/vld1q_p64_xN_1.c b/gcc/testsuite/gcc.target/arm/simd/vld1q_p64_xN_1.c
new file mode 100644
index 00000000000..04ceb5e4a24
--- /dev/null
+++ b/gcc/testsuite/gcc.target/arm/simd/vld1q_p64_xN_1.c
@@ -0,0 +1,14 @@
+/* { dg-do assemble } */
+/* { dg-require-effective-target arm_crypto_ok } */
+/* { dg-options "-save-temps -O2" } */
+/* { dg-add-options arm_crypto } */
+
+#include "arm_neon.h"
+
+poly64x2x2_t test_vld1q_p64_x2 (poly64_t * a)
+{
+ return vld1q_p64_x2 (a);
+}
+
+/* { dg-final { scan-assembler-times {vld1.64\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+:64\]\n} 1 } } */
+
--
2.25.1
^ permalink raw reply [flat|nested] 6+ messages in thread
* Re: [PATCH 0/3] [GCC] arm: vld1q_types_xN ACLE intrinsics
2023-10-06 9:49 [PATCH 0/3] [GCC] arm: vld1q_types_xN ACLE intrinsics Ezra.Sitorus
` (2 preceding siblings ...)
2023-10-06 9:49 ` [PATCH 3/3] [GCC] arm: vld1q_types_x4 " Ezra.Sitorus
@ 2023-11-27 15:06 ` Richard Earnshaw
3 siblings, 0 replies; 6+ messages in thread
From: Richard Earnshaw @ 2023-11-27 15:06 UTC (permalink / raw)
To: Ezra.Sitorus, gcc-patches; +Cc: richard.earnshaw, kyrylo.tkachov
On 06/10/2023 10:49, Ezra.Sitorus@arm.com wrote:
> Add xN variants of vld1q_types intrinsic.
>
>
These patches are all OK, but please fix commit message formatting in
line with the comments on the earlier series.
R.
^ permalink raw reply [flat|nested] 6+ messages in thread
end of thread, other threads:[~2023-11-27 15:06 UTC | newest]
Thread overview: 6+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2023-10-06 9:49 [PATCH 0/3] [GCC] arm: vld1q_types_xN ACLE intrinsics Ezra.Sitorus
2023-10-06 9:49 ` [PATCH 1/3] [GCC] arm: vld1q_types_x2 " Ezra.Sitorus
2023-10-24 10:20 ` [PING] " Ezra Sitorus
2023-10-06 9:49 ` [PATCH 2/3] [GCC] arm: vld1q_types_x3 " Ezra.Sitorus
2023-10-06 9:49 ` [PATCH 3/3] [GCC] arm: vld1q_types_x4 " Ezra.Sitorus
2023-11-27 15:06 ` [PATCH 0/3] [GCC] arm: vld1q_types_xN " Richard Earnshaw
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).