* [PATCH 0/3] [GCC] arm: vst1_types_xN ACLE intrinsics
@ 2023-10-06 11:55 Ezra.Sitorus
2023-10-06 11:55 ` [PATCH 1/3] [GCC] arm: vst1_types_x2 " Ezra.Sitorus
` (2 more replies)
0 siblings, 3 replies; 8+ messages in thread
From: Ezra.Sitorus @ 2023-10-06 11:55 UTC (permalink / raw)
To: gcc-patches; +Cc: richard.earnshaw, kyrylo.tkachov
Add xN variants of vst1_types intrinsic.
^ permalink raw reply [flat|nested] 8+ messages in thread
* [PATCH 1/3] [GCC] arm: vst1_types_x2 ACLE intrinsics
2023-10-06 11:55 [PATCH 0/3] [GCC] arm: vst1_types_xN ACLE intrinsics Ezra.Sitorus
@ 2023-10-06 11:55 ` Ezra.Sitorus
2023-11-27 14:56 ` Richard Earnshaw
2023-10-06 11:55 ` [PATCH 2/3] [GCC] arm: vst1_types_x3 " Ezra.Sitorus
2023-10-06 11:56 ` [PATCH 3/3] [GCC] arm: vst1_types_x4 " Ezra.Sitorus
2 siblings, 1 reply; 8+ messages in thread
From: Ezra.Sitorus @ 2023-10-06 11:55 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 vst1 intrinsic for arm32.
This patch adds the _x2 variants of the vst1 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
(vst1_u8_x2, vst1_u16_x2, vst1_u32_x2, vst1_u64_x32): New.
(vst1_s8_x2, vst1_s16_x2, vst1_s32_x2, vst1_s64_x2): New.
(vst1_f16_x2, vst1_f32_x2): New.
(vst1_p8_x2, vst1_p16_x2, vst1_p64_x2): New.
(vst1_bf16_x2): New.
* config/arm/arm_neon_builtins.def (vst1_x2): New entries.
* config/arm/neon.md (vst1_x2<mode>): New.
gcc/testsuite/ChangeLog:
* gcc.target/arm/simd/vst1_base_xN_1.c: Add new tests.
* gcc.target/arm/simd/vst1_bf16_xN_1.c: Add new tests.
* gcc.target/arm/simd/vst1_fp16_xN_1.c: Add new tests.
* gcc.target/arm/simd/vst1_p64_xN_1.c: Add new tests.
---
gcc/config/arm/arm_neon.h | 114 ++++++++++++++++++
gcc/config/arm/arm_neon_builtins.def | 1 +
gcc/config/arm/neon.md | 10 ++
.../gcc.target/arm/simd/vst1_base_xN_1.c | 67 ++++++++++
.../gcc.target/arm/simd/vst1_bf16_xN_1.c | 13 ++
.../gcc.target/arm/simd/vst1_fp16_xN_1.c | 13 ++
.../gcc.target/arm/simd/vst1_p64_xN_1.c | 13 ++
7 files changed, 231 insertions(+)
create mode 100644 gcc/testsuite/gcc.target/arm/simd/vst1_base_xN_1.c
create mode 100644 gcc/testsuite/gcc.target/arm/simd/vst1_bf16_xN_1.c
create mode 100644 gcc/testsuite/gcc.target/arm/simd/vst1_fp16_xN_1.c
create mode 100644 gcc/testsuite/gcc.target/arm/simd/vst1_p64_xN_1.c
diff --git a/gcc/config/arm/arm_neon.h b/gcc/config/arm/arm_neon.h
index c03be9912f8..4bd6093281b 100644
--- a/gcc/config/arm/arm_neon.h
+++ b/gcc/config/arm/arm_neon.h
@@ -11242,6 +11242,14 @@ vst1_p64 (poly64_t * __a, poly64x1_t __b)
__builtin_neon_vst1di ((__builtin_neon_di *) __a, __b);
}
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vst1_p64_x2 (poly64_t * __a, poly64x1x2_t __b)
+{
+ union { poly64x1x2_t __i; __builtin_neon_ti __o; } __bu = { __b };
+ __builtin_neon_vst1_x2di ((__builtin_neon_di *) __a, __bu.__o);
+}
+
#pragma GCC pop_options
__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
@@ -11271,6 +11279,38 @@ vst1_s64 (int64_t * __a, int64x1_t __b)
__builtin_neon_vst1di ((__builtin_neon_di *) __a, __b);
}
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vst1_s8_x2 (int8_t * __a, int8x8x2_t __b)
+{
+ union { int8x8x2_t __i; __builtin_neon_ti __o; } __bu = { __b };
+ __builtin_neon_vst1_x2v8qi ((__builtin_neon_qi *) __a, __bu.__o);
+}
+
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vst1_s16_x2 (int16_t * __a, int16x4x2_t __b)
+{
+ union { int16x4x2_t __i; __builtin_neon_ti __o; } __bu = { __b };
+ __builtin_neon_vst1_x2v4hi ((__builtin_neon_hi *) __a, __bu.__o);
+}
+
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vst1_s32_x2 (int32_t * __a, int32x2x2_t __b)
+{
+ union { int32x2x2_t __i; __builtin_neon_ti __o; } __bu = { __b };
+ __builtin_neon_vst1_x2v2si ((__builtin_neon_si *) __a, __bu.__o);
+}
+
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vst1_s64_x2 (int64_t * __a, int64x1x2_t __b)
+{
+ union { int64x1x2_t __i; __builtin_neon_ti __o; } __bu = { __b };
+ __builtin_neon_vst1_x2di ((__builtin_neon_di *) __a, __bu.__o);
+}
+
#if defined (__ARM_FP16_FORMAT_IEEE) || defined (__ARM_FP16_FORMAT_ALTERNATIVE)
__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
@@ -11287,6 +11327,24 @@ vst1_f32 (float32_t * __a, float32x2_t __b)
__builtin_neon_vst1v2sf ((__builtin_neon_sf *) __a, __b);
}
+#if defined (__ARM_FP16_FORMAT_IEEE) || defined (__ARM_FP16_FORMAT_ALTERNATIVE)
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vst1_f16_x2 (float16_t * __a, float16x4x2_t __b)
+{
+ union { float16x4x2_t __i; __builtin_neon_ti __o; } __bu = { __b };
+ __builtin_neon_vst1_x2v4hf (__a, __bu.__o);
+}
+#endif
+
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vst1_f32_x2 (float32_t * __a, float32x2x2_t __b)
+{
+ union { float32x2x2_t __i; __builtin_neon_ti __o; } __bu = { __b };
+ __builtin_neon_vst1_x2v2sf ((__builtin_neon_sf *) __a, __bu.__o);
+}
+
__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vst1_u8 (uint8_t * __a, uint8x8_t __b)
@@ -11315,6 +11373,38 @@ vst1_u64 (uint64_t * __a, uint64x1_t __b)
__builtin_neon_vst1di ((__builtin_neon_di *) __a, (int64x1_t) __b);
}
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vst1_u8_x2 (uint8_t * __a, uint8x8x2_t __b)
+{
+ union { uint8x8x2_t __i; __builtin_neon_ti __o; } __bu = { __b };
+ __builtin_neon_vst1_x2v8qi ((__builtin_neon_qi *) __a, __bu.__o);
+}
+
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vst1_u16_x2 (uint16_t * __a, uint16x4x2_t __b)
+{
+ union { uint16x4x2_t __i; __builtin_neon_ti __o; } __bu = { __b };
+ __builtin_neon_vst1_x2v4hi ((__builtin_neon_hi *) __a, __bu.__o);
+}
+
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vst1_u32_x2 (uint32_t * __a, uint32x2x2_t __b)
+{
+ union { uint32x2x2_t __i; __builtin_neon_ti __o; } __bu = { __b };
+ __builtin_neon_vst1_x2v2si ((__builtin_neon_si *) __a, __bu.__o);
+}
+
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vst1_u64_x2 (uint64_t * __a, uint64x1x2_t __b)
+{
+ union { uint64x1x2_t __i; __builtin_neon_ti __o; } __bu = { __b };
+ __builtin_neon_vst1_x2di ((__builtin_neon_di *) __a, __bu.__o);
+}
+
__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vst1_p8 (poly8_t * __a, poly8x8_t __b)
@@ -11329,6 +11419,22 @@ vst1_p16 (poly16_t * __a, poly16x4_t __b)
__builtin_neon_vst1v4hi ((__builtin_neon_hi *) __a, (int16x4_t) __b);
}
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vst1_p8_x2 (poly8_t * __a, poly8x8x2_t __b)
+{
+ union { poly8x8x2_t __i; __builtin_neon_ti __o; } __bu = { __b };
+ __builtin_neon_vst1_x2v8qi ((__builtin_neon_qi *) __a, __bu.__o);
+}
+
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vst1_p16_x2 (poly16_t * __a, poly16x4x2_t __b)
+{
+ union { poly16x4x2_t __i; __builtin_neon_ti __o; } __bu = { __b };
+ __builtin_neon_vst1_x2v4hi ((__builtin_neon_hi *) __a, __bu.__o);
+}
+
#pragma GCC push_options
#pragma GCC target ("fpu=crypto-neon-fp-armv8")
__extension__ extern __inline void
@@ -20070,6 +20176,14 @@ vst1_bf16 (bfloat16_t * __a, bfloat16x4_t __b)
__builtin_neon_vst1v4bf (__a, __b);
}
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vst1_bf16_x2 (bfloat16_t * __a, bfloat16x4x2_t __b)
+{
+ union { bfloat16x4x2_t __i; __builtin_neon_ti __o; } __bu = { __b };
+ __builtin_neon_vst1_x2v4hf ((__builtin_neon_bf *) __a, __bu.__o);
+}
+
__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vst1q_bf16 (bfloat16_t * __a, bfloat16x8_t __b)
diff --git a/gcc/config/arm/arm_neon_builtins.def b/gcc/config/arm/arm_neon_builtins.def
index f4001b298c4..7aef6f958cd 100644
--- a/gcc/config/arm/arm_neon_builtins.def
+++ b/gcc/config/arm/arm_neon_builtins.def
@@ -308,6 +308,7 @@ VAR12 (LOAD1LANE, vld1_lane,
v8qi, v4hi, v2si, v2sf, di, v16qi, v8hi, v4si, v4sf, v2di, v4bf, v8bf)
VAR10 (LOAD1, vld1_dup,
v8qi, v4hi, v2si, v2sf, di, v16qi, v8hi, v4si, v4sf, v2di)
+VAR7 (STORE1, vst1_x2, v8qi, v4hi, v2si, di, v4hf, v2sf, v4bf)
VAR14 (STORE1, vst1,
v8qi, v4hi, v4hf, v2si, v2sf, di, v16qi, v8hi, v8hf, v4si, v4sf, v2di,
v4bf, v8bf)
diff --git a/gcc/config/arm/neon.md b/gcc/config/arm/neon.md
index 62decab37a2..7a10e2cb61e 100644
--- a/gcc/config/arm/neon.md
+++ b/gcc/config/arm/neon.md
@@ -5125,6 +5125,16 @@ if (BYTES_BIG_ENDIAN)
UNSPEC_VST1))]
"TARGET_NEON")
+(define_insn "neon_vst1_x2<mode>"
+ [(set (match_operand:TI 0 "neon_struct_operand" "=Um")
+ (unspec:TI [(match_operand:TI 1 "s_register_operand" "w")
+ (unspec:VDQX [(const_int 0)] UNSPEC_VSTRUCTDUMMY)]
+ UNSPEC_VST1))]
+ "TARGET_NEON"
+ "vst1.<V_sz_elem>\t%h1, %A0"
+ [(set_attr "type" "neon_store1_2reg<q>")]
+)
+
(define_insn "neon_vst1<mode>"
[(set (match_operand:VDQX 0 "neon_struct_operand" "=Um")
(unspec:VDQX [(match_operand:VDQX 1 "s_register_operand" "w")]
diff --git a/gcc/testsuite/gcc.target/arm/simd/vst1_base_xN_1.c b/gcc/testsuite/gcc.target/arm/simd/vst1_base_xN_1.c
new file mode 100644
index 00000000000..575897fa422
--- /dev/null
+++ b/gcc/testsuite/gcc.target/arm/simd/vst1_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"
+
+void test_vst1_u8_x2 (uint8_t * ptr, uint8x8x2_t val)
+{
+ vst1_u8_x2 (ptr, val);
+}
+
+void test_vst1_u16_x2 (uint16_t * ptr, uint16x4x2_t val)
+{
+ vst1_u16_x2 (ptr, val);
+}
+
+void test_vst1_u32_x2 (uint32_t * ptr, uint32x2x2_t val)
+{
+ vst1_u32_x2 (ptr, val);
+}
+
+void test_vst1_u64_x2 (uint64_t * ptr, uint64x1x2_t val)
+{
+ vst1_u64_x2 (ptr, val);
+}
+
+void test_vst1_s8_x2 (int8_t * ptr, int8x8x2_t val)
+{
+ vst1_s8_x2 (ptr, val);
+}
+
+void test_vst1_s16_x2 (int16_t * ptr, int16x4x2_t val)
+{
+ vst1_s16_x2 (ptr, val);
+}
+
+void test_vst1_s32_x2 (int32_t * ptr, int32x2x2_t val)
+{
+ vst1_s32_x2 (ptr, val);
+}
+
+void test_vst1_s64_x2 (int64_t * ptr, int64x1x2_t val)
+{
+ vst1_s64_x2 (ptr, val);
+}
+
+void test_vst1_f32_x2 (float32_t * ptr, float32x2x2_t val)
+{
+ vst1_f32_x2 (ptr, val);
+}
+
+void test_vst1_p8_x2 (poly8_t * ptr, poly8x8x2_t val)
+{
+ vst1_p8_x2 (ptr, val);
+}
+
+void test_vst1_p16_x2 (poly16_t * ptr, poly16x4x2_t val)
+{
+ vst1_p16_x2 (ptr, val);
+}
+
+
+/* { dg-final { scan-assembler-times {vst1.8\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+\]\n} 3 } } */
+/* { dg-final { scan-assembler-times {vst1.16\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+\]\n} 3 } } */
+/* { dg-final { scan-assembler-times {vst1.32\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+\]\n} 3 } } */
+/* { dg-final { scan-assembler-times {vst1.64\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+:64\]\n} 2 } } */
diff --git a/gcc/testsuite/gcc.target/arm/simd/vst1_bf16_xN_1.c b/gcc/testsuite/gcc.target/arm/simd/vst1_bf16_xN_1.c
new file mode 100644
index 00000000000..213fd20ee65
--- /dev/null
+++ b/gcc/testsuite/gcc.target/arm/simd/vst1_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"
+
+void test_vst1_bf16_x2 (bfloat16_t * ptr, bfloat16x4x2_t val)
+{
+ vst1_bf16_x2 (ptr, val);
+}
+
+/* { dg-final { scan-assembler-times {vst1.16\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+\]\n} 1 } } */
diff --git a/gcc/testsuite/gcc.target/arm/simd/vst1_fp16_xN_1.c b/gcc/testsuite/gcc.target/arm/simd/vst1_fp16_xN_1.c
new file mode 100644
index 00000000000..523aec92db2
--- /dev/null
+++ b/gcc/testsuite/gcc.target/arm/simd/vst1_fp16_xN_1.c
@@ -0,0 +1,13 @@
+/* { 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"
+
+void test_vst1_f16_x2 (float16_t * ptr, float16x4x2_t val)
+{
+ vst1_f16_x2 (ptr, val);
+}
+
+/* { dg-final { scan-assembler-times {vst1.16\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+\]\n} 1 } } */
diff --git a/gcc/testsuite/gcc.target/arm/simd/vst1_p64_xN_1.c b/gcc/testsuite/gcc.target/arm/simd/vst1_p64_xN_1.c
new file mode 100644
index 00000000000..f590ebd7b94
--- /dev/null
+++ b/gcc/testsuite/gcc.target/arm/simd/vst1_p64_xN_1.c
@@ -0,0 +1,13 @@
+/* { dg-do assemble } */
+/* { dg-require-effective-target arm_crypto_ok } */
+/* { dg-options "-save-temps -O2" } */
+/* { dg-add-options arm_crypto } */
+
+#include "arm_neon.h"
+
+void test_vst1_p64_x2 (poly64_t * ptr, poly64x1x2_t val)
+{
+ vst1_p64_x2 (ptr, val);
+}
+
+/* { dg-final { scan-assembler-times {vst1.64\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+:64\]\n} 1 } } */
\ No newline at end of file
--
2.25.1
^ permalink raw reply [flat|nested] 8+ messages in thread
* [PATCH 2/3] [GCC] arm: vst1_types_x3 ACLE intrinsics
2023-10-06 11:55 [PATCH 0/3] [GCC] arm: vst1_types_xN ACLE intrinsics Ezra.Sitorus
2023-10-06 11:55 ` [PATCH 1/3] [GCC] arm: vst1_types_x2 " Ezra.Sitorus
@ 2023-10-06 11:55 ` Ezra.Sitorus
2023-11-27 14:58 ` Richard Earnshaw
2023-10-06 11:56 ` [PATCH 3/3] [GCC] arm: vst1_types_x4 " Ezra.Sitorus
2 siblings, 1 reply; 8+ messages in thread
From: Ezra.Sitorus @ 2023-10-06 11:55 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 vst1 intrinsic for arm32.
This patch adds the _x3 variants of the vst1 intrinsic.
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
(vst1_u8_x3, vst1_u16_x3, vst1_u32_x3, vst1_u64_x3): New.
(vst1_s8_x3, vst1_s16_x3, vst1_s32_x3, vst1_s64_x3): New.
(vst1_f16_x3, vst1_f32_x3): New.
(vst1_p8_x3, vst1_p16_x3, vst1_p64_x3): New.
(vst1_bf16_x3): New.
* config/arm/arm_neon_builtins.def (vst1_x3): New entries.
* config/arm/neon.md (vst1_x3<mode>): New.
gcc/testsuite/ChangeLog:
* gcc.target/arm/simd/vst1_base_xN_1.c: Add new test.
* gcc.target/arm/simd/vst1_bf16_xN_1.c: Add new test.
* gcc.target/arm/simd/vst1_fp16_xN_1.c: Add new test.
* gcc.target/arm/simd/vst1_p64_xN_1.c: Add new test.
---
gcc/config/arm/arm_neon.h | 114 ++++++++++++++++++
gcc/config/arm/arm_neon_builtins.def | 1 +
gcc/config/arm/neon.md | 10 ++
.../gcc.target/arm/simd/vst1_base_xN_1.c | 63 +++++++++-
.../gcc.target/arm/simd/vst1_bf16_xN_1.c | 7 +-
.../gcc.target/arm/simd/vst1_fp16_xN_1.c | 7 +-
.../gcc.target/arm/simd/vst1_p64_xN_1.c | 7 +-
7 files changed, 202 insertions(+), 7 deletions(-)
diff --git a/gcc/config/arm/arm_neon.h b/gcc/config/arm/arm_neon.h
index 4bd6093281b..b01171e5966 100644
--- a/gcc/config/arm/arm_neon.h
+++ b/gcc/config/arm/arm_neon.h
@@ -11250,6 +11250,14 @@ vst1_p64_x2 (poly64_t * __a, poly64x1x2_t __b)
__builtin_neon_vst1_x2di ((__builtin_neon_di *) __a, __bu.__o);
}
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vst1_p64_x3 (poly64_t * __a, poly64x1x3_t __b)
+{
+ union { poly64x1x3_t __i; __builtin_neon_ei __o; } __bu = { __b };
+ __builtin_neon_vst1_x3di ((__builtin_neon_di *) __a, __bu.__o);
+}
+
#pragma GCC pop_options
__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
@@ -11311,6 +11319,38 @@ vst1_s64_x2 (int64_t * __a, int64x1x2_t __b)
__builtin_neon_vst1_x2di ((__builtin_neon_di *) __a, __bu.__o);
}
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vst1_s8_x3 (int8_t * __a, int8x8x3_t __b)
+{
+ union { int8x8x3_t __i; __builtin_neon_ei __o; } __bu = { __b };
+ __builtin_neon_vst1_x3v8qi ((__builtin_neon_qi *) __a, __bu.__o);
+}
+
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vst1_s16_x3 (int16_t * __a, int16x4x3_t __b)
+{
+ union { int16x4x3_t __i; __builtin_neon_ei __o; } __bu = { __b };
+ __builtin_neon_vst1_x3v4hi ((__builtin_neon_hi *) __a, __bu.__o);
+}
+
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vst1_s32_x3 (int32_t * __a, int32x2x3_t __b)
+{
+ union { int32x2x3_t __i; __builtin_neon_ei __o; } __bu = { __b };
+ __builtin_neon_vst1_x3v2si ((__builtin_neon_si *) __a, __bu.__o);
+}
+
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vst1_s64_x3 (int64_t * __a, int64x1x3_t __b)
+{
+ union { int64x1x3_t __i; __builtin_neon_ei __o; } __bu = { __b };
+ __builtin_neon_vst1_x3di ((__builtin_neon_di *) __a, __bu.__o);
+}
+
#if defined (__ARM_FP16_FORMAT_IEEE) || defined (__ARM_FP16_FORMAT_ALTERNATIVE)
__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
@@ -11345,6 +11385,24 @@ vst1_f32_x2 (float32_t * __a, float32x2x2_t __b)
__builtin_neon_vst1_x2v2sf ((__builtin_neon_sf *) __a, __bu.__o);
}
+#if defined (__ARM_FP16_FORMAT_IEEE) || defined (__ARM_FP16_FORMAT_ALTERNATIVE)
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vst1_f16_x3 (float16_t * __a, float16x4x3_t __b)
+{
+ union { float16x4x3_t __i; __builtin_neon_ei __o; } __bu = { __b };
+ __builtin_neon_vst1_x3v4hf (__a, __bu.__o);
+}
+#endif
+
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vst1_f32_x3 (float32_t * __a, float32x2x3_t __b)
+{
+ union { float32x2x3_t __i; __builtin_neon_ei __o; } __bu = { __b };
+ __builtin_neon_vst1_x3v2sf ((__builtin_neon_sf *) __a, __bu.__o);
+}
+
__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vst1_u8 (uint8_t * __a, uint8x8_t __b)
@@ -11405,6 +11463,38 @@ vst1_u64_x2 (uint64_t * __a, uint64x1x2_t __b)
__builtin_neon_vst1_x2di ((__builtin_neon_di *) __a, __bu.__o);
}
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vst1_u8_x3 (uint8_t * __a, uint8x8x3_t __b)
+{
+ union { uint8x8x3_t __i; __builtin_neon_ei __o; } __bu = { __b };
+ __builtin_neon_vst1_x3v8qi ((__builtin_neon_qi *) __a, __bu.__o);
+}
+
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vst1_u16_x3 (uint16_t * __a, uint16x4x3_t __b)
+{
+ union { uint16x4x3_t __i; __builtin_neon_ei __o; } __bu = { __b };
+ __builtin_neon_vst1_x3v4hi ((__builtin_neon_hi *) __a, __bu.__o);
+}
+
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vst1_u32_x3 (uint32_t * __a, uint32x2x3_t __b)
+{
+ union { uint32x2x3_t __i; __builtin_neon_ei __o; } __bu = { __b };
+ __builtin_neon_vst1_x3v2si ((__builtin_neon_si *) __a, __bu.__o);
+}
+
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vst1_u64_x3 (uint64_t * __a, uint64x1x3_t __b)
+{
+ union { uint64x1x3_t __i; __builtin_neon_ei __o; } __bu = { __b };
+ __builtin_neon_vst1_x3di ((__builtin_neon_di *) __a, __bu.__o);
+}
+
__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vst1_p8 (poly8_t * __a, poly8x8_t __b)
@@ -11435,6 +11525,22 @@ vst1_p16_x2 (poly16_t * __a, poly16x4x2_t __b)
__builtin_neon_vst1_x2v4hi ((__builtin_neon_hi *) __a, __bu.__o);
}
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vst1_p8_x3 (poly8_t * __a, poly8x8x3_t __b)
+{
+ union { poly8x8x3_t __i; __builtin_neon_ei __o; } __bu = { __b };
+ __builtin_neon_vst1_x3v8qi ((__builtin_neon_qi *) __a, __bu.__o);
+}
+
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vst1_p16_x3 (poly16_t * __a, poly16x4x3_t __b)
+{
+ union { poly16x4x3_t __i; __builtin_neon_ei __o; } __bu = { __b };
+ __builtin_neon_vst1_x3v4hi ((__builtin_neon_hi *) __a, __bu.__o);
+}
+
#pragma GCC push_options
#pragma GCC target ("fpu=crypto-neon-fp-armv8")
__extension__ extern __inline void
@@ -20184,6 +20290,14 @@ vst1_bf16_x2 (bfloat16_t * __a, bfloat16x4x2_t __b)
__builtin_neon_vst1_x2v4hf ((__builtin_neon_bf *) __a, __bu.__o);
}
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vst1_bf16_x3 (bfloat16_t * __a, bfloat16x4x3_t __b)
+{
+ union { bfloat16x4x3_t __i; __builtin_neon_ei __o; } __bu = { __b };
+ __builtin_neon_vst1_x3v4hf ((__builtin_neon_bf *) __a, __bu.__o);
+}
+
__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vst1q_bf16 (bfloat16_t * __a, bfloat16x8_t __b)
diff --git a/gcc/config/arm/arm_neon_builtins.def b/gcc/config/arm/arm_neon_builtins.def
index 7aef6f958cd..e33b3429f65 100644
--- a/gcc/config/arm/arm_neon_builtins.def
+++ b/gcc/config/arm/arm_neon_builtins.def
@@ -309,6 +309,7 @@ VAR12 (LOAD1LANE, vld1_lane,
VAR10 (LOAD1, vld1_dup,
v8qi, v4hi, v2si, v2sf, di, v16qi, v8hi, v4si, v4sf, v2di)
VAR7 (STORE1, vst1_x2, v8qi, v4hi, v2si, di, v4hf, v2sf, v4bf)
+VAR7 (STORE1, vst1_x3, v8qi, v4hi, v2si, di, v4hf, v2sf, v4bf)
VAR14 (STORE1, vst1,
v8qi, v4hi, v4hf, v2si, v2sf, di, v16qi, v8hi, v8hf, v4si, v4sf, v2di,
v4bf, v8bf)
diff --git a/gcc/config/arm/neon.md b/gcc/config/arm/neon.md
index 7a10e2cb61e..3d1d2aa7d06 100644
--- a/gcc/config/arm/neon.md
+++ b/gcc/config/arm/neon.md
@@ -5135,6 +5135,16 @@ if (BYTES_BIG_ENDIAN)
[(set_attr "type" "neon_store1_2reg<q>")]
)
+(define_insn "neon_vst1_x3<mode>"
+ [(set (match_operand:EI 0 "neon_struct_operand" "=Um")
+ (unspec:EI [(match_operand:EI 1 "s_register_operand" "w")
+ (unspec:VDQX [(const_int 0)] UNSPEC_VSTRUCTDUMMY)]
+ UNSPEC_VST1))]
+ "TARGET_NEON"
+ "vst1.<V_sz_elem>\t%h1, %A0"
+ [(set_attr "type" "neon_store1_3reg<q>")]
+)
+
(define_insn "neon_vst1<mode>"
[(set (match_operand:VDQX 0 "neon_struct_operand" "=Um")
(unspec:VDQX [(match_operand:VDQX 1 "s_register_operand" "w")]
diff --git a/gcc/testsuite/gcc.target/arm/simd/vst1_base_xN_1.c b/gcc/testsuite/gcc.target/arm/simd/vst1_base_xN_1.c
index 575897fa422..5f820a6a496 100644
--- a/gcc/testsuite/gcc.target/arm/simd/vst1_base_xN_1.c
+++ b/gcc/testsuite/gcc.target/arm/simd/vst1_base_xN_1.c
@@ -60,8 +60,63 @@ void test_vst1_p16_x2 (poly16_t * ptr, poly16x4x2_t val)
vst1_p16_x2 (ptr, val);
}
+void test_vst1_u8_x3 (uint8_t * ptr, uint8x8x3_t val)
+{
+ vst1_u8_x3 (ptr, val);
+}
+
+void test_vst1_u16_x3 (uint16_t * ptr, uint16x4x3_t val)
+{
+ vst1_u16_x3 (ptr, val);
+}
+
+void test_vst1_u32_x3 (uint32_t * ptr, uint32x2x3_t val)
+{
+ vst1_u32_x3 (ptr, val);
+}
+
+void test_vst1_u64_x3 (uint64_t * ptr, uint64x1x3_t val)
+{
+ vst1_u64_x3 (ptr, val);
+}
+
+void test_vst1_s8_x3 (int8_t * ptr, int8x8x3_t val)
+{
+ vst1_s8_x3 (ptr, val);
+}
+
+void test_vst1_s16_x3 (int16_t * ptr, int16x4x3_t val)
+{
+ vst1_s16_x3 (ptr, val);
+}
+
+void test_vst1_s32_x3 (int32_t * ptr, int32x2x3_t val)
+{
+ vst1_s32_x3 (ptr, val);
+}
+
+void test_vst1_s64_x3 (int64_t * ptr, int64x1x3_t val)
+{
+ vst1_s64_x3 (ptr, val);
+}
+
+void test_vst1_f32_x3 (float32_t * ptr, float32x2x3_t val)
+{
+ vst1_f32_x3 (ptr, val);
+}
+
+void test_vst1_p8_x3 (poly8_t * ptr, poly8x8x3_t val)
+{
+ vst1_p8_x3 (ptr, val);
+}
+
+void test_vst1_p16_x3 (poly16_t * ptr, poly16x4x3_t val)
+{
+ vst1_p16_x3 (ptr, val);
+}
+
-/* { dg-final { scan-assembler-times {vst1.8\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+\]\n} 3 } } */
-/* { dg-final { scan-assembler-times {vst1.16\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+\]\n} 3 } } */
-/* { dg-final { scan-assembler-times {vst1.32\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+\]\n} 3 } } */
-/* { dg-final { scan-assembler-times {vst1.64\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+:64\]\n} 2 } } */
+/* { dg-final { scan-assembler-times {vst1.8\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+\]\n} 6 } } */
+/* { dg-final { scan-assembler-times {vst1.16\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+\]\n} 6 } } */
+/* { dg-final { scan-assembler-times {vst1.32\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+\]\n} 6 } } */
+/* { dg-final { scan-assembler-times {vst1.64\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+:64\]\n} 4 } } */
diff --git a/gcc/testsuite/gcc.target/arm/simd/vst1_bf16_xN_1.c b/gcc/testsuite/gcc.target/arm/simd/vst1_bf16_xN_1.c
index 213fd20ee65..a3a00ead468 100644
--- a/gcc/testsuite/gcc.target/arm/simd/vst1_bf16_xN_1.c
+++ b/gcc/testsuite/gcc.target/arm/simd/vst1_bf16_xN_1.c
@@ -10,4 +10,9 @@ void test_vst1_bf16_x2 (bfloat16_t * ptr, bfloat16x4x2_t val)
vst1_bf16_x2 (ptr, val);
}
-/* { dg-final { scan-assembler-times {vst1.16\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+\]\n} 1 } } */
+void test_vst1_bf16_x3 (bfloat16_t * ptr, bfloat16x4x3_t val)
+{
+ vst1_bf16_x3 (ptr, val);
+}
+
+/* { dg-final { scan-assembler-times {vst1.16\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+\]\n} 2 } } */
diff --git a/gcc/testsuite/gcc.target/arm/simd/vst1_fp16_xN_1.c b/gcc/testsuite/gcc.target/arm/simd/vst1_fp16_xN_1.c
index 523aec92db2..0a6863e24c6 100644
--- a/gcc/testsuite/gcc.target/arm/simd/vst1_fp16_xN_1.c
+++ b/gcc/testsuite/gcc.target/arm/simd/vst1_fp16_xN_1.c
@@ -10,4 +10,9 @@ void test_vst1_f16_x2 (float16_t * ptr, float16x4x2_t val)
vst1_f16_x2 (ptr, val);
}
-/* { dg-final { scan-assembler-times {vst1.16\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+\]\n} 1 } } */
+void test_vst1_f16_x3 (float16_t * ptr, float16x4x3_t val)
+{
+ vst1_f16_x3 (ptr, val);
+}
+
+/* { dg-final { scan-assembler-times {vst1.16\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+\]\n} 2 } } */
diff --git a/gcc/testsuite/gcc.target/arm/simd/vst1_p64_xN_1.c b/gcc/testsuite/gcc.target/arm/simd/vst1_p64_xN_1.c
index f590ebd7b94..5dbd6049bc9 100644
--- a/gcc/testsuite/gcc.target/arm/simd/vst1_p64_xN_1.c
+++ b/gcc/testsuite/gcc.target/arm/simd/vst1_p64_xN_1.c
@@ -10,4 +10,9 @@ void test_vst1_p64_x2 (poly64_t * ptr, poly64x1x2_t val)
vst1_p64_x2 (ptr, val);
}
-/* { dg-final { scan-assembler-times {vst1.64\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+:64\]\n} 1 } } */
\ No newline at end of file
+void test_vst1_p64_x3 (poly64_t * ptr, poly64x1x3_t val)
+{
+ vst1_p64_x3 (ptr, val);
+}
+
+/* { dg-final { scan-assembler-times {vst1.64\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+:64\]\n} 2 } } */
\ No newline at end of file
--
2.25.1
^ permalink raw reply [flat|nested] 8+ messages in thread
* [PATCH 3/3] [GCC] arm: vst1_types_x4 ACLE intrinsics
2023-10-06 11:55 [PATCH 0/3] [GCC] arm: vst1_types_xN ACLE intrinsics Ezra.Sitorus
2023-10-06 11:55 ` [PATCH 1/3] [GCC] arm: vst1_types_x2 " Ezra.Sitorus
2023-10-06 11:55 ` [PATCH 2/3] [GCC] arm: vst1_types_x3 " Ezra.Sitorus
@ 2023-10-06 11:56 ` Ezra.Sitorus
2023-11-27 14:59 ` Richard Earnshaw
2 siblings, 1 reply; 8+ messages in thread
From: Ezra.Sitorus @ 2023-10-06 11:56 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 vst1 intrinsic for arm32.
This patch adds the _x4 variants of the vst1 intrinsic.
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
(vst1_u8_x4, vst1_u16_x4, vst1_u32_x4, vst1_u64_x4): New.
(vst1_s8_x4, vst1_s16_x4, vst1_s32_x4, vst1_s64_x4): New.
(vst1_f16_x4, vst1_f32_x4): New.
(vst1_p8_x4, vst1_p16_x4, vst1_p64_x4): New.
(vst1_bf16_x4): New.
* config/arm/arm_neon_builtins.def (vst1_x4): New entries.
* config/arm/neon.md (vst1_x4<mode>): New.
gcc/testsuite/ChangeLog:
* gcc.target/arm/simd/vst1_base_xN_1.c: Add new test.
* gcc.target/arm/simd/vst1_bf16_xN_1.c: Add new test.
* gcc.target/arm/simd/vst1_fp16_xN_1.c: Add new test.
* gcc.target/arm/simd/vst1_p64_xN_1.c: Add new test.
---
gcc/config/arm/arm_neon.h | 114 ++++++++++++++++++
gcc/config/arm/arm_neon_builtins.def | 1 +
gcc/config/arm/neon.md | 10 ++
.../gcc.target/arm/simd/vst1_base_xN_1.c | 62 +++++++++-
.../gcc.target/arm/simd/vst1_bf16_xN_1.c | 6 +-
.../gcc.target/arm/simd/vst1_fp16_xN_1.c | 7 +-
.../gcc.target/arm/simd/vst1_p64_xN_1.c | 7 +-
7 files changed, 200 insertions(+), 7 deletions(-)
diff --git a/gcc/config/arm/arm_neon.h b/gcc/config/arm/arm_neon.h
index b01171e5966..41e645d8352 100644
--- a/gcc/config/arm/arm_neon.h
+++ b/gcc/config/arm/arm_neon.h
@@ -11258,6 +11258,14 @@ vst1_p64_x3 (poly64_t * __a, poly64x1x3_t __b)
__builtin_neon_vst1_x3di ((__builtin_neon_di *) __a, __bu.__o);
}
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vst1_p64_x4 (poly64_t * __a, poly64x1x4_t __b)
+{
+ union { poly64x1x4_t __i; __builtin_neon_oi __o; } __bu = { __b };
+ __builtin_neon_vst1_x3di ((__builtin_neon_di *) __a, __bu.__o);
+}
+
#pragma GCC pop_options
__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
@@ -11351,6 +11359,38 @@ vst1_s64_x3 (int64_t * __a, int64x1x3_t __b)
__builtin_neon_vst1_x3di ((__builtin_neon_di *) __a, __bu.__o);
}
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vst1_s8_x4 (int8_t * __a, int8x8x4_t __b)
+{
+ union { int8x8x4_t __i; __builtin_neon_oi __o; } __bu = { __b };
+ __builtin_neon_vst1_x4v8qi ((__builtin_neon_qi *) __a, __bu.__o);
+}
+
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vst1_s16_x4 (int16_t * __a, int16x4x4_t __b)
+{
+ union { int16x4x4_t __i; __builtin_neon_oi __o; } __bu = { __b };
+ __builtin_neon_vst1_x4v4hi ((__builtin_neon_hi *) __a, __bu.__o);
+}
+
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vst1_s32_x4 (int32_t * __a, int32x2x4_t __b)
+{
+ union { int32x2x4_t __i; __builtin_neon_oi __o; } __bu = { __b };
+ __builtin_neon_vst1_x4v2si ((__builtin_neon_si *) __a, __bu.__o);
+}
+
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vst1_s64_x4 (int64_t * __a, int64x1x4_t __b)
+{
+ union { int64x1x4_t __i; __builtin_neon_oi __o; } __bu = { __b };
+ __builtin_neon_vst1_x4di ((__builtin_neon_di *) __a, __bu.__o);
+}
+
#if defined (__ARM_FP16_FORMAT_IEEE) || defined (__ARM_FP16_FORMAT_ALTERNATIVE)
__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
@@ -11403,6 +11443,24 @@ vst1_f32_x3 (float32_t * __a, float32x2x3_t __b)
__builtin_neon_vst1_x3v2sf ((__builtin_neon_sf *) __a, __bu.__o);
}
+#if defined (__ARM_FP16_FORMAT_IEEE) || defined (__ARM_FP16_FORMAT_ALTERNATIVE)
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vst1_f16_x4 (float16_t * __a, float16x4x4_t __b)
+{
+ union { float16x4x4_t __i; __builtin_neon_oi __o; } __bu = { __b };
+ __builtin_neon_vst1_x4v4hf (__a, __bu.__o);
+}
+#endif
+
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vst1_f32_x4 (float32_t * __a, float32x2x4_t __b)
+{
+ union { float32x2x4_t __i; __builtin_neon_oi __o; } __bu = { __b };
+ __builtin_neon_vst1_x4v2sf ((__builtin_neon_sf *) __a, __bu.__o);
+}
+
__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vst1_u8 (uint8_t * __a, uint8x8_t __b)
@@ -11495,6 +11553,38 @@ vst1_u64_x3 (uint64_t * __a, uint64x1x3_t __b)
__builtin_neon_vst1_x3di ((__builtin_neon_di *) __a, __bu.__o);
}
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vst1_u8_x4 (uint8_t * __a, uint8x8x4_t __b)
+{
+ union { uint8x8x4_t __i; __builtin_neon_oi __o; } __bu = { __b };
+ __builtin_neon_vst1_x4v8qi ((__builtin_neon_qi *) __a, __bu.__o);
+}
+
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vst1_u16_x4 (uint16_t * __a, uint16x4x4_t __b)
+{
+ union { uint16x4x4_t __i; __builtin_neon_oi __o; } __bu = { __b };
+ __builtin_neon_vst1_x4v4hi ((__builtin_neon_hi *) __a, __bu.__o);
+}
+
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vst1_u32_x4 (uint32_t * __a, uint32x2x4_t __b)
+{
+ union { uint32x2x4_t __i; __builtin_neon_oi __o; } __bu = { __b };
+ __builtin_neon_vst1_x4v2si ((__builtin_neon_si *) __a, __bu.__o);
+}
+
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vst1_u64_x4 (uint64_t * __a, uint64x1x4_t __b)
+{
+ union { uint64x1x4_t __i; __builtin_neon_oi __o; } __bu = { __b };
+ __builtin_neon_vst1_x4di ((__builtin_neon_di *) __a, __bu.__o);
+}
+
__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vst1_p8 (poly8_t * __a, poly8x8_t __b)
@@ -11541,6 +11631,22 @@ vst1_p16_x3 (poly16_t * __a, poly16x4x3_t __b)
__builtin_neon_vst1_x3v4hi ((__builtin_neon_hi *) __a, __bu.__o);
}
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vst1_p8_x4 (poly8_t * __a, poly8x8x4_t __b)
+{
+ union { poly8x8x4_t __i; __builtin_neon_oi __o; } __bu = { __b };
+ __builtin_neon_vst1_x4v8qi ((__builtin_neon_qi *) __a, __bu.__o);
+}
+
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vst1_p16_x4 (poly16_t * __a, poly16x4x4_t __b)
+{
+ union { poly16x4x4_t __i; __builtin_neon_oi __o; } __bu = { __b };
+ __builtin_neon_vst1_x4v4hi ((__builtin_neon_hi *) __a, __bu.__o);
+}
+
#pragma GCC push_options
#pragma GCC target ("fpu=crypto-neon-fp-armv8")
__extension__ extern __inline void
@@ -20298,6 +20404,14 @@ vst1_bf16_x3 (bfloat16_t * __a, bfloat16x4x3_t __b)
__builtin_neon_vst1_x3v4hf ((__builtin_neon_bf *) __a, __bu.__o);
}
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vst1_bf16_x4 (bfloat16_t * __a, bfloat16x4x4_t __b)
+{
+ union { bfloat16x4x4_t __i; __builtin_neon_oi __o; } __bu = { __b };
+ __builtin_neon_vst1_x4v4hf ((__builtin_neon_bf *) __a, __bu.__o);
+}
+
__extension__ extern __inline void
__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
vst1q_bf16 (bfloat16_t * __a, bfloat16x8_t __b)
diff --git a/gcc/config/arm/arm_neon_builtins.def b/gcc/config/arm/arm_neon_builtins.def
index e33b3429f65..95300cb0fe4 100644
--- a/gcc/config/arm/arm_neon_builtins.def
+++ b/gcc/config/arm/arm_neon_builtins.def
@@ -310,6 +310,7 @@ VAR10 (LOAD1, vld1_dup,
v8qi, v4hi, v2si, v2sf, di, v16qi, v8hi, v4si, v4sf, v2di)
VAR7 (STORE1, vst1_x2, v8qi, v4hi, v2si, di, v4hf, v2sf, v4bf)
VAR7 (STORE1, vst1_x3, v8qi, v4hi, v2si, di, v4hf, v2sf, v4bf)
+VAR7 (STORE1, vst1_x4, v8qi, v4hi, v2si, di, v4hf, v2sf, v4bf)
VAR14 (STORE1, vst1,
v8qi, v4hi, v4hf, v2si, v2sf, di, v16qi, v8hi, v8hf, v4si, v4sf, v2di,
v4bf, v8bf)
diff --git a/gcc/config/arm/neon.md b/gcc/config/arm/neon.md
index 3d1d2aa7d06..f5d583129fa 100644
--- a/gcc/config/arm/neon.md
+++ b/gcc/config/arm/neon.md
@@ -5145,6 +5145,16 @@ if (BYTES_BIG_ENDIAN)
[(set_attr "type" "neon_store1_3reg<q>")]
)
+(define_insn "neon_vst1_x4<mode>"
+ [(set (match_operand:OI 0 "neon_struct_operand" "=Um")
+ (unspec:OI [(match_operand:OI 1 "s_register_operand" "w")
+ (unspec:VDQX [(const_int 0)] UNSPEC_VSTRUCTDUMMY)]
+ UNSPEC_VST1))]
+ "TARGET_NEON"
+ "vst1.<V_sz_elem>\t%h1, %A0"
+ [(set_attr "type" "neon_store1_4reg<q>")]
+)
+
(define_insn "neon_vst1<mode>"
[(set (match_operand:VDQX 0 "neon_struct_operand" "=Um")
(unspec:VDQX [(match_operand:VDQX 1 "s_register_operand" "w")]
diff --git a/gcc/testsuite/gcc.target/arm/simd/vst1_base_xN_1.c b/gcc/testsuite/gcc.target/arm/simd/vst1_base_xN_1.c
index 5f820a6a496..04ca6583552 100644
--- a/gcc/testsuite/gcc.target/arm/simd/vst1_base_xN_1.c
+++ b/gcc/testsuite/gcc.target/arm/simd/vst1_base_xN_1.c
@@ -115,8 +115,62 @@ void test_vst1_p16_x3 (poly16_t * ptr, poly16x4x3_t val)
vst1_p16_x3 (ptr, val);
}
+void test_vst1_u8_x4 (uint8_t * ptr, uint8x8x4_t val)
+{
+ vst1_u8_x4 (ptr, val);
+}
+
+void test_vst1_u16_x4 (uint16_t * ptr, uint16x4x4_t val)
+{
+ vst1_u16_x4 (ptr, val);
+}
+
+void test_vst1_u32_x4 (uint32_t * ptr, uint32x2x4_t val)
+{
+ vst1_u32_x4 (ptr, val);
+}
+
+void test_vst1_u64_x4 (uint64_t * ptr, uint64x1x4_t val)
+{
+ vst1_u64_x4 (ptr, val);
+}
+
+void test_vst1_s8_x4 (int8_t * ptr, int8x8x4_t val)
+{
+ vst1_s8_x4 (ptr, val);
+}
+
+void test_vst1_s16_x4 (int16_t * ptr, int16x4x4_t val)
+{
+ vst1_s16_x4 (ptr, val);
+}
+
+void test_vst1_s32_x4 (int32_t * ptr, int32x2x4_t val)
+{
+ vst1_s32_x4 (ptr, val);
+}
+
+void test_vst1_s64_x4 (int64_t * ptr, int64x1x4_t val)
+{
+ vst1_s64_x4 (ptr, val);
+}
+
+void test_vst1_f32_x4 (float32_t * ptr, float32x2x4_t val)
+{
+ vst1_f32_x4 (ptr, val);
+}
+
+void test_vst1_p8_x4 (poly8_t * ptr, poly8x8x4_t val)
+{
+ vst1_p8_x4 (ptr, val);
+}
+
+void test_vst1_p16_x4 (poly16_t * ptr, poly16x4x4_t val)
+{
+ vst1_p16_x4 (ptr, val);
+}
-/* { dg-final { scan-assembler-times {vst1.8\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+\]\n} 6 } } */
-/* { dg-final { scan-assembler-times {vst1.16\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+\]\n} 6 } } */
-/* { dg-final { scan-assembler-times {vst1.32\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+\]\n} 6 } } */
-/* { dg-final { scan-assembler-times {vst1.64\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+:64\]\n} 4 } } */
+/* { dg-final { scan-assembler-times {vst1.8\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+\]\n} 9 } } */
+/* { dg-final { scan-assembler-times {vst1.16\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+\]\n} 9 } } */
+/* { dg-final { scan-assembler-times {vst1.32\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+\]\n} 9 } } */
+/* { dg-final { scan-assembler-times {vst1.64\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+:64\]\n} 6 } } */
diff --git a/gcc/testsuite/gcc.target/arm/simd/vst1_bf16_xN_1.c b/gcc/testsuite/gcc.target/arm/simd/vst1_bf16_xN_1.c
index a3a00ead468..d919c7d060d 100644
--- a/gcc/testsuite/gcc.target/arm/simd/vst1_bf16_xN_1.c
+++ b/gcc/testsuite/gcc.target/arm/simd/vst1_bf16_xN_1.c
@@ -15,4 +15,8 @@ void test_vst1_bf16_x3 (bfloat16_t * ptr, bfloat16x4x3_t val)
vst1_bf16_x3 (ptr, val);
}
-/* { dg-final { scan-assembler-times {vst1.16\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+\]\n} 2 } } */
+void test_vst1_bf16_x4 (bfloat16_t * ptr, bfloat16x4x4_t val)
+{
+ vst1_bf16_x4 (ptr, val);
+}
+/* { dg-final { scan-assembler-times {vst1.16\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+\]\n} 3 } } */
diff --git a/gcc/testsuite/gcc.target/arm/simd/vst1_fp16_xN_1.c b/gcc/testsuite/gcc.target/arm/simd/vst1_fp16_xN_1.c
index 0a6863e24c6..3d1d1eb7ad1 100644
--- a/gcc/testsuite/gcc.target/arm/simd/vst1_fp16_xN_1.c
+++ b/gcc/testsuite/gcc.target/arm/simd/vst1_fp16_xN_1.c
@@ -15,4 +15,9 @@ void test_vst1_f16_x3 (float16_t * ptr, float16x4x3_t val)
vst1_f16_x3 (ptr, val);
}
-/* { dg-final { scan-assembler-times {vst1.16\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+\]\n} 2 } } */
+void test_vst1_f16_x4 (float16_t * ptr, float16x4x4_t val)
+{
+ vst1_f16_x4 (ptr, val);
+}
+
+/* { dg-final { scan-assembler-times {vst1.16\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+\]\n} 3 } } */
diff --git a/gcc/testsuite/gcc.target/arm/simd/vst1_p64_xN_1.c b/gcc/testsuite/gcc.target/arm/simd/vst1_p64_xN_1.c
index 5dbd6049bc9..62912143481 100644
--- a/gcc/testsuite/gcc.target/arm/simd/vst1_p64_xN_1.c
+++ b/gcc/testsuite/gcc.target/arm/simd/vst1_p64_xN_1.c
@@ -15,4 +15,9 @@ void test_vst1_p64_x3 (poly64_t * ptr, poly64x1x3_t val)
vst1_p64_x3 (ptr, val);
}
-/* { dg-final { scan-assembler-times {vst1.64\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+:64\]\n} 2 } } */
\ No newline at end of file
+void test_vst1_p64_x4 (poly64_t * ptr, poly64x1x4_t val)
+{
+ vst1_p64_x4 (ptr, val);
+}
+
+/* { dg-final { scan-assembler-times {vst1.64\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+:64\]\n} 3 } } */
\ No newline at end of file
--
2.25.1
^ permalink raw reply [flat|nested] 8+ messages in thread
* Re: [PATCH 1/3] [GCC] arm: vst1_types_x2 ACLE intrinsics
2023-10-06 11:55 ` [PATCH 1/3] [GCC] arm: vst1_types_x2 " Ezra.Sitorus
@ 2023-11-27 14:56 ` Richard Earnshaw
2023-11-27 14:57 ` Richard Earnshaw
0 siblings, 1 reply; 8+ messages in thread
From: Richard Earnshaw @ 2023-11-27 14:56 UTC (permalink / raw)
To: Ezra.Sitorus, gcc-patches; +Cc: richard.earnshaw, kyrylo.tkachov
On 06/10/2023 12:55, Ezra.Sitorus@arm.com wrote:
> From: Ezra Sitorus <ezra.sitorus@arm.com>
>
> This patch is part of a series of patches implementing the _xN variants of the vst1 intrinsic for arm32.
We avoid the terms arm32 and arm64 in the gnu toolchain because of the
potential for conflict with wild-card regexps for old CPUs (arm3* and
arm6*). Please just use 'arm' in this case.
> This patch adds the _x2 variants of the vst1 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/
>
Please reformat this text section to be no more than 70 columns, so that
it doesn't produce awkward line wraps in the commit message.
> gcc/ChangeLog:
> * config/arm/arm_neon.h
> (vst1_u8_x2, vst1_u16_x2, vst1_u32_x2, vst1_u64_x32): New.
This line appears to be a duplicate of the one below, except for the
bogus ..._x32 at the end. I think just drop it.
> (vst1_s8_x2, vst1_s16_x2, vst1_s32_x2, vst1_s64_x2): New.
> (vst1_f16_x2, vst1_f32_x2): New.
> (vst1_p8_x2, vst1_p16_x2, vst1_p64_x2): New.
> (vst1_bf16_x2): New.
> * config/arm/arm_neon_builtins.def (vst1_x2): New entries.
> * config/arm/neon.md (vst1_x2<mode>): New.
>
> gcc/testsuite/ChangeLog:
> * gcc.target/arm/simd/vst1_base_xN_1.c: Add new tests.
> * gcc.target/arm/simd/vst1_bf16_xN_1.c: Add new tests.
> * gcc.target/arm/simd/vst1_fp16_xN_1.c: Add new tests.
> * gcc.target/arm/simd/vst1_p64_xN_1.c: Add new tests.
OK with the above changes.
R.
> ---
> gcc/config/arm/arm_neon.h | 114 ++++++++++++++++++
> gcc/config/arm/arm_neon_builtins.def | 1 +
> gcc/config/arm/neon.md | 10 ++
> .../gcc.target/arm/simd/vst1_base_xN_1.c | 67 ++++++++++
> .../gcc.target/arm/simd/vst1_bf16_xN_1.c | 13 ++
> .../gcc.target/arm/simd/vst1_fp16_xN_1.c | 13 ++
> .../gcc.target/arm/simd/vst1_p64_xN_1.c | 13 ++
> 7 files changed, 231 insertions(+)
> create mode 100644 gcc/testsuite/gcc.target/arm/simd/vst1_base_xN_1.c
> create mode 100644 gcc/testsuite/gcc.target/arm/simd/vst1_bf16_xN_1.c
> create mode 100644 gcc/testsuite/gcc.target/arm/simd/vst1_fp16_xN_1.c
> create mode 100644 gcc/testsuite/gcc.target/arm/simd/vst1_p64_xN_1.c
>
> diff --git a/gcc/config/arm/arm_neon.h b/gcc/config/arm/arm_neon.h
> index c03be9912f8..4bd6093281b 100644
> --- a/gcc/config/arm/arm_neon.h
> +++ b/gcc/config/arm/arm_neon.h
> @@ -11242,6 +11242,14 @@ vst1_p64 (poly64_t * __a, poly64x1_t __b)
> __builtin_neon_vst1di ((__builtin_neon_di *) __a, __b);
> }
>
> +__extension__ extern __inline void
> +__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> +vst1_p64_x2 (poly64_t * __a, poly64x1x2_t __b)
> +{
> + union { poly64x1x2_t __i; __builtin_neon_ti __o; } __bu = { __b };
> + __builtin_neon_vst1_x2di ((__builtin_neon_di *) __a, __bu.__o);
> +}
> +
> #pragma GCC pop_options
> __extension__ extern __inline void
> __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> @@ -11271,6 +11279,38 @@ vst1_s64 (int64_t * __a, int64x1_t __b)
> __builtin_neon_vst1di ((__builtin_neon_di *) __a, __b);
> }
>
> +__extension__ extern __inline void
> +__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> +vst1_s8_x2 (int8_t * __a, int8x8x2_t __b)
> +{
> + union { int8x8x2_t __i; __builtin_neon_ti __o; } __bu = { __b };
> + __builtin_neon_vst1_x2v8qi ((__builtin_neon_qi *) __a, __bu.__o);
> +}
> +
> +__extension__ extern __inline void
> +__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> +vst1_s16_x2 (int16_t * __a, int16x4x2_t __b)
> +{
> + union { int16x4x2_t __i; __builtin_neon_ti __o; } __bu = { __b };
> + __builtin_neon_vst1_x2v4hi ((__builtin_neon_hi *) __a, __bu.__o);
> +}
> +
> +__extension__ extern __inline void
> +__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> +vst1_s32_x2 (int32_t * __a, int32x2x2_t __b)
> +{
> + union { int32x2x2_t __i; __builtin_neon_ti __o; } __bu = { __b };
> + __builtin_neon_vst1_x2v2si ((__builtin_neon_si *) __a, __bu.__o);
> +}
> +
> +__extension__ extern __inline void
> +__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> +vst1_s64_x2 (int64_t * __a, int64x1x2_t __b)
> +{
> + union { int64x1x2_t __i; __builtin_neon_ti __o; } __bu = { __b };
> + __builtin_neon_vst1_x2di ((__builtin_neon_di *) __a, __bu.__o);
> +}
> +
> #if defined (__ARM_FP16_FORMAT_IEEE) || defined (__ARM_FP16_FORMAT_ALTERNATIVE)
> __extension__ extern __inline void
> __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> @@ -11287,6 +11327,24 @@ vst1_f32 (float32_t * __a, float32x2_t __b)
> __builtin_neon_vst1v2sf ((__builtin_neon_sf *) __a, __b);
> }
>
> +#if defined (__ARM_FP16_FORMAT_IEEE) || defined (__ARM_FP16_FORMAT_ALTERNATIVE)
> +__extension__ extern __inline void
> +__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> +vst1_f16_x2 (float16_t * __a, float16x4x2_t __b)
> +{
> + union { float16x4x2_t __i; __builtin_neon_ti __o; } __bu = { __b };
> + __builtin_neon_vst1_x2v4hf (__a, __bu.__o);
> +}
> +#endif
> +
> +__extension__ extern __inline void
> +__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> +vst1_f32_x2 (float32_t * __a, float32x2x2_t __b)
> +{
> + union { float32x2x2_t __i; __builtin_neon_ti __o; } __bu = { __b };
> + __builtin_neon_vst1_x2v2sf ((__builtin_neon_sf *) __a, __bu.__o);
> +}
> +
> __extension__ extern __inline void
> __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> vst1_u8 (uint8_t * __a, uint8x8_t __b)
> @@ -11315,6 +11373,38 @@ vst1_u64 (uint64_t * __a, uint64x1_t __b)
> __builtin_neon_vst1di ((__builtin_neon_di *) __a, (int64x1_t) __b);
> }
>
> +__extension__ extern __inline void
> +__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> +vst1_u8_x2 (uint8_t * __a, uint8x8x2_t __b)
> +{
> + union { uint8x8x2_t __i; __builtin_neon_ti __o; } __bu = { __b };
> + __builtin_neon_vst1_x2v8qi ((__builtin_neon_qi *) __a, __bu.__o);
> +}
> +
> +__extension__ extern __inline void
> +__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> +vst1_u16_x2 (uint16_t * __a, uint16x4x2_t __b)
> +{
> + union { uint16x4x2_t __i; __builtin_neon_ti __o; } __bu = { __b };
> + __builtin_neon_vst1_x2v4hi ((__builtin_neon_hi *) __a, __bu.__o);
> +}
> +
> +__extension__ extern __inline void
> +__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> +vst1_u32_x2 (uint32_t * __a, uint32x2x2_t __b)
> +{
> + union { uint32x2x2_t __i; __builtin_neon_ti __o; } __bu = { __b };
> + __builtin_neon_vst1_x2v2si ((__builtin_neon_si *) __a, __bu.__o);
> +}
> +
> +__extension__ extern __inline void
> +__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> +vst1_u64_x2 (uint64_t * __a, uint64x1x2_t __b)
> +{
> + union { uint64x1x2_t __i; __builtin_neon_ti __o; } __bu = { __b };
> + __builtin_neon_vst1_x2di ((__builtin_neon_di *) __a, __bu.__o);
> +}
> +
> __extension__ extern __inline void
> __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> vst1_p8 (poly8_t * __a, poly8x8_t __b)
> @@ -11329,6 +11419,22 @@ vst1_p16 (poly16_t * __a, poly16x4_t __b)
> __builtin_neon_vst1v4hi ((__builtin_neon_hi *) __a, (int16x4_t) __b);
> }
>
> +__extension__ extern __inline void
> +__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> +vst1_p8_x2 (poly8_t * __a, poly8x8x2_t __b)
> +{
> + union { poly8x8x2_t __i; __builtin_neon_ti __o; } __bu = { __b };
> + __builtin_neon_vst1_x2v8qi ((__builtin_neon_qi *) __a, __bu.__o);
> +}
> +
> +__extension__ extern __inline void
> +__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> +vst1_p16_x2 (poly16_t * __a, poly16x4x2_t __b)
> +{
> + union { poly16x4x2_t __i; __builtin_neon_ti __o; } __bu = { __b };
> + __builtin_neon_vst1_x2v4hi ((__builtin_neon_hi *) __a, __bu.__o);
> +}
> +
> #pragma GCC push_options
> #pragma GCC target ("fpu=crypto-neon-fp-armv8")
> __extension__ extern __inline void
> @@ -20070,6 +20176,14 @@ vst1_bf16 (bfloat16_t * __a, bfloat16x4_t __b)
> __builtin_neon_vst1v4bf (__a, __b);
> }
>
> +__extension__ extern __inline void
> +__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> +vst1_bf16_x2 (bfloat16_t * __a, bfloat16x4x2_t __b)
> +{
> + union { bfloat16x4x2_t __i; __builtin_neon_ti __o; } __bu = { __b };
> + __builtin_neon_vst1_x2v4hf ((__builtin_neon_bf *) __a, __bu.__o);
> +}
> +
> __extension__ extern __inline void
> __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> vst1q_bf16 (bfloat16_t * __a, bfloat16x8_t __b)
> diff --git a/gcc/config/arm/arm_neon_builtins.def b/gcc/config/arm/arm_neon_builtins.def
> index f4001b298c4..7aef6f958cd 100644
> --- a/gcc/config/arm/arm_neon_builtins.def
> +++ b/gcc/config/arm/arm_neon_builtins.def
> @@ -308,6 +308,7 @@ VAR12 (LOAD1LANE, vld1_lane,
> v8qi, v4hi, v2si, v2sf, di, v16qi, v8hi, v4si, v4sf, v2di, v4bf, v8bf)
> VAR10 (LOAD1, vld1_dup,
> v8qi, v4hi, v2si, v2sf, di, v16qi, v8hi, v4si, v4sf, v2di)
> +VAR7 (STORE1, vst1_x2, v8qi, v4hi, v2si, di, v4hf, v2sf, v4bf)
> VAR14 (STORE1, vst1,
> v8qi, v4hi, v4hf, v2si, v2sf, di, v16qi, v8hi, v8hf, v4si, v4sf, v2di,
> v4bf, v8bf)
> diff --git a/gcc/config/arm/neon.md b/gcc/config/arm/neon.md
> index 62decab37a2..7a10e2cb61e 100644
> --- a/gcc/config/arm/neon.md
> +++ b/gcc/config/arm/neon.md
> @@ -5125,6 +5125,16 @@ if (BYTES_BIG_ENDIAN)
> UNSPEC_VST1))]
> "TARGET_NEON")
>
> +(define_insn "neon_vst1_x2<mode>"
> + [(set (match_operand:TI 0 "neon_struct_operand" "=Um")
> + (unspec:TI [(match_operand:TI 1 "s_register_operand" "w")
> + (unspec:VDQX [(const_int 0)] UNSPEC_VSTRUCTDUMMY)]
> + UNSPEC_VST1))]
> + "TARGET_NEON"
> + "vst1.<V_sz_elem>\t%h1, %A0"
> + [(set_attr "type" "neon_store1_2reg<q>")]
> +)
> +
> (define_insn "neon_vst1<mode>"
> [(set (match_operand:VDQX 0 "neon_struct_operand" "=Um")
> (unspec:VDQX [(match_operand:VDQX 1 "s_register_operand" "w")]
> diff --git a/gcc/testsuite/gcc.target/arm/simd/vst1_base_xN_1.c b/gcc/testsuite/gcc.target/arm/simd/vst1_base_xN_1.c
> new file mode 100644
> index 00000000000..575897fa422
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/arm/simd/vst1_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"
> +
> +void test_vst1_u8_x2 (uint8_t * ptr, uint8x8x2_t val)
> +{
> + vst1_u8_x2 (ptr, val);
> +}
> +
> +void test_vst1_u16_x2 (uint16_t * ptr, uint16x4x2_t val)
> +{
> + vst1_u16_x2 (ptr, val);
> +}
> +
> +void test_vst1_u32_x2 (uint32_t * ptr, uint32x2x2_t val)
> +{
> + vst1_u32_x2 (ptr, val);
> +}
> +
> +void test_vst1_u64_x2 (uint64_t * ptr, uint64x1x2_t val)
> +{
> + vst1_u64_x2 (ptr, val);
> +}
> +
> +void test_vst1_s8_x2 (int8_t * ptr, int8x8x2_t val)
> +{
> + vst1_s8_x2 (ptr, val);
> +}
> +
> +void test_vst1_s16_x2 (int16_t * ptr, int16x4x2_t val)
> +{
> + vst1_s16_x2 (ptr, val);
> +}
> +
> +void test_vst1_s32_x2 (int32_t * ptr, int32x2x2_t val)
> +{
> + vst1_s32_x2 (ptr, val);
> +}
> +
> +void test_vst1_s64_x2 (int64_t * ptr, int64x1x2_t val)
> +{
> + vst1_s64_x2 (ptr, val);
> +}
> +
> +void test_vst1_f32_x2 (float32_t * ptr, float32x2x2_t val)
> +{
> + vst1_f32_x2 (ptr, val);
> +}
> +
> +void test_vst1_p8_x2 (poly8_t * ptr, poly8x8x2_t val)
> +{
> + vst1_p8_x2 (ptr, val);
> +}
> +
> +void test_vst1_p16_x2 (poly16_t * ptr, poly16x4x2_t val)
> +{
> + vst1_p16_x2 (ptr, val);
> +}
> +
> +
> +/* { dg-final { scan-assembler-times {vst1.8\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+\]\n} 3 } } */
> +/* { dg-final { scan-assembler-times {vst1.16\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+\]\n} 3 } } */
> +/* { dg-final { scan-assembler-times {vst1.32\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+\]\n} 3 } } */
> +/* { dg-final { scan-assembler-times {vst1.64\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+:64\]\n} 2 } } */
> diff --git a/gcc/testsuite/gcc.target/arm/simd/vst1_bf16_xN_1.c b/gcc/testsuite/gcc.target/arm/simd/vst1_bf16_xN_1.c
> new file mode 100644
> index 00000000000..213fd20ee65
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/arm/simd/vst1_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"
> +
> +void test_vst1_bf16_x2 (bfloat16_t * ptr, bfloat16x4x2_t val)
> +{
> + vst1_bf16_x2 (ptr, val);
> +}
> +
> +/* { dg-final { scan-assembler-times {vst1.16\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+\]\n} 1 } } */
> diff --git a/gcc/testsuite/gcc.target/arm/simd/vst1_fp16_xN_1.c b/gcc/testsuite/gcc.target/arm/simd/vst1_fp16_xN_1.c
> new file mode 100644
> index 00000000000..523aec92db2
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/arm/simd/vst1_fp16_xN_1.c
> @@ -0,0 +1,13 @@
> +/* { 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"
> +
> +void test_vst1_f16_x2 (float16_t * ptr, float16x4x2_t val)
> +{
> + vst1_f16_x2 (ptr, val);
> +}
> +
> +/* { dg-final { scan-assembler-times {vst1.16\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+\]\n} 1 } } */
> diff --git a/gcc/testsuite/gcc.target/arm/simd/vst1_p64_xN_1.c b/gcc/testsuite/gcc.target/arm/simd/vst1_p64_xN_1.c
> new file mode 100644
> index 00000000000..f590ebd7b94
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/arm/simd/vst1_p64_xN_1.c
> @@ -0,0 +1,13 @@
> +/* { dg-do assemble } */
> +/* { dg-require-effective-target arm_crypto_ok } */
> +/* { dg-options "-save-temps -O2" } */
> +/* { dg-add-options arm_crypto } */
> +
> +#include "arm_neon.h"
> +
> +void test_vst1_p64_x2 (poly64_t * ptr, poly64x1x2_t val)
> +{
> + vst1_p64_x2 (ptr, val);
> +}
> +
> +/* { dg-final { scan-assembler-times {vst1.64\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+:64\]\n} 1 } } */
> \ No newline at end of file
^ permalink raw reply [flat|nested] 8+ messages in thread
* Re: [PATCH 1/3] [GCC] arm: vst1_types_x2 ACLE intrinsics
2023-11-27 14:56 ` Richard Earnshaw
@ 2023-11-27 14:57 ` Richard Earnshaw
0 siblings, 0 replies; 8+ messages in thread
From: Richard Earnshaw @ 2023-11-27 14:57 UTC (permalink / raw)
To: Ezra.Sitorus, gcc-patches; +Cc: richard.earnshaw, kyrylo.tkachov
On 27/11/2023 14:56, Richard Earnshaw wrote:
>
>
> On 06/10/2023 12:55, Ezra.Sitorus@arm.com wrote:
>> From: Ezra Sitorus <ezra.sitorus@arm.com>
>>
>> This patch is part of a series of patches implementing the _xN
>> variants of the vst1 intrinsic for arm32.
>
> We avoid the terms arm32 and arm64 in the gnu toolchain because of the
> potential for conflict with wild-card regexps for old CPUs (arm3* and
> arm6*). Please just use 'arm' in this case.
>
>> This patch adds the _x2 variants of the vst1 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/
>>
>
> Please reformat this text section to be no more than 70 columns, so that
> it doesn't produce awkward line wraps in the commit message.
>
>> gcc/ChangeLog:
>> * config/arm/arm_neon.h
>> (vst1_u8_x2, vst1_u16_x2, vst1_u32_x2, vst1_u64_x32): New.
>
> This line appears to be a duplicate of the one below, except for the
> bogus ..._x32 at the end. I think just drop it.
Ah, the difference is u<size> vs s<size>. So please fix the last entry.
R.
>
>> (vst1_s8_x2, vst1_s16_x2, vst1_s32_x2, vst1_s64_x2): New.
>> (vst1_f16_x2, vst1_f32_x2): New.
>> (vst1_p8_x2, vst1_p16_x2, vst1_p64_x2): New.
>> (vst1_bf16_x2): New.
>> * config/arm/arm_neon_builtins.def (vst1_x2): New entries.
>> * config/arm/neon.md (vst1_x2<mode>): New.
>>
>> gcc/testsuite/ChangeLog:
>> * gcc.target/arm/simd/vst1_base_xN_1.c: Add new tests.
>> * gcc.target/arm/simd/vst1_bf16_xN_1.c: Add new tests.
>> * gcc.target/arm/simd/vst1_fp16_xN_1.c: Add new tests.
>> * gcc.target/arm/simd/vst1_p64_xN_1.c: Add new tests.
>
> OK with the above changes.
>
> R.
>
>> ---
>> gcc/config/arm/arm_neon.h | 114 ++++++++++++++++++
>> gcc/config/arm/arm_neon_builtins.def | 1 +
>> gcc/config/arm/neon.md | 10 ++
>> .../gcc.target/arm/simd/vst1_base_xN_1.c | 67 ++++++++++
>> .../gcc.target/arm/simd/vst1_bf16_xN_1.c | 13 ++
>> .../gcc.target/arm/simd/vst1_fp16_xN_1.c | 13 ++
>> .../gcc.target/arm/simd/vst1_p64_xN_1.c | 13 ++
>> 7 files changed, 231 insertions(+)
>> create mode 100644 gcc/testsuite/gcc.target/arm/simd/vst1_base_xN_1.c
>> create mode 100644 gcc/testsuite/gcc.target/arm/simd/vst1_bf16_xN_1.c
>> create mode 100644 gcc/testsuite/gcc.target/arm/simd/vst1_fp16_xN_1.c
>> create mode 100644 gcc/testsuite/gcc.target/arm/simd/vst1_p64_xN_1.c
>>
>> diff --git a/gcc/config/arm/arm_neon.h b/gcc/config/arm/arm_neon.h
>> index c03be9912f8..4bd6093281b 100644
>> --- a/gcc/config/arm/arm_neon.h
>> +++ b/gcc/config/arm/arm_neon.h
>> @@ -11242,6 +11242,14 @@ vst1_p64 (poly64_t * __a, poly64x1_t __b)
>> __builtin_neon_vst1di ((__builtin_neon_di *) __a, __b);
>> }
>> +__extension__ extern __inline void
>> +__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
>> +vst1_p64_x2 (poly64_t * __a, poly64x1x2_t __b)
>> +{
>> + union { poly64x1x2_t __i; __builtin_neon_ti __o; } __bu = { __b };
>> + __builtin_neon_vst1_x2di ((__builtin_neon_di *) __a, __bu.__o);
>> +}
>> +
>> #pragma GCC pop_options
>> __extension__ extern __inline void
>> __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
>> @@ -11271,6 +11279,38 @@ vst1_s64 (int64_t * __a, int64x1_t __b)
>> __builtin_neon_vst1di ((__builtin_neon_di *) __a, __b);
>> }
>> +__extension__ extern __inline void
>> +__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
>> +vst1_s8_x2 (int8_t * __a, int8x8x2_t __b)
>> +{
>> + union { int8x8x2_t __i; __builtin_neon_ti __o; } __bu = { __b };
>> + __builtin_neon_vst1_x2v8qi ((__builtin_neon_qi *) __a, __bu.__o);
>> +}
>> +
>> +__extension__ extern __inline void
>> +__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
>> +vst1_s16_x2 (int16_t * __a, int16x4x2_t __b)
>> +{
>> + union { int16x4x2_t __i; __builtin_neon_ti __o; } __bu = { __b };
>> + __builtin_neon_vst1_x2v4hi ((__builtin_neon_hi *) __a, __bu.__o);
>> +}
>> +
>> +__extension__ extern __inline void
>> +__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
>> +vst1_s32_x2 (int32_t * __a, int32x2x2_t __b)
>> +{
>> + union { int32x2x2_t __i; __builtin_neon_ti __o; } __bu = { __b };
>> + __builtin_neon_vst1_x2v2si ((__builtin_neon_si *) __a, __bu.__o);
>> +}
>> +
>> +__extension__ extern __inline void
>> +__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
>> +vst1_s64_x2 (int64_t * __a, int64x1x2_t __b)
>> +{
>> + union { int64x1x2_t __i; __builtin_neon_ti __o; } __bu = { __b };
>> + __builtin_neon_vst1_x2di ((__builtin_neon_di *) __a, __bu.__o);
>> +}
>> +
>> #if defined (__ARM_FP16_FORMAT_IEEE) || defined
>> (__ARM_FP16_FORMAT_ALTERNATIVE)
>> __extension__ extern __inline void
>> __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
>> @@ -11287,6 +11327,24 @@ vst1_f32 (float32_t * __a, float32x2_t __b)
>> __builtin_neon_vst1v2sf ((__builtin_neon_sf *) __a, __b);
>> }
>> +#if defined (__ARM_FP16_FORMAT_IEEE) || defined
>> (__ARM_FP16_FORMAT_ALTERNATIVE)
>> +__extension__ extern __inline void
>> +__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
>> +vst1_f16_x2 (float16_t * __a, float16x4x2_t __b)
>> +{
>> + union { float16x4x2_t __i; __builtin_neon_ti __o; } __bu = { __b };
>> + __builtin_neon_vst1_x2v4hf (__a, __bu.__o);
>> +}
>> +#endif
>> +
>> +__extension__ extern __inline void
>> +__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
>> +vst1_f32_x2 (float32_t * __a, float32x2x2_t __b)
>> +{
>> + union { float32x2x2_t __i; __builtin_neon_ti __o; } __bu = { __b };
>> + __builtin_neon_vst1_x2v2sf ((__builtin_neon_sf *) __a, __bu.__o);
>> +}
>> +
>> __extension__ extern __inline void
>> __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
>> vst1_u8 (uint8_t * __a, uint8x8_t __b)
>> @@ -11315,6 +11373,38 @@ vst1_u64 (uint64_t * __a, uint64x1_t __b)
>> __builtin_neon_vst1di ((__builtin_neon_di *) __a, (int64x1_t) __b);
>> }
>> +__extension__ extern __inline void
>> +__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
>> +vst1_u8_x2 (uint8_t * __a, uint8x8x2_t __b)
>> +{
>> + union { uint8x8x2_t __i; __builtin_neon_ti __o; } __bu = { __b };
>> + __builtin_neon_vst1_x2v8qi ((__builtin_neon_qi *) __a, __bu.__o);
>> +}
>> +
>> +__extension__ extern __inline void
>> +__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
>> +vst1_u16_x2 (uint16_t * __a, uint16x4x2_t __b)
>> +{
>> + union { uint16x4x2_t __i; __builtin_neon_ti __o; } __bu = { __b };
>> + __builtin_neon_vst1_x2v4hi ((__builtin_neon_hi *) __a, __bu.__o);
>> +}
>> +
>> +__extension__ extern __inline void
>> +__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
>> +vst1_u32_x2 (uint32_t * __a, uint32x2x2_t __b)
>> +{
>> + union { uint32x2x2_t __i; __builtin_neon_ti __o; } __bu = { __b };
>> + __builtin_neon_vst1_x2v2si ((__builtin_neon_si *) __a, __bu.__o);
>> +}
>> +
>> +__extension__ extern __inline void
>> +__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
>> +vst1_u64_x2 (uint64_t * __a, uint64x1x2_t __b)
>> +{
>> + union { uint64x1x2_t __i; __builtin_neon_ti __o; } __bu = { __b };
>> + __builtin_neon_vst1_x2di ((__builtin_neon_di *) __a, __bu.__o);
>> +}
>> +
>> __extension__ extern __inline void
>> __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
>> vst1_p8 (poly8_t * __a, poly8x8_t __b)
>> @@ -11329,6 +11419,22 @@ vst1_p16 (poly16_t * __a, poly16x4_t __b)
>> __builtin_neon_vst1v4hi ((__builtin_neon_hi *) __a, (int16x4_t) __b);
>> }
>> +__extension__ extern __inline void
>> +__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
>> +vst1_p8_x2 (poly8_t * __a, poly8x8x2_t __b)
>> +{
>> + union { poly8x8x2_t __i; __builtin_neon_ti __o; } __bu = { __b };
>> + __builtin_neon_vst1_x2v8qi ((__builtin_neon_qi *) __a, __bu.__o);
>> +}
>> +
>> +__extension__ extern __inline void
>> +__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
>> +vst1_p16_x2 (poly16_t * __a, poly16x4x2_t __b)
>> +{
>> + union { poly16x4x2_t __i; __builtin_neon_ti __o; } __bu = { __b };
>> + __builtin_neon_vst1_x2v4hi ((__builtin_neon_hi *) __a, __bu.__o);
>> +}
>> +
>> #pragma GCC push_options
>> #pragma GCC target ("fpu=crypto-neon-fp-armv8")
>> __extension__ extern __inline void
>> @@ -20070,6 +20176,14 @@ vst1_bf16 (bfloat16_t * __a, bfloat16x4_t __b)
>> __builtin_neon_vst1v4bf (__a, __b);
>> }
>> +__extension__ extern __inline void
>> +__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
>> +vst1_bf16_x2 (bfloat16_t * __a, bfloat16x4x2_t __b)
>> +{
>> + union { bfloat16x4x2_t __i; __builtin_neon_ti __o; } __bu = { __b };
>> + __builtin_neon_vst1_x2v4hf ((__builtin_neon_bf *) __a, __bu.__o);
>> +}
>> +
>> __extension__ extern __inline void
>> __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
>> vst1q_bf16 (bfloat16_t * __a, bfloat16x8_t __b)
>> diff --git a/gcc/config/arm/arm_neon_builtins.def
>> b/gcc/config/arm/arm_neon_builtins.def
>> index f4001b298c4..7aef6f958cd 100644
>> --- a/gcc/config/arm/arm_neon_builtins.def
>> +++ b/gcc/config/arm/arm_neon_builtins.def
>> @@ -308,6 +308,7 @@ VAR12 (LOAD1LANE, vld1_lane,
>> v8qi, v4hi, v2si, v2sf, di, v16qi, v8hi, v4si, v4sf, v2di, v4bf,
>> v8bf)
>> VAR10 (LOAD1, vld1_dup,
>> v8qi, v4hi, v2si, v2sf, di, v16qi, v8hi, v4si, v4sf, v2di)
>> +VAR7 (STORE1, vst1_x2, v8qi, v4hi, v2si, di, v4hf, v2sf, v4bf)
>> VAR14 (STORE1, vst1,
>> v8qi, v4hi, v4hf, v2si, v2sf, di, v16qi, v8hi, v8hf, v4si,
>> v4sf, v2di,
>> v4bf, v8bf)
>> diff --git a/gcc/config/arm/neon.md b/gcc/config/arm/neon.md
>> index 62decab37a2..7a10e2cb61e 100644
>> --- a/gcc/config/arm/neon.md
>> +++ b/gcc/config/arm/neon.md
>> @@ -5125,6 +5125,16 @@ if (BYTES_BIG_ENDIAN)
>> UNSPEC_VST1))]
>> "TARGET_NEON")
>> +(define_insn "neon_vst1_x2<mode>"
>> + [(set (match_operand:TI 0 "neon_struct_operand" "=Um")
>> + (unspec:TI [(match_operand:TI 1 "s_register_operand" "w")
>> + (unspec:VDQX [(const_int 0)] UNSPEC_VSTRUCTDUMMY)]
>> + UNSPEC_VST1))]
>> + "TARGET_NEON"
>> + "vst1.<V_sz_elem>\t%h1, %A0"
>> + [(set_attr "type" "neon_store1_2reg<q>")]
>> +)
>> +
>> (define_insn "neon_vst1<mode>"
>> [(set (match_operand:VDQX 0 "neon_struct_operand" "=Um")
>> (unspec:VDQX [(match_operand:VDQX 1 "s_register_operand" "w")]
>> diff --git a/gcc/testsuite/gcc.target/arm/simd/vst1_base_xN_1.c
>> b/gcc/testsuite/gcc.target/arm/simd/vst1_base_xN_1.c
>> new file mode 100644
>> index 00000000000..575897fa422
>> --- /dev/null
>> +++ b/gcc/testsuite/gcc.target/arm/simd/vst1_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"
>> +
>> +void test_vst1_u8_x2 (uint8_t * ptr, uint8x8x2_t val)
>> +{
>> + vst1_u8_x2 (ptr, val);
>> +}
>> +
>> +void test_vst1_u16_x2 (uint16_t * ptr, uint16x4x2_t val)
>> +{
>> + vst1_u16_x2 (ptr, val);
>> +}
>> +
>> +void test_vst1_u32_x2 (uint32_t * ptr, uint32x2x2_t val)
>> +{
>> + vst1_u32_x2 (ptr, val);
>> +}
>> +
>> +void test_vst1_u64_x2 (uint64_t * ptr, uint64x1x2_t val)
>> +{
>> + vst1_u64_x2 (ptr, val);
>> +}
>> +
>> +void test_vst1_s8_x2 (int8_t * ptr, int8x8x2_t val)
>> +{
>> + vst1_s8_x2 (ptr, val);
>> +}
>> +
>> +void test_vst1_s16_x2 (int16_t * ptr, int16x4x2_t val)
>> +{
>> + vst1_s16_x2 (ptr, val);
>> +}
>> +
>> +void test_vst1_s32_x2 (int32_t * ptr, int32x2x2_t val)
>> +{
>> + vst1_s32_x2 (ptr, val);
>> +}
>> +
>> +void test_vst1_s64_x2 (int64_t * ptr, int64x1x2_t val)
>> +{
>> + vst1_s64_x2 (ptr, val);
>> +}
>> +
>> +void test_vst1_f32_x2 (float32_t * ptr, float32x2x2_t val)
>> +{
>> + vst1_f32_x2 (ptr, val);
>> +}
>> +
>> +void test_vst1_p8_x2 (poly8_t * ptr, poly8x8x2_t val)
>> +{
>> + vst1_p8_x2 (ptr, val);
>> +}
>> +
>> +void test_vst1_p16_x2 (poly16_t * ptr, poly16x4x2_t val)
>> +{
>> + vst1_p16_x2 (ptr, val);
>> +}
>> +
>> +
>> +/* { dg-final { scan-assembler-times {vst1.8\t\{d[0-9]+-d[0-9]+\},
>> \[r[0-9]+\]\n} 3 } } */
>> +/* { dg-final { scan-assembler-times {vst1.16\t\{d[0-9]+-d[0-9]+\},
>> \[r[0-9]+\]\n} 3 } } */
>> +/* { dg-final { scan-assembler-times {vst1.32\t\{d[0-9]+-d[0-9]+\},
>> \[r[0-9]+\]\n} 3 } } */
>> +/* { dg-final { scan-assembler-times {vst1.64\t\{d[0-9]+-d[0-9]+\},
>> \[r[0-9]+:64\]\n} 2 } } */
>> diff --git a/gcc/testsuite/gcc.target/arm/simd/vst1_bf16_xN_1.c
>> b/gcc/testsuite/gcc.target/arm/simd/vst1_bf16_xN_1.c
>> new file mode 100644
>> index 00000000000..213fd20ee65
>> --- /dev/null
>> +++ b/gcc/testsuite/gcc.target/arm/simd/vst1_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"
>> +
>> +void test_vst1_bf16_x2 (bfloat16_t * ptr, bfloat16x4x2_t val)
>> +{
>> + vst1_bf16_x2 (ptr, val);
>> +}
>> +
>> +/* { dg-final { scan-assembler-times {vst1.16\t\{d[0-9]+-d[0-9]+\},
>> \[r[0-9]+\]\n} 1 } } */
>> diff --git a/gcc/testsuite/gcc.target/arm/simd/vst1_fp16_xN_1.c
>> b/gcc/testsuite/gcc.target/arm/simd/vst1_fp16_xN_1.c
>> new file mode 100644
>> index 00000000000..523aec92db2
>> --- /dev/null
>> +++ b/gcc/testsuite/gcc.target/arm/simd/vst1_fp16_xN_1.c
>> @@ -0,0 +1,13 @@
>> +/* { 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"
>> +
>> +void test_vst1_f16_x2 (float16_t * ptr, float16x4x2_t val)
>> +{
>> + vst1_f16_x2 (ptr, val);
>> +}
>> +
>> +/* { dg-final { scan-assembler-times {vst1.16\t\{d[0-9]+-d[0-9]+\},
>> \[r[0-9]+\]\n} 1 } } */
>> diff --git a/gcc/testsuite/gcc.target/arm/simd/vst1_p64_xN_1.c
>> b/gcc/testsuite/gcc.target/arm/simd/vst1_p64_xN_1.c
>> new file mode 100644
>> index 00000000000..f590ebd7b94
>> --- /dev/null
>> +++ b/gcc/testsuite/gcc.target/arm/simd/vst1_p64_xN_1.c
>> @@ -0,0 +1,13 @@
>> +/* { dg-do assemble } */
>> +/* { dg-require-effective-target arm_crypto_ok } */
>> +/* { dg-options "-save-temps -O2" } */
>> +/* { dg-add-options arm_crypto } */
>> +
>> +#include "arm_neon.h"
>> +
>> +void test_vst1_p64_x2 (poly64_t * ptr, poly64x1x2_t val)
>> +{
>> + vst1_p64_x2 (ptr, val);
>> +}
>> +
>> +/* { dg-final { scan-assembler-times {vst1.64\t\{d[0-9]+-d[0-9]+\},
>> \[r[0-9]+:64\]\n} 1 } } */
>> \ No newline at end of file
^ permalink raw reply [flat|nested] 8+ messages in thread
* Re: [PATCH 2/3] [GCC] arm: vst1_types_x3 ACLE intrinsics
2023-10-06 11:55 ` [PATCH 2/3] [GCC] arm: vst1_types_x3 " Ezra.Sitorus
@ 2023-11-27 14:58 ` Richard Earnshaw
0 siblings, 0 replies; 8+ messages in thread
From: Richard Earnshaw @ 2023-11-27 14:58 UTC (permalink / raw)
To: Ezra.Sitorus, gcc-patches; +Cc: richard.earnshaw, kyrylo.tkachov
On 06/10/2023 12:55, Ezra.Sitorus@arm.com wrote:
> From: Ezra Sitorus <ezra.sitorus@arm.com>
>
> This patch is part of a series of patches implementing the _xN variants of the vst1 intrinsic for arm32.
> This patch adds the _x3 variants of the vst1 intrinsic.
>
OK, but see comments on the first patch about naming and formatting.
R.
> 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
> (vst1_u8_x3, vst1_u16_x3, vst1_u32_x3, vst1_u64_x3): New.
> (vst1_s8_x3, vst1_s16_x3, vst1_s32_x3, vst1_s64_x3): New.
> (vst1_f16_x3, vst1_f32_x3): New.
> (vst1_p8_x3, vst1_p16_x3, vst1_p64_x3): New.
> (vst1_bf16_x3): New.
> * config/arm/arm_neon_builtins.def (vst1_x3): New entries.
> * config/arm/neon.md (vst1_x3<mode>): New.
>
> gcc/testsuite/ChangeLog:
> * gcc.target/arm/simd/vst1_base_xN_1.c: Add new test.
> * gcc.target/arm/simd/vst1_bf16_xN_1.c: Add new test.
> * gcc.target/arm/simd/vst1_fp16_xN_1.c: Add new test.
> * gcc.target/arm/simd/vst1_p64_xN_1.c: Add new test.
> ---
> gcc/config/arm/arm_neon.h | 114 ++++++++++++++++++
> gcc/config/arm/arm_neon_builtins.def | 1 +
> gcc/config/arm/neon.md | 10 ++
> .../gcc.target/arm/simd/vst1_base_xN_1.c | 63 +++++++++-
> .../gcc.target/arm/simd/vst1_bf16_xN_1.c | 7 +-
> .../gcc.target/arm/simd/vst1_fp16_xN_1.c | 7 +-
> .../gcc.target/arm/simd/vst1_p64_xN_1.c | 7 +-
> 7 files changed, 202 insertions(+), 7 deletions(-)
>
> diff --git a/gcc/config/arm/arm_neon.h b/gcc/config/arm/arm_neon.h
> index 4bd6093281b..b01171e5966 100644
> --- a/gcc/config/arm/arm_neon.h
> +++ b/gcc/config/arm/arm_neon.h
> @@ -11250,6 +11250,14 @@ vst1_p64_x2 (poly64_t * __a, poly64x1x2_t __b)
> __builtin_neon_vst1_x2di ((__builtin_neon_di *) __a, __bu.__o);
> }
>
> +__extension__ extern __inline void
> +__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> +vst1_p64_x3 (poly64_t * __a, poly64x1x3_t __b)
> +{
> + union { poly64x1x3_t __i; __builtin_neon_ei __o; } __bu = { __b };
> + __builtin_neon_vst1_x3di ((__builtin_neon_di *) __a, __bu.__o);
> +}
> +
> #pragma GCC pop_options
> __extension__ extern __inline void
> __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> @@ -11311,6 +11319,38 @@ vst1_s64_x2 (int64_t * __a, int64x1x2_t __b)
> __builtin_neon_vst1_x2di ((__builtin_neon_di *) __a, __bu.__o);
> }
>
> +__extension__ extern __inline void
> +__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> +vst1_s8_x3 (int8_t * __a, int8x8x3_t __b)
> +{
> + union { int8x8x3_t __i; __builtin_neon_ei __o; } __bu = { __b };
> + __builtin_neon_vst1_x3v8qi ((__builtin_neon_qi *) __a, __bu.__o);
> +}
> +
> +__extension__ extern __inline void
> +__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> +vst1_s16_x3 (int16_t * __a, int16x4x3_t __b)
> +{
> + union { int16x4x3_t __i; __builtin_neon_ei __o; } __bu = { __b };
> + __builtin_neon_vst1_x3v4hi ((__builtin_neon_hi *) __a, __bu.__o);
> +}
> +
> +__extension__ extern __inline void
> +__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> +vst1_s32_x3 (int32_t * __a, int32x2x3_t __b)
> +{
> + union { int32x2x3_t __i; __builtin_neon_ei __o; } __bu = { __b };
> + __builtin_neon_vst1_x3v2si ((__builtin_neon_si *) __a, __bu.__o);
> +}
> +
> +__extension__ extern __inline void
> +__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> +vst1_s64_x3 (int64_t * __a, int64x1x3_t __b)
> +{
> + union { int64x1x3_t __i; __builtin_neon_ei __o; } __bu = { __b };
> + __builtin_neon_vst1_x3di ((__builtin_neon_di *) __a, __bu.__o);
> +}
> +
> #if defined (__ARM_FP16_FORMAT_IEEE) || defined (__ARM_FP16_FORMAT_ALTERNATIVE)
> __extension__ extern __inline void
> __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> @@ -11345,6 +11385,24 @@ vst1_f32_x2 (float32_t * __a, float32x2x2_t __b)
> __builtin_neon_vst1_x2v2sf ((__builtin_neon_sf *) __a, __bu.__o);
> }
>
> +#if defined (__ARM_FP16_FORMAT_IEEE) || defined (__ARM_FP16_FORMAT_ALTERNATIVE)
> +__extension__ extern __inline void
> +__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> +vst1_f16_x3 (float16_t * __a, float16x4x3_t __b)
> +{
> + union { float16x4x3_t __i; __builtin_neon_ei __o; } __bu = { __b };
> + __builtin_neon_vst1_x3v4hf (__a, __bu.__o);
> +}
> +#endif
> +
> +__extension__ extern __inline void
> +__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> +vst1_f32_x3 (float32_t * __a, float32x2x3_t __b)
> +{
> + union { float32x2x3_t __i; __builtin_neon_ei __o; } __bu = { __b };
> + __builtin_neon_vst1_x3v2sf ((__builtin_neon_sf *) __a, __bu.__o);
> +}
> +
> __extension__ extern __inline void
> __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> vst1_u8 (uint8_t * __a, uint8x8_t __b)
> @@ -11405,6 +11463,38 @@ vst1_u64_x2 (uint64_t * __a, uint64x1x2_t __b)
> __builtin_neon_vst1_x2di ((__builtin_neon_di *) __a, __bu.__o);
> }
>
> +__extension__ extern __inline void
> +__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> +vst1_u8_x3 (uint8_t * __a, uint8x8x3_t __b)
> +{
> + union { uint8x8x3_t __i; __builtin_neon_ei __o; } __bu = { __b };
> + __builtin_neon_vst1_x3v8qi ((__builtin_neon_qi *) __a, __bu.__o);
> +}
> +
> +__extension__ extern __inline void
> +__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> +vst1_u16_x3 (uint16_t * __a, uint16x4x3_t __b)
> +{
> + union { uint16x4x3_t __i; __builtin_neon_ei __o; } __bu = { __b };
> + __builtin_neon_vst1_x3v4hi ((__builtin_neon_hi *) __a, __bu.__o);
> +}
> +
> +__extension__ extern __inline void
> +__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> +vst1_u32_x3 (uint32_t * __a, uint32x2x3_t __b)
> +{
> + union { uint32x2x3_t __i; __builtin_neon_ei __o; } __bu = { __b };
> + __builtin_neon_vst1_x3v2si ((__builtin_neon_si *) __a, __bu.__o);
> +}
> +
> +__extension__ extern __inline void
> +__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> +vst1_u64_x3 (uint64_t * __a, uint64x1x3_t __b)
> +{
> + union { uint64x1x3_t __i; __builtin_neon_ei __o; } __bu = { __b };
> + __builtin_neon_vst1_x3di ((__builtin_neon_di *) __a, __bu.__o);
> +}
> +
> __extension__ extern __inline void
> __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> vst1_p8 (poly8_t * __a, poly8x8_t __b)
> @@ -11435,6 +11525,22 @@ vst1_p16_x2 (poly16_t * __a, poly16x4x2_t __b)
> __builtin_neon_vst1_x2v4hi ((__builtin_neon_hi *) __a, __bu.__o);
> }
>
> +__extension__ extern __inline void
> +__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> +vst1_p8_x3 (poly8_t * __a, poly8x8x3_t __b)
> +{
> + union { poly8x8x3_t __i; __builtin_neon_ei __o; } __bu = { __b };
> + __builtin_neon_vst1_x3v8qi ((__builtin_neon_qi *) __a, __bu.__o);
> +}
> +
> +__extension__ extern __inline void
> +__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> +vst1_p16_x3 (poly16_t * __a, poly16x4x3_t __b)
> +{
> + union { poly16x4x3_t __i; __builtin_neon_ei __o; } __bu = { __b };
> + __builtin_neon_vst1_x3v4hi ((__builtin_neon_hi *) __a, __bu.__o);
> +}
> +
> #pragma GCC push_options
> #pragma GCC target ("fpu=crypto-neon-fp-armv8")
> __extension__ extern __inline void
> @@ -20184,6 +20290,14 @@ vst1_bf16_x2 (bfloat16_t * __a, bfloat16x4x2_t __b)
> __builtin_neon_vst1_x2v4hf ((__builtin_neon_bf *) __a, __bu.__o);
> }
>
> +__extension__ extern __inline void
> +__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> +vst1_bf16_x3 (bfloat16_t * __a, bfloat16x4x3_t __b)
> +{
> + union { bfloat16x4x3_t __i; __builtin_neon_ei __o; } __bu = { __b };
> + __builtin_neon_vst1_x3v4hf ((__builtin_neon_bf *) __a, __bu.__o);
> +}
> +
> __extension__ extern __inline void
> __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> vst1q_bf16 (bfloat16_t * __a, bfloat16x8_t __b)
> diff --git a/gcc/config/arm/arm_neon_builtins.def b/gcc/config/arm/arm_neon_builtins.def
> index 7aef6f958cd..e33b3429f65 100644
> --- a/gcc/config/arm/arm_neon_builtins.def
> +++ b/gcc/config/arm/arm_neon_builtins.def
> @@ -309,6 +309,7 @@ VAR12 (LOAD1LANE, vld1_lane,
> VAR10 (LOAD1, vld1_dup,
> v8qi, v4hi, v2si, v2sf, di, v16qi, v8hi, v4si, v4sf, v2di)
> VAR7 (STORE1, vst1_x2, v8qi, v4hi, v2si, di, v4hf, v2sf, v4bf)
> +VAR7 (STORE1, vst1_x3, v8qi, v4hi, v2si, di, v4hf, v2sf, v4bf)
> VAR14 (STORE1, vst1,
> v8qi, v4hi, v4hf, v2si, v2sf, di, v16qi, v8hi, v8hf, v4si, v4sf, v2di,
> v4bf, v8bf)
> diff --git a/gcc/config/arm/neon.md b/gcc/config/arm/neon.md
> index 7a10e2cb61e..3d1d2aa7d06 100644
> --- a/gcc/config/arm/neon.md
> +++ b/gcc/config/arm/neon.md
> @@ -5135,6 +5135,16 @@ if (BYTES_BIG_ENDIAN)
> [(set_attr "type" "neon_store1_2reg<q>")]
> )
>
> +(define_insn "neon_vst1_x3<mode>"
> + [(set (match_operand:EI 0 "neon_struct_operand" "=Um")
> + (unspec:EI [(match_operand:EI 1 "s_register_operand" "w")
> + (unspec:VDQX [(const_int 0)] UNSPEC_VSTRUCTDUMMY)]
> + UNSPEC_VST1))]
> + "TARGET_NEON"
> + "vst1.<V_sz_elem>\t%h1, %A0"
> + [(set_attr "type" "neon_store1_3reg<q>")]
> +)
> +
> (define_insn "neon_vst1<mode>"
> [(set (match_operand:VDQX 0 "neon_struct_operand" "=Um")
> (unspec:VDQX [(match_operand:VDQX 1 "s_register_operand" "w")]
> diff --git a/gcc/testsuite/gcc.target/arm/simd/vst1_base_xN_1.c b/gcc/testsuite/gcc.target/arm/simd/vst1_base_xN_1.c
> index 575897fa422..5f820a6a496 100644
> --- a/gcc/testsuite/gcc.target/arm/simd/vst1_base_xN_1.c
> +++ b/gcc/testsuite/gcc.target/arm/simd/vst1_base_xN_1.c
> @@ -60,8 +60,63 @@ void test_vst1_p16_x2 (poly16_t * ptr, poly16x4x2_t val)
> vst1_p16_x2 (ptr, val);
> }
>
> +void test_vst1_u8_x3 (uint8_t * ptr, uint8x8x3_t val)
> +{
> + vst1_u8_x3 (ptr, val);
> +}
> +
> +void test_vst1_u16_x3 (uint16_t * ptr, uint16x4x3_t val)
> +{
> + vst1_u16_x3 (ptr, val);
> +}
> +
> +void test_vst1_u32_x3 (uint32_t * ptr, uint32x2x3_t val)
> +{
> + vst1_u32_x3 (ptr, val);
> +}
> +
> +void test_vst1_u64_x3 (uint64_t * ptr, uint64x1x3_t val)
> +{
> + vst1_u64_x3 (ptr, val);
> +}
> +
> +void test_vst1_s8_x3 (int8_t * ptr, int8x8x3_t val)
> +{
> + vst1_s8_x3 (ptr, val);
> +}
> +
> +void test_vst1_s16_x3 (int16_t * ptr, int16x4x3_t val)
> +{
> + vst1_s16_x3 (ptr, val);
> +}
> +
> +void test_vst1_s32_x3 (int32_t * ptr, int32x2x3_t val)
> +{
> + vst1_s32_x3 (ptr, val);
> +}
> +
> +void test_vst1_s64_x3 (int64_t * ptr, int64x1x3_t val)
> +{
> + vst1_s64_x3 (ptr, val);
> +}
> +
> +void test_vst1_f32_x3 (float32_t * ptr, float32x2x3_t val)
> +{
> + vst1_f32_x3 (ptr, val);
> +}
> +
> +void test_vst1_p8_x3 (poly8_t * ptr, poly8x8x3_t val)
> +{
> + vst1_p8_x3 (ptr, val);
> +}
> +
> +void test_vst1_p16_x3 (poly16_t * ptr, poly16x4x3_t val)
> +{
> + vst1_p16_x3 (ptr, val);
> +}
> +
>
> -/* { dg-final { scan-assembler-times {vst1.8\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+\]\n} 3 } } */
> -/* { dg-final { scan-assembler-times {vst1.16\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+\]\n} 3 } } */
> -/* { dg-final { scan-assembler-times {vst1.32\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+\]\n} 3 } } */
> -/* { dg-final { scan-assembler-times {vst1.64\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+:64\]\n} 2 } } */
> +/* { dg-final { scan-assembler-times {vst1.8\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+\]\n} 6 } } */
> +/* { dg-final { scan-assembler-times {vst1.16\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+\]\n} 6 } } */
> +/* { dg-final { scan-assembler-times {vst1.32\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+\]\n} 6 } } */
> +/* { dg-final { scan-assembler-times {vst1.64\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+:64\]\n} 4 } } */
> diff --git a/gcc/testsuite/gcc.target/arm/simd/vst1_bf16_xN_1.c b/gcc/testsuite/gcc.target/arm/simd/vst1_bf16_xN_1.c
> index 213fd20ee65..a3a00ead468 100644
> --- a/gcc/testsuite/gcc.target/arm/simd/vst1_bf16_xN_1.c
> +++ b/gcc/testsuite/gcc.target/arm/simd/vst1_bf16_xN_1.c
> @@ -10,4 +10,9 @@ void test_vst1_bf16_x2 (bfloat16_t * ptr, bfloat16x4x2_t val)
> vst1_bf16_x2 (ptr, val);
> }
>
> -/* { dg-final { scan-assembler-times {vst1.16\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+\]\n} 1 } } */
> +void test_vst1_bf16_x3 (bfloat16_t * ptr, bfloat16x4x3_t val)
> +{
> + vst1_bf16_x3 (ptr, val);
> +}
> +
> +/* { dg-final { scan-assembler-times {vst1.16\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+\]\n} 2 } } */
> diff --git a/gcc/testsuite/gcc.target/arm/simd/vst1_fp16_xN_1.c b/gcc/testsuite/gcc.target/arm/simd/vst1_fp16_xN_1.c
> index 523aec92db2..0a6863e24c6 100644
> --- a/gcc/testsuite/gcc.target/arm/simd/vst1_fp16_xN_1.c
> +++ b/gcc/testsuite/gcc.target/arm/simd/vst1_fp16_xN_1.c
> @@ -10,4 +10,9 @@ void test_vst1_f16_x2 (float16_t * ptr, float16x4x2_t val)
> vst1_f16_x2 (ptr, val);
> }
>
> -/* { dg-final { scan-assembler-times {vst1.16\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+\]\n} 1 } } */
> +void test_vst1_f16_x3 (float16_t * ptr, float16x4x3_t val)
> +{
> + vst1_f16_x3 (ptr, val);
> +}
> +
> +/* { dg-final { scan-assembler-times {vst1.16\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+\]\n} 2 } } */
> diff --git a/gcc/testsuite/gcc.target/arm/simd/vst1_p64_xN_1.c b/gcc/testsuite/gcc.target/arm/simd/vst1_p64_xN_1.c
> index f590ebd7b94..5dbd6049bc9 100644
> --- a/gcc/testsuite/gcc.target/arm/simd/vst1_p64_xN_1.c
> +++ b/gcc/testsuite/gcc.target/arm/simd/vst1_p64_xN_1.c
> @@ -10,4 +10,9 @@ void test_vst1_p64_x2 (poly64_t * ptr, poly64x1x2_t val)
> vst1_p64_x2 (ptr, val);
> }
>
> -/* { dg-final { scan-assembler-times {vst1.64\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+:64\]\n} 1 } } */
> \ No newline at end of file
> +void test_vst1_p64_x3 (poly64_t * ptr, poly64x1x3_t val)
> +{
> + vst1_p64_x3 (ptr, val);
> +}
> +
> +/* { dg-final { scan-assembler-times {vst1.64\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+:64\]\n} 2 } } */
> \ No newline at end of file
^ permalink raw reply [flat|nested] 8+ messages in thread
* Re: [PATCH 3/3] [GCC] arm: vst1_types_x4 ACLE intrinsics
2023-10-06 11:56 ` [PATCH 3/3] [GCC] arm: vst1_types_x4 " Ezra.Sitorus
@ 2023-11-27 14:59 ` Richard Earnshaw
0 siblings, 0 replies; 8+ messages in thread
From: Richard Earnshaw @ 2023-11-27 14:59 UTC (permalink / raw)
To: Ezra.Sitorus, gcc-patches; +Cc: richard.earnshaw, kyrylo.tkachov
On 06/10/2023 12:56, Ezra.Sitorus@arm.com wrote:
> From: Ezra Sitorus <ezra.sitorus@arm.com>
>
> This patch is part of a series of patches implementing the _xN variants of the vst1 intrinsic for arm32.
> This patch adds the _x4 variants of the vst1 intrinsic.
OK, but please see comment on first patch about naming and formatting.
R.
>
> 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
> (vst1_u8_x4, vst1_u16_x4, vst1_u32_x4, vst1_u64_x4): New.
> (vst1_s8_x4, vst1_s16_x4, vst1_s32_x4, vst1_s64_x4): New.
> (vst1_f16_x4, vst1_f32_x4): New.
> (vst1_p8_x4, vst1_p16_x4, vst1_p64_x4): New.
> (vst1_bf16_x4): New.
> * config/arm/arm_neon_builtins.def (vst1_x4): New entries.
> * config/arm/neon.md (vst1_x4<mode>): New.
>
> gcc/testsuite/ChangeLog:
> * gcc.target/arm/simd/vst1_base_xN_1.c: Add new test.
> * gcc.target/arm/simd/vst1_bf16_xN_1.c: Add new test.
> * gcc.target/arm/simd/vst1_fp16_xN_1.c: Add new test.
> * gcc.target/arm/simd/vst1_p64_xN_1.c: Add new test.
> ---
> gcc/config/arm/arm_neon.h | 114 ++++++++++++++++++
> gcc/config/arm/arm_neon_builtins.def | 1 +
> gcc/config/arm/neon.md | 10 ++
> .../gcc.target/arm/simd/vst1_base_xN_1.c | 62 +++++++++-
> .../gcc.target/arm/simd/vst1_bf16_xN_1.c | 6 +-
> .../gcc.target/arm/simd/vst1_fp16_xN_1.c | 7 +-
> .../gcc.target/arm/simd/vst1_p64_xN_1.c | 7 +-
> 7 files changed, 200 insertions(+), 7 deletions(-)
>
> diff --git a/gcc/config/arm/arm_neon.h b/gcc/config/arm/arm_neon.h
> index b01171e5966..41e645d8352 100644
> --- a/gcc/config/arm/arm_neon.h
> +++ b/gcc/config/arm/arm_neon.h
> @@ -11258,6 +11258,14 @@ vst1_p64_x3 (poly64_t * __a, poly64x1x3_t __b)
> __builtin_neon_vst1_x3di ((__builtin_neon_di *) __a, __bu.__o);
> }
>
> +__extension__ extern __inline void
> +__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> +vst1_p64_x4 (poly64_t * __a, poly64x1x4_t __b)
> +{
> + union { poly64x1x4_t __i; __builtin_neon_oi __o; } __bu = { __b };
> + __builtin_neon_vst1_x3di ((__builtin_neon_di *) __a, __bu.__o);
> +}
> +
> #pragma GCC pop_options
> __extension__ extern __inline void
> __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> @@ -11351,6 +11359,38 @@ vst1_s64_x3 (int64_t * __a, int64x1x3_t __b)
> __builtin_neon_vst1_x3di ((__builtin_neon_di *) __a, __bu.__o);
> }
>
> +__extension__ extern __inline void
> +__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> +vst1_s8_x4 (int8_t * __a, int8x8x4_t __b)
> +{
> + union { int8x8x4_t __i; __builtin_neon_oi __o; } __bu = { __b };
> + __builtin_neon_vst1_x4v8qi ((__builtin_neon_qi *) __a, __bu.__o);
> +}
> +
> +__extension__ extern __inline void
> +__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> +vst1_s16_x4 (int16_t * __a, int16x4x4_t __b)
> +{
> + union { int16x4x4_t __i; __builtin_neon_oi __o; } __bu = { __b };
> + __builtin_neon_vst1_x4v4hi ((__builtin_neon_hi *) __a, __bu.__o);
> +}
> +
> +__extension__ extern __inline void
> +__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> +vst1_s32_x4 (int32_t * __a, int32x2x4_t __b)
> +{
> + union { int32x2x4_t __i; __builtin_neon_oi __o; } __bu = { __b };
> + __builtin_neon_vst1_x4v2si ((__builtin_neon_si *) __a, __bu.__o);
> +}
> +
> +__extension__ extern __inline void
> +__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> +vst1_s64_x4 (int64_t * __a, int64x1x4_t __b)
> +{
> + union { int64x1x4_t __i; __builtin_neon_oi __o; } __bu = { __b };
> + __builtin_neon_vst1_x4di ((__builtin_neon_di *) __a, __bu.__o);
> +}
> +
> #if defined (__ARM_FP16_FORMAT_IEEE) || defined (__ARM_FP16_FORMAT_ALTERNATIVE)
> __extension__ extern __inline void
> __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> @@ -11403,6 +11443,24 @@ vst1_f32_x3 (float32_t * __a, float32x2x3_t __b)
> __builtin_neon_vst1_x3v2sf ((__builtin_neon_sf *) __a, __bu.__o);
> }
>
> +#if defined (__ARM_FP16_FORMAT_IEEE) || defined (__ARM_FP16_FORMAT_ALTERNATIVE)
> +__extension__ extern __inline void
> +__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> +vst1_f16_x4 (float16_t * __a, float16x4x4_t __b)
> +{
> + union { float16x4x4_t __i; __builtin_neon_oi __o; } __bu = { __b };
> + __builtin_neon_vst1_x4v4hf (__a, __bu.__o);
> +}
> +#endif
> +
> +__extension__ extern __inline void
> +__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> +vst1_f32_x4 (float32_t * __a, float32x2x4_t __b)
> +{
> + union { float32x2x4_t __i; __builtin_neon_oi __o; } __bu = { __b };
> + __builtin_neon_vst1_x4v2sf ((__builtin_neon_sf *) __a, __bu.__o);
> +}
> +
> __extension__ extern __inline void
> __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> vst1_u8 (uint8_t * __a, uint8x8_t __b)
> @@ -11495,6 +11553,38 @@ vst1_u64_x3 (uint64_t * __a, uint64x1x3_t __b)
> __builtin_neon_vst1_x3di ((__builtin_neon_di *) __a, __bu.__o);
> }
>
> +__extension__ extern __inline void
> +__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> +vst1_u8_x4 (uint8_t * __a, uint8x8x4_t __b)
> +{
> + union { uint8x8x4_t __i; __builtin_neon_oi __o; } __bu = { __b };
> + __builtin_neon_vst1_x4v8qi ((__builtin_neon_qi *) __a, __bu.__o);
> +}
> +
> +__extension__ extern __inline void
> +__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> +vst1_u16_x4 (uint16_t * __a, uint16x4x4_t __b)
> +{
> + union { uint16x4x4_t __i; __builtin_neon_oi __o; } __bu = { __b };
> + __builtin_neon_vst1_x4v4hi ((__builtin_neon_hi *) __a, __bu.__o);
> +}
> +
> +__extension__ extern __inline void
> +__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> +vst1_u32_x4 (uint32_t * __a, uint32x2x4_t __b)
> +{
> + union { uint32x2x4_t __i; __builtin_neon_oi __o; } __bu = { __b };
> + __builtin_neon_vst1_x4v2si ((__builtin_neon_si *) __a, __bu.__o);
> +}
> +
> +__extension__ extern __inline void
> +__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> +vst1_u64_x4 (uint64_t * __a, uint64x1x4_t __b)
> +{
> + union { uint64x1x4_t __i; __builtin_neon_oi __o; } __bu = { __b };
> + __builtin_neon_vst1_x4di ((__builtin_neon_di *) __a, __bu.__o);
> +}
> +
> __extension__ extern __inline void
> __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> vst1_p8 (poly8_t * __a, poly8x8_t __b)
> @@ -11541,6 +11631,22 @@ vst1_p16_x3 (poly16_t * __a, poly16x4x3_t __b)
> __builtin_neon_vst1_x3v4hi ((__builtin_neon_hi *) __a, __bu.__o);
> }
>
> +__extension__ extern __inline void
> +__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> +vst1_p8_x4 (poly8_t * __a, poly8x8x4_t __b)
> +{
> + union { poly8x8x4_t __i; __builtin_neon_oi __o; } __bu = { __b };
> + __builtin_neon_vst1_x4v8qi ((__builtin_neon_qi *) __a, __bu.__o);
> +}
> +
> +__extension__ extern __inline void
> +__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> +vst1_p16_x4 (poly16_t * __a, poly16x4x4_t __b)
> +{
> + union { poly16x4x4_t __i; __builtin_neon_oi __o; } __bu = { __b };
> + __builtin_neon_vst1_x4v4hi ((__builtin_neon_hi *) __a, __bu.__o);
> +}
> +
> #pragma GCC push_options
> #pragma GCC target ("fpu=crypto-neon-fp-armv8")
> __extension__ extern __inline void
> @@ -20298,6 +20404,14 @@ vst1_bf16_x3 (bfloat16_t * __a, bfloat16x4x3_t __b)
> __builtin_neon_vst1_x3v4hf ((__builtin_neon_bf *) __a, __bu.__o);
> }
>
> +__extension__ extern __inline void
> +__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> +vst1_bf16_x4 (bfloat16_t * __a, bfloat16x4x4_t __b)
> +{
> + union { bfloat16x4x4_t __i; __builtin_neon_oi __o; } __bu = { __b };
> + __builtin_neon_vst1_x4v4hf ((__builtin_neon_bf *) __a, __bu.__o);
> +}
> +
> __extension__ extern __inline void
> __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
> vst1q_bf16 (bfloat16_t * __a, bfloat16x8_t __b)
> diff --git a/gcc/config/arm/arm_neon_builtins.def b/gcc/config/arm/arm_neon_builtins.def
> index e33b3429f65..95300cb0fe4 100644
> --- a/gcc/config/arm/arm_neon_builtins.def
> +++ b/gcc/config/arm/arm_neon_builtins.def
> @@ -310,6 +310,7 @@ VAR10 (LOAD1, vld1_dup,
> v8qi, v4hi, v2si, v2sf, di, v16qi, v8hi, v4si, v4sf, v2di)
> VAR7 (STORE1, vst1_x2, v8qi, v4hi, v2si, di, v4hf, v2sf, v4bf)
> VAR7 (STORE1, vst1_x3, v8qi, v4hi, v2si, di, v4hf, v2sf, v4bf)
> +VAR7 (STORE1, vst1_x4, v8qi, v4hi, v2si, di, v4hf, v2sf, v4bf)
> VAR14 (STORE1, vst1,
> v8qi, v4hi, v4hf, v2si, v2sf, di, v16qi, v8hi, v8hf, v4si, v4sf, v2di,
> v4bf, v8bf)
> diff --git a/gcc/config/arm/neon.md b/gcc/config/arm/neon.md
> index 3d1d2aa7d06..f5d583129fa 100644
> --- a/gcc/config/arm/neon.md
> +++ b/gcc/config/arm/neon.md
> @@ -5145,6 +5145,16 @@ if (BYTES_BIG_ENDIAN)
> [(set_attr "type" "neon_store1_3reg<q>")]
> )
>
> +(define_insn "neon_vst1_x4<mode>"
> + [(set (match_operand:OI 0 "neon_struct_operand" "=Um")
> + (unspec:OI [(match_operand:OI 1 "s_register_operand" "w")
> + (unspec:VDQX [(const_int 0)] UNSPEC_VSTRUCTDUMMY)]
> + UNSPEC_VST1))]
> + "TARGET_NEON"
> + "vst1.<V_sz_elem>\t%h1, %A0"
> + [(set_attr "type" "neon_store1_4reg<q>")]
> +)
> +
> (define_insn "neon_vst1<mode>"
> [(set (match_operand:VDQX 0 "neon_struct_operand" "=Um")
> (unspec:VDQX [(match_operand:VDQX 1 "s_register_operand" "w")]
> diff --git a/gcc/testsuite/gcc.target/arm/simd/vst1_base_xN_1.c b/gcc/testsuite/gcc.target/arm/simd/vst1_base_xN_1.c
> index 5f820a6a496..04ca6583552 100644
> --- a/gcc/testsuite/gcc.target/arm/simd/vst1_base_xN_1.c
> +++ b/gcc/testsuite/gcc.target/arm/simd/vst1_base_xN_1.c
> @@ -115,8 +115,62 @@ void test_vst1_p16_x3 (poly16_t * ptr, poly16x4x3_t val)
> vst1_p16_x3 (ptr, val);
> }
>
> +void test_vst1_u8_x4 (uint8_t * ptr, uint8x8x4_t val)
> +{
> + vst1_u8_x4 (ptr, val);
> +}
> +
> +void test_vst1_u16_x4 (uint16_t * ptr, uint16x4x4_t val)
> +{
> + vst1_u16_x4 (ptr, val);
> +}
> +
> +void test_vst1_u32_x4 (uint32_t * ptr, uint32x2x4_t val)
> +{
> + vst1_u32_x4 (ptr, val);
> +}
> +
> +void test_vst1_u64_x4 (uint64_t * ptr, uint64x1x4_t val)
> +{
> + vst1_u64_x4 (ptr, val);
> +}
> +
> +void test_vst1_s8_x4 (int8_t * ptr, int8x8x4_t val)
> +{
> + vst1_s8_x4 (ptr, val);
> +}
> +
> +void test_vst1_s16_x4 (int16_t * ptr, int16x4x4_t val)
> +{
> + vst1_s16_x4 (ptr, val);
> +}
> +
> +void test_vst1_s32_x4 (int32_t * ptr, int32x2x4_t val)
> +{
> + vst1_s32_x4 (ptr, val);
> +}
> +
> +void test_vst1_s64_x4 (int64_t * ptr, int64x1x4_t val)
> +{
> + vst1_s64_x4 (ptr, val);
> +}
> +
> +void test_vst1_f32_x4 (float32_t * ptr, float32x2x4_t val)
> +{
> + vst1_f32_x4 (ptr, val);
> +}
> +
> +void test_vst1_p8_x4 (poly8_t * ptr, poly8x8x4_t val)
> +{
> + vst1_p8_x4 (ptr, val);
> +}
> +
> +void test_vst1_p16_x4 (poly16_t * ptr, poly16x4x4_t val)
> +{
> + vst1_p16_x4 (ptr, val);
> +}
>
> -/* { dg-final { scan-assembler-times {vst1.8\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+\]\n} 6 } } */
> -/* { dg-final { scan-assembler-times {vst1.16\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+\]\n} 6 } } */
> -/* { dg-final { scan-assembler-times {vst1.32\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+\]\n} 6 } } */
> -/* { dg-final { scan-assembler-times {vst1.64\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+:64\]\n} 4 } } */
> +/* { dg-final { scan-assembler-times {vst1.8\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+\]\n} 9 } } */
> +/* { dg-final { scan-assembler-times {vst1.16\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+\]\n} 9 } } */
> +/* { dg-final { scan-assembler-times {vst1.32\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+\]\n} 9 } } */
> +/* { dg-final { scan-assembler-times {vst1.64\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+:64\]\n} 6 } } */
> diff --git a/gcc/testsuite/gcc.target/arm/simd/vst1_bf16_xN_1.c b/gcc/testsuite/gcc.target/arm/simd/vst1_bf16_xN_1.c
> index a3a00ead468..d919c7d060d 100644
> --- a/gcc/testsuite/gcc.target/arm/simd/vst1_bf16_xN_1.c
> +++ b/gcc/testsuite/gcc.target/arm/simd/vst1_bf16_xN_1.c
> @@ -15,4 +15,8 @@ void test_vst1_bf16_x3 (bfloat16_t * ptr, bfloat16x4x3_t val)
> vst1_bf16_x3 (ptr, val);
> }
>
> -/* { dg-final { scan-assembler-times {vst1.16\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+\]\n} 2 } } */
> +void test_vst1_bf16_x4 (bfloat16_t * ptr, bfloat16x4x4_t val)
> +{
> + vst1_bf16_x4 (ptr, val);
> +}
> +/* { dg-final { scan-assembler-times {vst1.16\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+\]\n} 3 } } */
> diff --git a/gcc/testsuite/gcc.target/arm/simd/vst1_fp16_xN_1.c b/gcc/testsuite/gcc.target/arm/simd/vst1_fp16_xN_1.c
> index 0a6863e24c6..3d1d1eb7ad1 100644
> --- a/gcc/testsuite/gcc.target/arm/simd/vst1_fp16_xN_1.c
> +++ b/gcc/testsuite/gcc.target/arm/simd/vst1_fp16_xN_1.c
> @@ -15,4 +15,9 @@ void test_vst1_f16_x3 (float16_t * ptr, float16x4x3_t val)
> vst1_f16_x3 (ptr, val);
> }
>
> -/* { dg-final { scan-assembler-times {vst1.16\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+\]\n} 2 } } */
> +void test_vst1_f16_x4 (float16_t * ptr, float16x4x4_t val)
> +{
> + vst1_f16_x4 (ptr, val);
> +}
> +
> +/* { dg-final { scan-assembler-times {vst1.16\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+\]\n} 3 } } */
> diff --git a/gcc/testsuite/gcc.target/arm/simd/vst1_p64_xN_1.c b/gcc/testsuite/gcc.target/arm/simd/vst1_p64_xN_1.c
> index 5dbd6049bc9..62912143481 100644
> --- a/gcc/testsuite/gcc.target/arm/simd/vst1_p64_xN_1.c
> +++ b/gcc/testsuite/gcc.target/arm/simd/vst1_p64_xN_1.c
> @@ -15,4 +15,9 @@ void test_vst1_p64_x3 (poly64_t * ptr, poly64x1x3_t val)
> vst1_p64_x3 (ptr, val);
> }
>
> -/* { dg-final { scan-assembler-times {vst1.64\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+:64\]\n} 2 } } */
> \ No newline at end of file
> +void test_vst1_p64_x4 (poly64_t * ptr, poly64x1x4_t val)
> +{
> + vst1_p64_x4 (ptr, val);
> +}
> +
> +/* { dg-final { scan-assembler-times {vst1.64\t\{d[0-9]+-d[0-9]+\}, \[r[0-9]+:64\]\n} 3 } } */
> \ No newline at end of file
^ permalink raw reply [flat|nested] 8+ messages in thread
end of thread, other threads:[~2023-11-27 14:59 UTC | newest]
Thread overview: 8+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2023-10-06 11:55 [PATCH 0/3] [GCC] arm: vst1_types_xN ACLE intrinsics Ezra.Sitorus
2023-10-06 11:55 ` [PATCH 1/3] [GCC] arm: vst1_types_x2 " Ezra.Sitorus
2023-11-27 14:56 ` Richard Earnshaw
2023-11-27 14:57 ` Richard Earnshaw
2023-10-06 11:55 ` [PATCH 2/3] [GCC] arm: vst1_types_x3 " Ezra.Sitorus
2023-11-27 14:58 ` Richard Earnshaw
2023-10-06 11:56 ` [PATCH 3/3] [GCC] arm: vst1_types_x4 " Ezra.Sitorus
2023-11-27 14:59 ` 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).