@@ -1898,6 +1898,14 @@ typedef struct { uint8x16_t val[4]; } uint8x16x4_t;
#define vstrwq_scatter_shifted_offset_p_u32(__base, __offset, __value, __p) __arm_vstrwq_scatter_shifted_offset_p_u32(__base, __offset, __value, __p)
#define vstrwq_scatter_shifted_offset_s32(__base, __offset, __value) __arm_vstrwq_scatter_shifted_offset_s32(__base, __offset, __value)
#define vstrwq_scatter_shifted_offset_u32(__base, __offset, __value) __arm_vstrwq_scatter_shifted_offset_u32(__base, __offset, __value)
+#define vaddq_s8(__a, __b) __arm_vaddq_s8(__a, __b)
+#define vaddq_s16(__a, __b) __arm_vaddq_s16(__a, __b)
+#define vaddq_s32(__a, __b) __arm_vaddq_s32(__a, __b)
+#define vaddq_u8(__a, __b) __arm_vaddq_u8(__a, __b)
+#define vaddq_u16(__a, __b) __arm_vaddq_u16(__a, __b)
+#define vaddq_u32(__a, __b) __arm_vaddq_u32(__a, __b)
+#define vaddq_f16(__a, __b) __arm_vaddq_f16(__a, __b)
+#define vaddq_f32(__a, __b) __arm_vaddq_f32(__a, __b)
#endif
__extension__ extern __inline void
@@ -12341,6 +12349,48 @@ __arm_vstrwq_scatter_shifted_offset_u32 (uint32_t * __base, uint32x4_t __offset,
__builtin_mve_vstrwq_scatter_shifted_offset_uv4si ((__builtin_neon_si *) __base, __offset, __value);
}
+__extension__ extern __inline int8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+__arm_vaddq_s8 (int8x16_t __a, int8x16_t __b)
+{
+ return __a + __b;
+}
+
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+__arm_vaddq_s16 (int16x8_t __a, int16x8_t __b)
+{
+ return __a + __b;
+}
+
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+__arm_vaddq_s32 (int32x4_t __a, int32x4_t __b)
+{
+ return __a + __b;
+}
+
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+__arm_vaddq_u8 (uint8x16_t __a, uint8x16_t __b)
+{
+ return __a + __b;
+}
+
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+__arm_vaddq_u16 (uint16x8_t __a, uint16x8_t __b)
+{
+ return __a + __b;
+}
+
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+__arm_vaddq_u32 (uint32x4_t __a, uint32x4_t __b)
+{
+ return __a + __b;
+}
+
#if (__ARM_FEATURE_MVE & 2) /* MVE Floating point. */
__extension__ extern __inline void
@@ -14707,6 +14757,20 @@ __arm_vstrwq_scatter_shifted_offset_p_f32 (float32_t * __base, uint32x4_t __offs
__builtin_mve_vstrwq_scatter_shifted_offset_p_fv4sf (__base, __offset, __value, __p);
}
+__extension__ extern __inline float16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+__arm_vaddq_f16 (float16x8_t __a, float16x8_t __b)
+{
+ return __a + __b;
+}
+
+__extension__ extern __inline float32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+__arm_vaddq_f32 (float32x4_t __a, float32x4_t __b)
+{
+ return __a + __b;
+}
+
#endif
enum {
@@ -16557,6 +16621,25 @@ extern void *__ARM_undef;
int (*)[__ARM_mve_type_uint32_t_ptr][__ARM_mve_type_uint32x4_t]: __arm_vstrwq_scatter_shifted_offset_u32 (__ARM_mve_coerce(__p0, uint32_t *), __p1, __ARM_mve_coerce(__p2, uint32x4_t)), \
int (*)[__ARM_mve_type_float32_t_ptr][__ARM_mve_type_float32x4_t]: __arm_vstrwq_scatter_shifted_offset_f32 (__ARM_mve_coerce(__p0, float32_t *), __p1, __ARM_mve_coerce(__p2, float32x4_t)));})
+#define vaddq(p0,p1) __arm_vaddq(p0,p1)
+#define __arm_vaddq(p0,p1) ({ __typeof(p0) __p0 = (p0); \
+ __typeof(p1) __p1 = (p1); \
+ _Generic( (int (*)[__ARM_mve_typeid(__p0)][__ARM_mve_typeid(__p1)])0, \
+ int (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int8x16_t]: __arm_vaddq_s8 (__ARM_mve_coerce(__p0, int8x16_t), __ARM_mve_coerce(__p1, int8x16_t)), \
+ int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int16x8_t]: __arm_vaddq_s16 (__ARM_mve_coerce(__p0, int16x8_t), __ARM_mve_coerce(__p1, int16x8_t)), \
+ int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int32x4_t]: __arm_vaddq_s32 (__ARM_mve_coerce(__p0, int32x4_t), __ARM_mve_coerce(__p1, int32x4_t)), \
+ int (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_uint8x16_t]: __arm_vaddq_u8 (__ARM_mve_coerce(__p0, uint8x16_t), __ARM_mve_coerce(__p1, uint8x16_t)), \
+ int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint16x8_t]: __arm_vaddq_u16 (__ARM_mve_coerce(__p0, uint16x8_t), __ARM_mve_coerce(__p1, uint16x8_t)), \
+ int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint32x4_t]: __arm_vaddq_u32 (__ARM_mve_coerce(__p0, uint32x4_t), __ARM_mve_coerce(__p1, uint32x4_t)), \
+ int (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_uint8_t]: __arm_vaddq_n_u8 (__ARM_mve_coerce(__p0, uint8x16_t), __ARM_mve_coerce(__p1, uint8_t)), \
+ int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint16_t]: __arm_vaddq_n_u16 (__ARM_mve_coerce(__p0, uint16x8_t), __ARM_mve_coerce(__p1, uint16_t)), \
+ int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint32_t]: __arm_vaddq_n_u32 (__ARM_mve_coerce(__p0, uint32x4_t), __ARM_mve_coerce(__p1, uint32_t)), \
+ int (*)[__ARM_mve_type_int8x16_t][__ARM_mve_type_int8_t]: __arm_vaddq_n_s8 (__ARM_mve_coerce(__p0, int8x16_t), __ARM_mve_coerce(__p1, int8_t)), \
+ int (*)[__ARM_mve_type_int16x8_t][__ARM_mve_type_int16_t]: __arm_vaddq_n_s16 (__ARM_mve_coerce(__p0, int16x8_t), __ARM_mve_coerce(__p1, int16_t)), \
+ int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int32_t]: __arm_vaddq_n_s32 (__ARM_mve_coerce(__p0, int32x4_t), __ARM_mve_coerce(__p1, int32_t)), \
+ int (*)[__ARM_mve_type_float16x8_t][__ARM_mve_type_float16x8_t]: __arm_vaddq_f16 (__ARM_mve_coerce(p0, float16x8_t), __ARM_mve_coerce(p1, float16x8_t)), \
+ int (*)[__ARM_mve_type_float32x4_t][__ARM_mve_type_float32x4_t]: __arm_vaddq_f32 (__ARM_mve_coerce(p0, float32x4_t), __ARM_mve_coerce(p1, float32x4_t)));})
+
#else /* MVE Interger. */
#define vst4q(p0,p1) __arm_vst4q(p0,p1)
@@ -17213,12 +17296,7 @@ extern void *__ARM_undef;
int (*)[__ARM_mve_type_int32x4_t][__ARM_mve_type_int32x4_t]: __arm_vaddq_s32 (__ARM_mve_coerce(__p0, int32x4_t), __ARM_mve_coerce(__p1, int32x4_t)), \
int (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_uint8x16_t]: __arm_vaddq_u8 (__ARM_mve_coerce(__p0, uint8x16_t), __ARM_mve_coerce(__p1, uint8x16_t)), \
int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint16x8_t]: __arm_vaddq_u16 (__ARM_mve_coerce(__p0, uint16x8_t), __ARM_mve_coerce(__p1, uint16x8_t)), \
- int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint32x4_t]: __arm_vaddq_u32 (__ARM_mve_coerce(__p0, uint32x4_t), __ARM_mve_coerce(__p1, uint32x4_t)));})
-
-#define vaddq(p0,p1) __arm_vaddq(p0,p1)
-#define __arm_vaddq(p0,p1) ({ __typeof(p0) __p0 = (p0); \
- __typeof(p1) __p1 = (p1); \
- _Generic( (int (*)[__ARM_mve_typeid(__p0)][__ARM_mve_typeid(__p1)])0, \
+ int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint32x4_t]: __arm_vaddq_u32 (__ARM_mve_coerce(__p0, uint32x4_t), __ARM_mve_coerce(__p1, uint32x4_t)), \
int (*)[__ARM_mve_type_uint8x16_t][__ARM_mve_type_uint8_t]: __arm_vaddq_n_u8 (__ARM_mve_coerce(__p0, uint8x16_t), __ARM_mve_coerce(__p1, uint8_t)), \
int (*)[__ARM_mve_type_uint16x8_t][__ARM_mve_type_uint16_t]: __arm_vaddq_n_u16 (__ARM_mve_coerce(__p0, uint16x8_t), __ARM_mve_coerce(__p1, uint16_t)), \
int (*)[__ARM_mve_type_uint32x4_t][__ARM_mve_type_uint32_t]: __arm_vaddq_n_u32 (__ARM_mve_coerce(__p0, uint32x4_t), __ARM_mve_coerce(__p1, uint32_t)), \
@@ -62,6 +62,14 @@
;; Integer and float modes supported by Neon and IWMMXT.
(define_mode_iterator VALL [V2DI V2SI V4HI V8QI V2SF V4SI V8HI V16QI V4SF])
+;; Integer and float modes supported by Neon, IWMMXT and MVE, used by
+;; arithmetic epxand patterns.
+(define_mode_iterator VNIM [V16QI V8HI V4SI V4SF])
+
+;; Integer and float modes supported by Neon and IWMMXT but not MVE, used by
+;; arithmetic epxand patterns.
+(define_mode_iterator VNINOTM [V2SI V4HI V8QI V2SF V2DI])
+
;; Integer and float modes supported by Neon, IWMMXT and MVE.
(define_mode_iterator VNIM1 [V16QI V8HI V4SI V4SF V2DI])
@@ -9636,3 +9636,31 @@
return "";
}
[(set_attr "length" "4")])
+
+;;
+;; [vaddq_s, vaddq_u])
+;;
+(define_insn "mve_vaddq<mode>"
+ [
+ (set (match_operand:MVE_2 0 "s_register_operand" "=w")
+ (plus:MVE_2 (match_operand:MVE_2 1 "s_register_operand" "w")
+ (match_operand:MVE_2 2 "s_register_operand" "w")))
+ ]
+ "TARGET_HAVE_MVE"
+ "vadd.i%#<V_sz_elem> %q0, %q1, %q2"
+ [(set_attr "type" "mve_move")
+])
+
+;;
+;; [vaddq_f])
+;;
+(define_insn "mve_vaddq_f<mode>"
+ [
+ (set (match_operand:MVE_0 0 "s_register_operand" "=w")
+ (plus:MVE_0 (match_operand:MVE_0 1 "s_register_operand" "w")
+ (match_operand:MVE_0 2 "s_register_operand" "w")))
+ ]
+ "TARGET_HAVE_MVE && TARGET_HAVE_MVE_FLOAT"
+ "vadd.f%#<V_sz_elem> %q0, %q1, %q2"
+ [(set_attr "type" "mve_move")
+])
@@ -519,18 +519,30 @@
;; As with SFmode, full support for HFmode vector arithmetic is only available
;; when flag-unsafe-math-optimizations is enabled.
-(define_insn "add<mode>3"
+;; Add pattern with modes V8HF and V4HF is split into separate patterns to add
+;; support for standard pattern addv8hf3 in MVE. Following pattern is called
+;; from "addv8hf3" standard pattern inside vec-common.md file.
+
+(define_insn "addv8hf3_neon"
[(set
- (match_operand:VH 0 "s_register_operand" "=w")
- (plus:VH
- (match_operand:VH 1 "s_register_operand" "w")
- (match_operand:VH 2 "s_register_operand" "w")))]
+ (match_operand:V8HF 0 "s_register_operand" "=w")
+ (plus:V8HF
+ (match_operand:V8HF 1 "s_register_operand" "w")
+ (match_operand:V8HF 2 "s_register_operand" "w")))]
"TARGET_NEON_FP16INST && flag_unsafe_math_optimizations"
- "vadd.<V_if_elem>\t%<V_reg>0, %<V_reg>1, %<V_reg>2"
- [(set (attr "type")
- (if_then_else (match_test "<Is_float_mode>")
- (const_string "neon_fp_addsub_s<q>")
- (const_string "neon_add<q>")))]
+ "vadd.f16\t%<V_reg>0, %<V_reg>1, %<V_reg>2"
+ [(set_attr "type" "neon_fp_addsub_s_q")]
+)
+
+(define_insn "addv4hf3"
+ [(set
+ (match_operand:V4HF 0 "s_register_operand" "=w")
+ (plus:V4HF
+ (match_operand:V4HF 1 "s_register_operand" "w")
+ (match_operand:V4HF 2 "s_register_operand" "w")))]
+ "TARGET_NEON_FP16INST && flag_unsafe_math_optimizations"
+ "vadd.f16\t%<V_reg>0, %<V_reg>1, %<V_reg>2"
+ [(set_attr "type" "neon_fp_addsub_s_q")]
)
(define_insn "add<mode>3_fp16"
@@ -77,19 +77,51 @@
}
})
+;; Vector arithmetic. Expanders are blank, then unnamed insns implement
+;; patterns separately for Neon, IWMMXT and MVE.
+
+(define_expand "add<mode>3"
+ [(set (match_operand:VNIM 0 "s_register_operand")
+ (plus:VNIM (match_operand:VNIM 1 "s_register_operand")
+ (match_operand:VNIM 2 "s_register_operand")))]
+ "(TARGET_NEON && ((<MODE>mode != V2SFmode && <MODE>mode != V4SFmode)
+ || flag_unsafe_math_optimizations))
+ || (TARGET_REALLY_IWMMXT && VALID_IWMMXT_REG_MODE (<MODE>mode))
+ || (TARGET_HAVE_MVE && VALID_MVE_SI_MODE(<MODE>mode))
+ || (TARGET_HAVE_MVE_FLOAT && VALID_MVE_SF_MODE(<MODE>mode))"
+{
+})
+
+;; Vector arithmetic. Expanders are blank, then unnamed insns implement
+;; patterns separately for Neon and MVE.
+
+(define_expand "addv8hf3"
+ [(set (match_operand:V8HF 0 "s_register_operand")
+ (plus:V8HF (match_operand:V8HF 1 "s_register_operand")
+ (match_operand:V8HF 2 "s_register_operand")))]
+ "(TARGET_HAVE_MVE_FLOAT && VALID_MVE_SF_MODE(V8HFmode))
+ || (TARGET_NEON_FP16INST && flag_unsafe_math_optimizations)"
+{
+ if (TARGET_NEON_FP16INST && flag_unsafe_math_optimizations)
+ emit_insn (gen_addv8hf3_neon (operands[0], operands[1], operands[2]));
+})
+
+;; Vector arithmetic. Expanders are blank, then unnamed insns implement
+;; patterns separately for Neon and IWMMXT.
+
+(define_expand "add<mode>3"
+ [(set (match_operand:VNINOTM 0 "s_register_operand")
+ (plus:VNINOTM (match_operand:VNINOTM 1 "s_register_operand")
+ (match_operand:VNINOTM 2 "s_register_operand")))]
+ "(TARGET_NEON && ((<MODE>mode != V2SFmode && <MODE>mode != V4SFmode)
+ || flag_unsafe_math_optimizations))
+ || (TARGET_REALLY_IWMMXT && VALID_IWMMXT_REG_MODE (<MODE>mode))"
+{
+})
+
;; Vector arithmetic. Expanders are blank, then unnamed insns implement
;; patterns separately for IWMMXT and Neon.
-(define_expand "add<mode>3"
- [(set (match_operand:VALL 0 "s_register_operand")
- (plus:VALL (match_operand:VALL 1 "s_register_operand")
- (match_operand:VALL 2 "s_register_operand")))]
- "(TARGET_NEON && ((<MODE>mode != V2SFmode && <MODE>mode != V4SFmode)
- || flag_unsafe_math_optimizations))
- || (TARGET_REALLY_IWMMXT && VALID_IWMMXT_REG_MODE (<MODE>mode))"
-{
-})
-
(define_expand "sub<mode>3"
[(set (match_operand:VALL 0 "s_register_operand")
(minus:VALL (match_operand:VALL 1 "s_register_operand")
new file mode 100644
@@ -0,0 +1,21 @@
+/* { dg-do compile } */
+/* { dg-additional-options "-march=armv8.1-m.main+mve.fp -mfloat-abi=hard -O2" } */
+/* { dg-skip-if "Skip if not auto" {*-*-*} {"-mfpu=*"} {"-mfpu=auto"} } */
+
+#include "arm_mve.h"
+
+float16x8_t
+foo (float16x8_t a, float16x8_t b)
+{
+ return vaddq_f16 (a, b);
+}
+
+/* { dg-final { scan-assembler "vadd.f16" } } */
+
+float16x8_t
+foo1 (float16x8_t a, float16x8_t b)
+{
+ return vaddq (a, b);
+}
+
+/* { dg-final { scan-assembler "vadd.f16" } } */
new file mode 100644
@@ -0,0 +1,21 @@
+/* { dg-do compile } */
+/* { dg-additional-options "-march=armv8.1-m.main+mve.fp -mfloat-abi=hard -O2" } */
+/* { dg-skip-if "Skip if not auto" {*-*-*} {"-mfpu=*"} {"-mfpu=auto"} } */
+
+#include "arm_mve.h"
+
+float32x4_t
+foo (float32x4_t a, float32x4_t b)
+{
+ return vaddq_f32 (a, b);
+}
+
+/* { dg-final { scan-assembler "vadd.f32" } } */
+
+float32x4_t
+foo1 (float32x4_t a, float32x4_t b)
+{
+ return vaddq (a, b);
+}
+
+/* { dg-final { scan-assembler "vadd.f32" } } */
new file mode 100644
@@ -0,0 +1,21 @@
+/* { dg-do compile } */
+/* { dg-additional-options "-march=armv8.1-m.main+mve -mfloat-abi=hard -O2" } */
+/* { dg-skip-if "Skip if not auto" {*-*-*} {"-mfpu=*"} {"-mfpu=auto"} } */
+
+#include "arm_mve.h"
+
+int16x8_t
+foo (int16x8_t a, int16x8_t b)
+{
+ return vaddq_s16 (a, b);
+}
+
+/* { dg-final { scan-assembler "vadd.i16" } } */
+
+int16x8_t
+foo1 (int16x8_t a, int16x8_t b)
+{
+ return vaddq (a, b);
+}
+
+/* { dg-final { scan-assembler "vadd.i16" } } */
new file mode 100644
@@ -0,0 +1,21 @@
+/* { dg-do compile } */
+/* { dg-additional-options "-march=armv8.1-m.main+mve -mfloat-abi=hard -O2" } */
+/* { dg-skip-if "Skip if not auto" {*-*-*} {"-mfpu=*"} {"-mfpu=auto"} } */
+
+#include "arm_mve.h"
+
+int32x4_t
+foo (int32x4_t a, int32x4_t b)
+{
+ return vaddq_s32 (a, b);
+}
+
+/* { dg-final { scan-assembler "vadd.i32" } } */
+
+int32x4_t
+foo1 (int32x4_t a, int32x4_t b)
+{
+ return vaddq (a, b);
+}
+
+/* { dg-final { scan-assembler "vadd.i32" } } */
new file mode 100644
@@ -0,0 +1,21 @@
+/* { dg-do compile } */
+/* { dg-additional-options "-march=armv8.1-m.main+mve -mfloat-abi=hard -O2" } */
+/* { dg-skip-if "Skip if not auto" {*-*-*} {"-mfpu=*"} {"-mfpu=auto"} } */
+
+#include "arm_mve.h"
+
+int8x16_t
+foo (int8x16_t a, int8x16_t b)
+{
+ return vaddq_s8 (a, b);
+}
+
+/* { dg-final { scan-assembler "vadd.i8" } } */
+
+int8x16_t
+foo1 (int8x16_t a, int8x16_t b)
+{
+ return vaddq (a, b);
+}
+
+/* { dg-final { scan-assembler "vadd.i8" } } */
new file mode 100644
@@ -0,0 +1,21 @@
+/* { dg-do compile } */
+/* { dg-additional-options "-march=armv8.1-m.main+mve -mfloat-abi=hard -O2" } */
+/* { dg-skip-if "Skip if not auto" {*-*-*} {"-mfpu=*"} {"-mfpu=auto"} } */
+
+#include "arm_mve.h"
+
+uint16x8_t
+foo (uint16x8_t a, uint16x8_t b)
+{
+ return vaddq_u16 (a, b);
+}
+
+/* { dg-final { scan-assembler "vadd.i16" } } */
+
+uint16x8_t
+foo1 (uint16x8_t a, uint16x8_t b)
+{
+ return vaddq (a, b);
+}
+
+/* { dg-final { scan-assembler "vadd.i16" } } */
new file mode 100644
@@ -0,0 +1,21 @@
+/* { dg-do compile } */
+/* { dg-additional-options "-march=armv8.1-m.main+mve -mfloat-abi=hard -O2" } */
+/* { dg-skip-if "Skip if not auto" {*-*-*} {"-mfpu=*"} {"-mfpu=auto"} } */
+
+#include "arm_mve.h"
+
+uint32x4_t
+foo (uint32x4_t a, uint32x4_t b)
+{
+ return vaddq_u32 (a, b);
+}
+
+/* { dg-final { scan-assembler "vadd.i32" } } */
+
+uint32x4_t
+foo1 (uint32x4_t a, uint32x4_t b)
+{
+ return vaddq (a, b);
+}
+
+/* { dg-final { scan-assembler "vadd.i32" } } */
new file mode 100644
@@ -0,0 +1,21 @@
+/* { dg-do compile } */
+/* { dg-additional-options "-march=armv8.1-m.main+mve -mfloat-abi=hard -O2" } */
+/* { dg-skip-if "Skip if not auto" {*-*-*} {"-mfpu=*"} {"-mfpu=auto"} } */
+
+#include "arm_mve.h"
+
+uint8x16_t
+foo (uint8x16_t a, uint8x16_t b)
+{
+ return vaddq_u8 (a, b);
+}
+
+/* { dg-final { scan-assembler "vadd.i8" } } */
+
+uint8x16_t
+foo1 (uint8x16_t a, uint8x16_t b)
+{
+ return vaddq (a, b);
+}
+
+/* { dg-final { scan-assembler "vadd.i8" } } */