@@ -356,6 +356,323 @@ _mm512_maskz_fnmsubne_pbh (__mmask32 __U, __m512bh __A,
__builtin_ia32_fnmsubnepbf16512_maskz (__A, __B, __C, __U);
}
+extern __inline__ __m512bh
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm512_rsqrt_pbh (__m512bh __A)
+{
+ return (__m512bh)
+ __builtin_ia32_rsqrtpbf16512_mask (__A,
+ (__v32bf) _mm512_setzero_si512 (),
+ (__mmask32) -1);
+
+}
+
+extern __inline__ __m512bh
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm512_mask_rsqrt_pbh (__m512bh __W, __mmask32 __U, __m512bh __A)
+{
+ return (__m512bh)
+ __builtin_ia32_rsqrtpbf16512_mask (__A, __W, __U);
+}
+
+extern __inline__ __m512bh
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm512_maskz_rsqrt_pbh (__mmask32 __U, __m512bh __A)
+{
+ return (__m512bh)
+ __builtin_ia32_rsqrtpbf16512_mask (__A,
+ (__v32bf) _mm512_setzero_si512 (),
+ __U);
+}
+
+extern __inline__ __m512bh
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm512_sqrtne_pbh (__m512bh __A)
+{
+ return (__m512bh)
+ __builtin_ia32_sqrtnepbf16512_mask (__A,
+ (__v32bf) _mm512_setzero_si512 (),
+ (__mmask32) -1);
+}
+
+extern __inline__ __m512bh
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm512_mask_sqrtne_pbh (__m512bh __W, __mmask32 __U, __m512bh __A)
+{
+ return (__m512bh)
+ __builtin_ia32_sqrtnepbf16512_mask (__A, __W, __U);
+}
+
+extern __inline__ __m512bh
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm512_maskz_sqrtne_pbh (__mmask32 __U, __m512bh __A)
+{
+ return (__m512bh)
+ __builtin_ia32_sqrtnepbf16512_mask (__A,
+ (__v32bf) _mm512_setzero_si512 (),
+ __U);
+}
+
+extern __inline__ __m512bh
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm512_rcp_pbh (__m512bh __A)
+{
+ return (__m512bh)
+ __builtin_ia32_rcppbf16512_mask (__A,
+ (__v32bf) _mm512_setzero_si512 (),
+ (__mmask32) -1);
+}
+
+extern __inline__ __m512bh
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm512_mask_rcp_pbh (__m512bh __W, __mmask32 __U, __m512bh __A)
+{
+ return (__m512bh)
+ __builtin_ia32_rcppbf16512_mask (__A, __W, __U);
+}
+
+extern __inline__ __m512bh
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm512_maskz_rcp_pbh (__mmask32 __U, __m512bh __A)
+{
+ return (__m512bh)
+ __builtin_ia32_rcppbf16512_mask (__A,
+ (__v32bf) _mm512_setzero_si512 (),
+ __U);
+}
+
+extern __inline__ __m512bh
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm512_getexp_pbh (__m512bh __A)
+{
+ return (__m512bh)
+ __builtin_ia32_getexppbf16512_mask (__A,
+ (__v32bf) _mm512_setzero_si512 (),
+ (__mmask32) -1);
+}
+
+extern __inline__ __m512bh
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm512_mask_getexp_pbh (__m512bh __W, __mmask32 __U, __m512bh __A)
+{
+ return (__m512bh) __builtin_ia32_getexppbf16512_mask (__A, __W, __U);
+}
+
+extern __inline__ __m512bh
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm512_maskz_getexp_pbh (__mmask32 __U, __m512bh __A)
+{
+ return (__m512bh)
+ __builtin_ia32_getexppbf16512_mask (__A,
+ (__v32bf) _mm512_setzero_si512 (),
+ __U);
+}
+
+/* Intrinsics vrndscalepbf16. */
+#ifdef __OPTIMIZE__
+extern __inline__ __m512bh
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm512_roundscalene_pbh (__m512bh __A, int B)
+{
+ return (__m512bh)
+ __builtin_ia32_rndscalenepbf16512_mask (__A, B,
+ (__v32bf) _mm512_setzero_si512 (),
+ (__mmask32) -1);
+}
+
+extern __inline__ __m512bh
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm512_mask_roundscalene_pbh (__m512bh __W, __mmask32 __U, __m512bh __A, int B)
+{
+ return (__m512bh)
+ __builtin_ia32_rndscalenepbf16512_mask (__A, B, __W, __U);
+}
+
+extern __inline__ __m512bh
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm512_maskz_roundscalene_pbh (__mmask32 __U, __m512bh __A, int B)
+{
+ return (__m512bh)
+ __builtin_ia32_rndscalenepbf16512_mask (__A, B,
+ (__v32bf) _mm512_setzero_si512 (),
+ __U);
+}
+
+#else
+#define _mm512_roundscalene_pbh(A, B) \
+ (__builtin_ia32_rndscalenepbf16512_mask ((A), (B), \
+ (__v32bf) _mm512_setzero_si512 (), \
+ (__mmask32) -1))
+
+#define _mm512_mask_roundscalene_pbh(A, B, C, D) \
+ (__builtin_ia32_rndscalenepbf16512_mask ((C), (D), (A), (B)))
+
+#define _mm512_maskz_roundscalene_pbh(A, B, C) \
+ (__builtin_ia32_rndscalenepbf16512_mask ((B), (C), \
+ (__v32bf) _mm512_setzero_si512 (), \
+ (A)))
+
+#endif /* __OPTIMIZE__ */
+
+/* Intrinsics vreducepbf16. */
+#ifdef __OPTIMIZE__
+extern __inline__ __m512bh
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm512_reducene_pbh (__m512bh __A, int B)
+{
+ return (__m512bh)
+ __builtin_ia32_reducenepbf16512_mask (__A, B,
+ (__v32bf) _mm512_setzero_si512 (),
+ (__mmask32) -1);
+}
+
+extern __inline__ __m512bh
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm512_mask_reducene_pbh (__m512bh __W, __mmask32 __U,
+ __m512bh __A, int B)
+{
+ return (__m512bh)
+ __builtin_ia32_reducenepbf16512_mask (__A, B, __W, __U);
+}
+
+extern __inline__ __m512bh
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm512_maskz_reducene_pbh (__mmask32 __U, __m512bh __A, int B)
+{
+ return (__m512bh)
+ __builtin_ia32_reducenepbf16512_mask (__A, B,
+ (__v32bf) _mm512_setzero_si512 (),
+ __U);
+}
+
+#else
+#define _mm512_reducene_pbh(A, B) \
+ (__builtin_ia32_reducenepbf16512_mask ((A), (B), \
+ (__v32bf) _mm512_setzero_si512 (), \
+ (__mmask32) -1))
+
+#define _mm512_mask_reducene_pbh(A, B, C, D) \
+ (__builtin_ia32_reducenepbf16512_mask ((C), (D), (A), (B)))
+
+#define _mm512_maskz_reducene_pbh(A, B, C) \
+ (__builtin_ia32_reducenepbf16512_mask ((B), (C), \
+ (__v32bf) _mm512_setzero_si512 (), \
+ (A)))
+
+#endif /* __OPTIMIZE__ */
+
+/* Intrinsics vgetmantpbf16. */
+#ifdef __OPTIMIZE__
+extern __inline__ __m512bh
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm512_getmant_pbh (__m512bh __A, _MM_MANTISSA_NORM_ENUM __B,
+ _MM_MANTISSA_SIGN_ENUM __C)
+{
+ return (__m512bh)
+ __builtin_ia32_getmantpbf16512_mask (__A, (int) (__C << 2) | __B,
+ (__v32bf) _mm512_setzero_si512 (),
+ (__mmask32) -1);
+}
+
+extern __inline__ __m512bh
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm512_mask_getmant_pbh (__m512bh __W, __mmask32 __U, __m512bh __A,
+ _MM_MANTISSA_NORM_ENUM __B,
+ _MM_MANTISSA_SIGN_ENUM __C)
+{
+ return (__m512bh)
+ __builtin_ia32_getmantpbf16512_mask (__A, (int) (__C << 2) | __B,
+ __W, __U);
+}
+
+extern __inline__ __m512bh
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm512_maskz_getmant_pbh (__mmask32 __U, __m512bh __A,
+ _MM_MANTISSA_NORM_ENUM __B,
+ _MM_MANTISSA_SIGN_ENUM __C)
+{
+ return (__m512bh)
+ __builtin_ia32_getmantpbf16512_mask (__A, (int) (__C << 2) | __B,
+ (__v32bf) _mm512_setzero_si512 (),
+ __U);
+}
+
+#else
+#define _mm512_getmant_pbh(A, B, C) \
+ (__builtin_ia32_getmantpbf16512_mask ((A), (int)(((C)<<2) | (B)), \
+ (__v32bf) _mm512_setzero_si512 (), \
+ (__mmask32) -1))
+
+#define _mm512_mask_getmant_pbh(A, B, C, D, E) \
+ (__builtin_ia32_getmantpbf16512_mask ((C), (int)(((D)<<2) | (E)), (A), (B)))
+
+#define _mm512_maskz_getmant_pbh(A, B, C, D) \
+ (__builtin_ia32_getmantpbf16512_mask ((B), (int)(((C)<<2) | (D)), \
+ (__v32bf) _mm512_setzero_si512 (), \
+ (A)))
+
+#endif /* __OPTIMIZE__ */
+
+/* Intrinsics vfpclasspbf16. */
+#ifdef __OPTIMIZE__
+extern __inline __mmask32
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm512_mask_fpclass_pbh_mask (__mmask32 __U, __m512bh __A,
+ const int __imm)
+{
+ return (__mmask32)
+ __builtin_ia32_fpclasspbf16512_mask (__A, __imm, __U);
+}
+
+extern __inline __mmask32
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm512_fpclass_pbh_mask (__m512bh __A, const int __imm)
+{
+ return (__mmask32)
+ __builtin_ia32_fpclasspbf16512_mask (__A, __imm,
+ (__mmask32) -1);
+}
+
+#else
+#define _mm512_mask_fpclass_pbh_mask(U, X, C) \
+ ((__mmask32) __builtin_ia32_fpclasspbf16512_mask ( \
+ (__v32bf) (__m512bh) (X), (int) (C), (__mmask32) (U)))
+
+#define _mm512_fpclass_pbh_mask(X, C) \
+ ((__mmask32) __builtin_ia32_fpclasspbf16512_mask ( \
+ (__v32bf) (__m512bh) (X), (int) (C), (__mmask32) (-1)))
+#endif /* __OPIMTIZE__ */
+
+
+/* Intrinsics vcmppbf16. */
+#ifdef __OPTIMIZE__
+extern __inline __mmask32
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm512_mask_cmp_pbh_mask (__mmask32 __U, __m512bh __A, __m512bh __B,
+ const int __imm)
+{
+ return (__mmask32)
+ __builtin_ia32_cmppbf16512_mask (__A, __B, __imm, __U);
+}
+
+extern __inline __mmask32
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm512_cmp_pbh_mask (__m512bh __A, __m512bh __B, const int __imm)
+{
+ return (__mmask32)
+ __builtin_ia32_cmppbf16512_mask (__A, __B, __imm,
+ (__mmask32) -1);
+}
+
+#else
+#define _mm512_mask_cmp_pbh_mask(A, B, C, D) \
+ ((__mmask32) __builtin_ia32_cmppbf16512_mask ((B), (C), (D), (A)))
+
+#define _mm512_cmp_pbh_mask(A, B, C) \
+ ((__mmask32) __builtin_ia32_cmppbf16512_mask ((A), (B), (C), (-1)))
+
+#endif /* __OPIMTIZE__ */
+
#ifdef __DISABLE_AVX10_2_512__
#undef __DISABLE_AVX10_2_512__
#pragma GCC pop_options
@@ -677,6 +677,656 @@ _mm_maskz_fnmsubne_pbh (__mmask8 __U, __m128bh __A,
__builtin_ia32_fnmsubnepbf16128_maskz (__A, __B, __C, __U);
}
+extern __inline__ __m256bh
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm256_rsqrt_pbh (__m256bh __A)
+{
+ return (__m256bh)
+ __builtin_ia32_rsqrtpbf16256_mask (__A,
+ (__v16bf) _mm256_setzero_si256 (),
+ (__mmask16) -1);
+}
+
+extern __inline__ __m256bh
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm256_mask_rsqrt_pbh (__m256bh __W, __mmask16 __U, __m256bh __A)
+{
+ return (__m256bh)
+ __builtin_ia32_rsqrtpbf16256_mask (__A, __W, __U);
+}
+
+extern __inline__ __m256bh
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm256_maskz_rsqrt_pbh (__mmask16 __U, __m256bh __A)
+{
+ return (__m256bh)
+ __builtin_ia32_rsqrtpbf16256_mask (__A,
+ (__v16bf) _mm256_setzero_si256 (),
+ __U);
+}
+
+extern __inline__ __m128bh
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm_rsqrt_pbh (__m128bh __A)
+{
+ return (__m128bh)
+ __builtin_ia32_rsqrtpbf16128_mask (__A,
+ (__v8bf) _mm_setzero_si128 (),
+ (__mmask8) -1);
+}
+
+extern __inline__ __m128bh
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm_mask_rsqrt_pbh (__m128bh __W, __mmask8 __U, __m128bh __A)
+{
+ return (__m128bh)
+ __builtin_ia32_rsqrtpbf16128_mask (__A, __W, __U);
+}
+
+extern __inline__ __m128bh
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm_maskz_rsqrt_pbh (__mmask8 __U, __m128bh __A)
+{
+ return (__m128bh)
+ __builtin_ia32_rsqrtpbf16128_mask (__A,
+ (__v8bf) _mm_setzero_si128 (),
+ __U);
+}
+
+extern __inline__ __m256bh
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm256_sqrtne_pbh (__m256bh __A)
+{
+ return (__m256bh)
+ __builtin_ia32_sqrtnepbf16256_mask (__A,
+ (__v16bf) _mm256_setzero_si256 (),
+ (__mmask16) -1);
+}
+
+extern __inline__ __m256bh
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm256_mask_sqrtne_pbh (__m256bh __W, __mmask16 __U, __m256bh __A)
+{
+ return (__m256bh)
+ __builtin_ia32_sqrtnepbf16256_mask (__A, __W, __U);
+}
+
+extern __inline__ __m256bh
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm256_maskz_sqrtne_pbh (__mmask16 __U, __m256bh __A)
+{
+ return (__m256bh)
+ __builtin_ia32_sqrtnepbf16256_mask (__A,
+ (__v16bf) _mm256_setzero_si256 (),
+ __U);
+}
+
+extern __inline__ __m128bh
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm_sqrtne_pbh (__m128bh __A)
+{
+ return (__m128bh)
+ __builtin_ia32_sqrtnepbf16128_mask (__A,
+ (__v8bf) _mm_setzero_si128 (),
+ (__mmask8) -1);
+}
+
+extern __inline__ __m128bh
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm_mask_sqrtne_pbh (__m128bh __W, __mmask8 __U, __m128bh __A)
+{
+ return (__m128bh)
+ __builtin_ia32_sqrtnepbf16128_mask (__A, __W, __U);
+}
+
+extern __inline__ __m128bh
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm_maskz_sqrtne_pbh (__mmask8 __U, __m128bh __A)
+{
+ return (__m128bh)
+ __builtin_ia32_sqrtnepbf16128_mask (__A,
+ (__v8bf) _mm_setzero_si128 (),
+ __U);
+}
+
+extern __inline__ __m256bh
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm256_rcp_pbh (__m256bh __A)
+{
+ return (__m256bh)
+ __builtin_ia32_rcppbf16256_mask (__A,
+ (__v16bf) _mm256_setzero_si256 (),
+ (__mmask16) -1);
+}
+
+extern __inline__ __m256bh
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm256_mask_rcp_pbh (__m256bh __W, __mmask16 __U, __m256bh __A)
+{
+ return (__m256bh)
+ __builtin_ia32_rcppbf16256_mask (__A, __W, __U);
+}
+
+extern __inline__ __m256bh
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm256_maskz_rcp_pbh (__mmask16 __U, __m256bh __A)
+{
+ return (__m256bh)
+ __builtin_ia32_rcppbf16256_mask (__A,
+ (__v16bf) _mm256_setzero_si256 (),
+ __U);
+}
+
+extern __inline__ __m128bh
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm_rcp_pbh (__m128bh __A)
+{
+ return (__m128bh)
+ __builtin_ia32_rcppbf16128_mask (__A,
+ (__v8bf) _mm_setzero_si128 (),
+ (__mmask8) -1);
+}
+
+extern __inline__ __m128bh
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm_mask_rcp_pbh (__m128bh __W, __mmask8 __U, __m128bh __A)
+{
+ return (__m128bh)
+ __builtin_ia32_rcppbf16128_mask (__A, __W, __U);
+}
+
+extern __inline__ __m128bh
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm_maskz_rcp_pbh (__mmask8 __U, __m128bh __A)
+{
+ return (__m128bh)
+ __builtin_ia32_rcppbf16128_mask (__A,
+ (__v8bf) _mm_setzero_si128 (),
+ __U);
+}
+
+extern __inline__ __m256bh
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm256_getexp_pbh (__m256bh __A)
+{
+ return (__m256bh)
+ __builtin_ia32_getexppbf16256_mask (__A,
+ (__v16bf) _mm256_setzero_si256 (),
+ (__mmask16) -1);
+}
+
+extern __inline__ __m256bh
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm256_mask_getexp_pbh (__m256bh __W, __mmask16 __U, __m256bh __A)
+{
+ return (__m256bh)
+ __builtin_ia32_getexppbf16256_mask (__A, __W, __U);
+}
+
+extern __inline__ __m256bh
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm256_maskz_getexp_pbh (__mmask16 __U, __m256bh __A)
+{
+ return (__m256bh)
+ __builtin_ia32_getexppbf16256_mask (__A,
+ (__v16bf) _mm256_setzero_si256 (),
+ __U);
+}
+
+extern __inline__ __m128bh
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm_getexp_pbh (__m128bh __A)
+{
+ return (__m128bh)
+ __builtin_ia32_getexppbf16128_mask (__A,
+ (__v8bf) _mm_setzero_si128 (),
+ (__mmask8) -1);
+}
+
+extern __inline__ __m128bh
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm_mask_getexp_pbh (__m128bh __W, __mmask8 __U, __m128bh __A)
+{
+ return (__m128bh)
+ __builtin_ia32_getexppbf16128_mask (__A, __W, __U);
+}
+
+extern __inline__ __m128bh
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm_maskz_getexp_pbh (__mmask8 __U, __m128bh __A)
+{
+ return (__m128bh)
+ __builtin_ia32_getexppbf16128_mask (__A,
+ (__v8bf) _mm_setzero_si128 (),
+ __U);
+}
+
+/* Intrinsics vrndscalepbf16. */
+#ifdef __OPTIMIZE__
+extern __inline__ __m256bh
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm256_roundscalene_pbh (__m256bh __A, int B)
+{
+ return (__m256bh)
+ __builtin_ia32_rndscalenepbf16256_mask (__A, B,
+ (__v16bf) _mm256_setzero_si256 (),
+ (__mmask16) -1);
+}
+
+extern __inline__ __m256bh
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm256_mask_roundscalene_pbh (__m256bh __W, __mmask16 __U,
+ __m256bh __A, int B)
+{
+ return (__m256bh)
+ __builtin_ia32_rndscalenepbf16256_mask (__A, B, __W, __U);
+}
+
+extern __inline__ __m256bh
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm256_maskz_roundscalene_pbh (__mmask16 __U, __m256bh __A, int B)
+{
+ return (__m256bh)
+ __builtin_ia32_rndscalenepbf16256_mask (__A, B,
+ (__v16bf) _mm256_setzero_si256 (),
+ __U);
+}
+
+extern __inline__ __m128bh
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm_roundscalene_pbh (__m128bh __A, int B)
+{
+ return (__m128bh)
+ __builtin_ia32_rndscalenepbf16128_mask (__A, B,
+ (__v8bf) _mm_setzero_si128 (),
+ (__mmask8) -1);
+}
+
+extern __inline__ __m128bh
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm_mask_roundscalene_pbh (__m128bh __W, __mmask8 __U,
+ __m128bh __A, int B)
+{
+ return (__m128bh)
+ __builtin_ia32_rndscalenepbf16128_mask (__A, B, __W, __U);
+}
+
+extern __inline__ __m128bh
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm_maskz_roundscalene_pbh (__mmask8 __U, __m128bh __A, int B)
+{
+ return (__m128bh)
+ __builtin_ia32_rndscalenepbf16128_mask (__A, B,
+ (__v8bf) _mm_setzero_si128 (),
+ __U);
+}
+
+#else
+#define _mm256_roundscalene_pbh(A, B) \
+ (__builtin_ia32_rndscalenepbf16256_mask ((A), (B), \
+ (__v16bf) _mm256_setzero_si256 (), \
+ (__mmask16) -1))
+
+#define _mm256_mask_roundscalene_pbh(A, B, C, D) \
+ (__builtin_ia32_rndscalenepbf16256_mask ((C), (D), (A), (B)))
+
+#define _mm256_maskz_roundscalene_pbh(A, B, C) \
+ (__builtin_ia32_rndscalenepbf16256_mask ((B), (C), \
+ (__v16bf) _mm256_setzero_si256 (), \
+ (A)))
+
+#define _mm_roundscalene_pbh(A, B) \
+ (__builtin_ia32_rndscalenepbf16128_mask ((A), (B), \
+ (__v8bf) _mm_setzero_si128 (), \
+ (__mmask8) -1))
+
+#define _mm_mask_roundscalene_pbh(A, B, C, D) \
+ (__builtin_ia32_rndscalenepbf16128_mask ((C), (D), (A), (B)))
+
+#define _mm_maskz_roundscalene_pbh(A, B, C) \
+ (__builtin_ia32_rndscalenepbf16128_mask ((B), (C), \
+ (__v8bf) _mm_setzero_si128 (), \
+ (A)))
+
+#endif /* __OPTIMIZE__ */
+
+/* Intrinsics vreducepbf16. */
+#ifdef __OPTIMIZE__
+extern __inline__ __m256bh
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm256_reducene_pbh (__m256bh __A, int B)
+{
+ return (__m256bh)
+ __builtin_ia32_reducenepbf16256_mask (__A, B,
+ (__v16bf) _mm256_setzero_si256 (),
+ (__mmask16) -1);
+}
+
+extern __inline__ __m256bh
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm256_mask_reducene_pbh (__m256bh __W, __mmask16 __U,
+ __m256bh __A, int B)
+{
+ return (__m256bh)
+ __builtin_ia32_reducenepbf16256_mask (__A, B, __W, __U);
+}
+
+extern __inline__ __m256bh
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm256_maskz_reducene_pbh (__mmask16 __U, __m256bh __A, int B)
+{
+ return (__m256bh)
+ __builtin_ia32_reducenepbf16256_mask (__A, B,
+ (__v16bf) _mm256_setzero_si256 (),
+ __U);
+}
+
+extern __inline__ __m128bh
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm_reducene_pbh (__m128bh __A, int B)
+{
+ return (__m128bh)
+ __builtin_ia32_reducenepbf16128_mask (__A, B,
+ (__v8bf) _mm_setzero_si128 (),
+ (__mmask8) -1);
+}
+
+extern __inline__ __m128bh
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm_mask_reducene_pbh (__m128bh __W, __mmask8 __U,
+ __m128bh __A, int B)
+{
+ return (__m128bh)
+ __builtin_ia32_reducenepbf16128_mask (__A, B, __W, __U);
+}
+
+extern __inline__ __m128bh
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm_maskz_reducene_pbh (__mmask8 __U, __m128bh __A, int B)
+{
+ return (__m128bh)
+ __builtin_ia32_reducenepbf16128_mask (__A, B,
+ (__v8bf) _mm_setzero_si128 (),
+ __U);
+}
+
+#else
+#define _mm256_reducene_pbh(A, B) \
+ (__builtin_ia32_reducenepbf16256_mask ((A), (B), \
+ (__v16bf) _mm256_setzero_si256 (), \
+ (__mmask16) -1))
+
+#define _mm256_mask_reducene_pbh(A, B, C, D) \
+ (__builtin_ia32_reducenepbf16256_mask ((C), (D), (A), (B)))
+
+#define _mm256_maskz_reducene_pbh(A, B, C) \
+ (__builtin_ia32_reducenepbf16256_mask ((B), (C), \
+ (__v16bf) _mm256_setzero_si256 (), \
+ (A)))
+
+#define _mm_reducene_pbh(A, B) \
+ (__builtin_ia32_reducenepbf16128_mask ((A), (B), \
+ (__v8bf) _mm_setzero_si128 (), \
+ (__mmask8) -1))
+
+#define _mm_mask_reducene_pbh(A, B, C, D) \
+ (__builtin_ia32_reducenepbf16128_mask ((C), (D), (A), (B)))
+
+#define _mm_maskz_reducene_pbh(A, B, C) \
+ (__builtin_ia32_reducenepbf16128_mask ((B), (C), \
+ (__v8bf) _mm_setzero_si128 (), \
+ (A)))
+
+#endif /* __OPTIMIZE__ */
+
+
+/* Intrinsics vgetmantpbf16. */
+#ifdef __OPTIMIZE__
+extern __inline__ __m256bh
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm256_getmant_pbh (__m256bh __A, _MM_MANTISSA_NORM_ENUM __B,
+ _MM_MANTISSA_SIGN_ENUM __C)
+{
+ return (__m256bh)
+ __builtin_ia32_getmantpbf16256_mask (__A, (int) (__C << 2) | __B,
+ (__v16bf) _mm256_setzero_si256 (),
+ (__mmask16) -1);
+}
+
+extern __inline__ __m256bh
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm256_mask_getmant_pbh (__m256bh __W, __mmask16 __U, __m256bh __A,
+ _MM_MANTISSA_NORM_ENUM __B,
+ _MM_MANTISSA_SIGN_ENUM __C)
+{
+ return (__m256bh)
+ __builtin_ia32_getmantpbf16256_mask (__A, (int) (__C << 2) | __B,
+ __W, __U);
+}
+
+extern __inline__ __m256bh
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm256_maskz_getmant_pbh (__mmask16 __U, __m256bh __A,
+ _MM_MANTISSA_NORM_ENUM __B,
+ _MM_MANTISSA_SIGN_ENUM __C)
+{
+ return (__m256bh)
+ __builtin_ia32_getmantpbf16256_mask (__A, (int) (__C << 2) | __B,
+ (__v16bf) _mm256_setzero_si256 (),
+ __U);
+}
+
+extern __inline__ __m128bh
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm_getmant_pbh (__m128bh __A, _MM_MANTISSA_NORM_ENUM __B,
+ _MM_MANTISSA_SIGN_ENUM __C)
+{
+ return (__m128bh)
+ __builtin_ia32_getmantpbf16128_mask (__A, (int) (__C << 2) | __B,
+ (__v8bf) _mm_setzero_si128 (),
+ (__mmask8) -1);
+}
+
+extern __inline__ __m128bh
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm_mask_getmant_pbh (__m128bh __W, __mmask8 __U, __m128bh __A,
+ _MM_MANTISSA_NORM_ENUM __B,
+ _MM_MANTISSA_SIGN_ENUM __C)
+{
+ return (__m128bh)
+ __builtin_ia32_getmantpbf16128_mask (__A, (int) (__C << 2) | __B,
+ __W, __U);
+}
+
+extern __inline__ __m128bh
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm_maskz_getmant_pbh (__mmask8 __U, __m128bh __A,
+ _MM_MANTISSA_NORM_ENUM __B,
+ _MM_MANTISSA_SIGN_ENUM __C)
+{
+ return (__m128bh)
+ __builtin_ia32_getmantpbf16128_mask (__A, (int) (__C << 2) | __B,
+ (__v8bf) _mm_setzero_si128 (),
+ __U);
+}
+
+#else
+#define _mm256_getmant_pbh(A, B, C) \
+ (__builtin_ia32_getmantpbf16256_mask ((A), (int)(((C)<<2) | (B)), \
+ (__v16bf) _mm256_setzero_si256 (), \
+ (__mmask16) (-1)))
+
+#define _mm256_mask_getmant_pbh(A, B, C, D, E) \
+ (__builtin_ia32_getmantpbf16256_mask ((C), (int)(((D)<<2) | (E)), (A), (B)))
+
+#define _mm256_maskz_getmant_pbh(A, B, C, D) \
+ (__builtin_ia32_getmantpbf16256_mask ((B), (int)(((C)<<2) | (D)), \
+ (__v16bf) _mm256_setzero_si256 (), \
+ (A)))
+
+#define _mm_getmant_pbh(A, B, C) \
+ (__builtin_ia32_getmantpbf16128_mask ((A), (int)(((C)<<2) | (B)), \
+ (__v8bf) _mm_setzero_si128 (), \
+ (__mmask8) (-1)))
+
+#define _mm_mask_getmant_pbh(A, B, C, D, E) \
+ (__builtin_ia32_getmantpbf16128_mask ((C), (int)(((D)<<2) | (E)), (A), (B)))
+
+#define _mm_maskz_getmant_pbh(A, B, C, D) \
+ (__builtin_ia32_getmantpbf16128_mask ((B), (int)(((C)<<2) | (D)), \
+ (__v8bf) _mm_setzero_si128 (), (A)))
+
+#endif /* __OPTIMIZE__ */
+
+/* Intrinsics vfpclasspbf16. */
+#ifdef __OPTIMIZE__
+extern __inline __mmask16
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm256_mask_fpclass_pbh_mask (__mmask16 __U, __m256bh __A,
+ const int __imm)
+{
+ return (__mmask16)
+ __builtin_ia32_fpclasspbf16256_mask (__A, __imm, __U);
+}
+
+extern __inline __mmask16
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm256_fpclass_pbh_mask (__m256bh __A, const int __imm)
+{
+ return (__mmask16)
+ __builtin_ia32_fpclasspbf16256_mask (__A, __imm, (__mmask16) -1);
+}
+
+extern __inline __mmask8
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm_mask_fpclass_pbh_mask (__mmask8 __U, __m128bh __A, const int __imm)
+{
+ return (__mmask8)
+ __builtin_ia32_fpclasspbf16128_mask (__A, __imm, __U);
+}
+
+extern __inline __mmask8
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm_fpclass_pbh_mask (__m128bh __A, const int __imm)
+{
+ return (__mmask8)
+ __builtin_ia32_fpclasspbf16128_mask (__A, __imm, (__mmask8) -1);
+}
+
+#else
+#define _mm256_mask_fpclass_pbh_mask(U, A, B) \
+ ((__mmask16) __builtin_ia32_fpclasspbf16256_mask ((A), (B), (U)))
+
+#define _mm256_fpclass_pbh_mask(A, B) \
+ ((__mmask16) __builtin_ia32_fpclasspbf16256_mask ((A), (B), \
+ (__mmask16) (-1)))
+
+#define _mm_mask_fpclass_pbh_mask(U, A, B) \
+ ((__mmask8) __builtin_ia32_fpclasspbf16128_mask ((A), (B), (U)))
+
+#define _mm_fpclass_pbh_mask(A, B) \
+ ((__mmask8) __builtin_ia32_fpclasspbf16128_mask ((A), (B), \
+ (__mmask8) (-1)))
+
+#endif /* __OPIMTIZE__ */
+
+
+/* Intrinsics vcmppbf16. */
+#ifdef __OPTIMIZE__
+extern __inline __mmask16
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm256_mask_cmp_pbh_mask (__mmask16 __U, __m256bh __A,
+ __m256bh __B, const int __imm)
+{
+ return (__mmask16)
+ __builtin_ia32_cmppbf16256_mask (__A, __B, __imm, __U);
+}
+
+extern __inline __mmask16
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm256_cmp_pbh_mask (__m256bh __A, __m256bh __B, const int __imm)
+{
+ return (__mmask16)
+ __builtin_ia32_cmppbf16256_mask (__A, __B, __imm, (__mmask16) -1);
+}
+
+extern __inline __mmask8
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm_mask_cmp_pbh_mask (__mmask8 __U, __m128bh __A,
+ __m128bh __B, const int __imm)
+{
+ return (__mmask8)
+ __builtin_ia32_cmppbf16128_mask (__A, __B, __imm, __U);
+}
+
+extern __inline __mmask8
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm_cmp_pbh_mask (__m128bh __A, __m128bh __B, const int __imm)
+{
+ return (__mmask8)
+ __builtin_ia32_cmppbf16128_mask (__A, __B, __imm, (__mmask8) -1);
+}
+
+#else
+#define _mm256_mask_cmp_pbh_mask(A, B, C, D) \
+ ((__mmask16) __builtin_ia32_cmppbf16256_mask ((B), (C), (D), (A)))
+
+#define _mm256_cmp_pbh_mask(A, B, C) \
+ ((__mmask16) __builtin_ia32_cmppbf16256_mask ((A), (B), (C), \
+ (__mmask16) (-1)))
+
+#define _mm_mask_cmp_pbh_mask(A, B, C, D) \
+ ((__mmask8) __builtin_ia32_cmppbf16128_mask ((B), (C), (D), (A)))
+
+#define _mm_cmp_pbh_mask(A, B, C) \
+ ((__mmask8) __builtin_ia32_cmppbf16128_mask ((A), (B), (C), \
+ (__mmask8) (-1)))
+
+#endif /* __OPIMTIZE__ */
+
+/* Intrinsics vcomsbf16. */
+extern __inline int
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm_comeq_sbh (__m128bh __A, __m128bh __B)
+{
+ return __builtin_ia32_vcomsbf16eq (__A, __B);
+}
+
+extern __inline int
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm_comlt_sbh (__m128bh __A, __m128bh __B)
+{
+ return __builtin_ia32_vcomsbf16lt (__A, __B);
+}
+
+extern __inline int
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm_comle_sbh (__m128bh __A, __m128bh __B)
+{
+ return __builtin_ia32_vcomsbf16le (__A, __B);
+}
+
+extern __inline int
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm_comgt_sbh (__m128bh __A, __m128bh __B)
+{
+ return __builtin_ia32_vcomsbf16gt (__A, __B);
+}
+
+extern __inline int
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm_comge_sbh (__m128bh __A, __m128bh __B)
+{
+ return __builtin_ia32_vcomsbf16ge (__A, __B);
+}
+
+extern __inline int
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm_comneq_sbh (__m128bh __A, __m128bh __B)
+{
+ return __builtin_ia32_vcomsbf16neq (__A, __B);
+}
+
#ifdef __DISABLE_AVX10_2_256__
#undef __DISABLE_AVX10_2_256__
#pragma GCC pop_options
@@ -1483,3 +1483,13 @@ DEF_FUNCTION_TYPE (V8BF, V8BF, V8BF, UQI)
DEF_FUNCTION_TYPE (V32BF, V32BF, V32BF, V32BF, USI)
DEF_FUNCTION_TYPE (V16BF, V16BF, V16BF, V16BF, UHI)
DEF_FUNCTION_TYPE (V8BF, V8BF, V8BF, V8BF, UQI)
+DEF_FUNCTION_TYPE (V32BF, V32BF, INT, V32BF, USI)
+DEF_FUNCTION_TYPE (V16BF, V16BF, INT, V16BF, UHI)
+DEF_FUNCTION_TYPE (V8BF, V8BF, INT, V8BF, UQI)
+DEF_FUNCTION_TYPE (QI, V8BF, INT, UQI)
+DEF_FUNCTION_TYPE (HI, V16BF, INT, UHI)
+DEF_FUNCTION_TYPE (SI, V32BF, INT, USI)
+DEF_FUNCTION_TYPE (USI, V32BF, V32BF, INT, USI)
+DEF_FUNCTION_TYPE (UHI, V16BF, V16BF, INT, UHI)
+DEF_FUNCTION_TYPE (UQI, V8BF, V8BF, INT, UQI)
+DEF_FUNCTION_TYPE (INT, V8BF, V8BF)
@@ -3237,6 +3237,39 @@ BDESC (0, OPTION_MASK_ISA2_AVX10_2_256, CODE_FOR_avx10_2_fnmsubnepbf16_v16bf_mas
BDESC (0, OPTION_MASK_ISA2_AVX10_2_256, CODE_FOR_avx10_2_fnmsubnepbf16_v8bf_mask, "__builtin_ia32_fnmsubnepbf16128_mask", IX86_BUILTIN_FNMSUBNEPBF16128_MASK, UNKNOWN, (int) V8BF_FTYPE_V8BF_V8BF_V8BF_UQI)
BDESC (0, OPTION_MASK_ISA2_AVX10_2_256, CODE_FOR_avx10_2_fnmsubnepbf16_v8bf_mask3, "__builtin_ia32_fnmsubnepbf16128_mask3", IX86_BUILTIN_FNMSUBNEPBF16128_MASK3, UNKNOWN, (int) V8BF_FTYPE_V8BF_V8BF_V8BF_UQI)
BDESC (0, OPTION_MASK_ISA2_AVX10_2_256, CODE_FOR_avx10_2_fnmsubnepbf16_v8bf_maskz, "__builtin_ia32_fnmsubnepbf16128_maskz", IX86_BUILTIN_FNMSUBNEPBF16128_MASKZ, UNKNOWN, (int) V8BF_FTYPE_V8BF_V8BF_V8BF_UQI)
+BDESC (0, OPTION_MASK_ISA2_AVX10_2_512, CODE_FOR_avx10_2_rsqrtpbf16_v32bf_mask, "__builtin_ia32_rsqrtpbf16512_mask", IX86_BUILTIN_RSQRTPBF16512_MASK, UNKNOWN, (int) V32BF_FTYPE_V32BF_V32BF_USI)
+BDESC (0, OPTION_MASK_ISA2_AVX10_2_256, CODE_FOR_avx10_2_rsqrtpbf16_v16bf_mask, "__builtin_ia32_rsqrtpbf16256_mask", IX86_BUILTIN_RSQRTPBF16256_MASK, UNKNOWN, (int) V16BF_FTYPE_V16BF_V16BF_UHI)
+BDESC (0, OPTION_MASK_ISA2_AVX10_2_256, CODE_FOR_avx10_2_rsqrtpbf16_v8bf_mask, "__builtin_ia32_rsqrtpbf16128_mask", IX86_BUILTIN_RSQRTPBF16128_MASK, UNKNOWN, (int) V8BF_FTYPE_V8BF_V8BF_UQI)
+BDESC (0, OPTION_MASK_ISA2_AVX10_2_512, CODE_FOR_avx10_2_sqrtnepbf16_v32bf_mask, "__builtin_ia32_sqrtnepbf16512_mask", IX86_BUILTIN_SQRTNEPBF16512_MASK, UNKNOWN, (int) V32BF_FTYPE_V32BF_V32BF_USI)
+BDESC (0, OPTION_MASK_ISA2_AVX10_2_256, CODE_FOR_avx10_2_sqrtnepbf16_v16bf_mask, "__builtin_ia32_sqrtnepbf16256_mask", IX86_BUILTIN_SQRTNEPBF16256_MASK, UNKNOWN, (int) V16BF_FTYPE_V16BF_V16BF_UHI)
+BDESC (0, OPTION_MASK_ISA2_AVX10_2_256, CODE_FOR_avx10_2_sqrtnepbf16_v8bf_mask, "__builtin_ia32_sqrtnepbf16128_mask", IX86_BUILTIN_SQRTNEPBF16128_MASK, UNKNOWN, (int) V8BF_FTYPE_V8BF_V8BF_UQI)
+BDESC (0, OPTION_MASK_ISA2_AVX10_2_512, CODE_FOR_avx10_2_rcppbf16_v32bf_mask, "__builtin_ia32_rcppbf16512_mask", IX86_BUILTIN_RCPPBF16512_MASK, UNKNOWN, (int) V32BF_FTYPE_V32BF_V32BF_USI)
+BDESC (0, OPTION_MASK_ISA2_AVX10_2_256, CODE_FOR_avx10_2_rcppbf16_v16bf_mask, "__builtin_ia32_rcppbf16256_mask", IX86_BUILTIN_RCPPBF16256_MASK, UNKNOWN, (int) V16BF_FTYPE_V16BF_V16BF_UHI)
+BDESC (0, OPTION_MASK_ISA2_AVX10_2_256, CODE_FOR_avx10_2_rcppbf16_v8bf_mask, "__builtin_ia32_rcppbf16128_mask", IX86_BUILTIN_RCPPBF16128_MASK, UNKNOWN, (int) V8BF_FTYPE_V8BF_V8BF_UQI)
+BDESC (0, OPTION_MASK_ISA2_AVX10_2_512, CODE_FOR_avx10_2_getexppbf16_v32bf_mask, "__builtin_ia32_getexppbf16512_mask", IX86_BUILTIN_GETEXPPBF16512_MASK, UNKNOWN, (int) V32BF_FTYPE_V32BF_V32BF_USI)
+BDESC (0, OPTION_MASK_ISA2_AVX10_2_256, CODE_FOR_avx10_2_getexppbf16_v16bf_mask, "__builtin_ia32_getexppbf16256_mask", IX86_BUILTIN_GETEXPPBF16256_MASK, UNKNOWN, (int) V16BF_FTYPE_V16BF_V16BF_UHI)
+BDESC (0, OPTION_MASK_ISA2_AVX10_2_256, CODE_FOR_avx10_2_getexppbf16_v8bf_mask, "__builtin_ia32_getexppbf16128_mask", IX86_BUILTIN_GETEXPPBF16128_MASK, UNKNOWN, (int) V8BF_FTYPE_V8BF_V8BF_UQI)
+BDESC (0, OPTION_MASK_ISA2_AVX10_2_512, CODE_FOR_avx10_2_rndscalenepbf16_v32bf_mask, "__builtin_ia32_rndscalenepbf16512_mask", IX86_BUILTIN_RNDSCALENEPBF16512_MASK, UNKNOWN, (int) V32BF_FTYPE_V32BF_INT_V32BF_USI)
+BDESC (0, OPTION_MASK_ISA2_AVX10_2_256, CODE_FOR_avx10_2_rndscalenepbf16_v16bf_mask, "__builtin_ia32_rndscalenepbf16256_mask", IX86_BUILTIN_RNDSCALENEPBF16256_MASK, UNKNOWN, (int) V16BF_FTYPE_V16BF_INT_V16BF_UHI)
+BDESC (0, OPTION_MASK_ISA2_AVX10_2_256, CODE_FOR_avx10_2_rndscalenepbf16_v8bf_mask, "__builtin_ia32_rndscalenepbf16128_mask", IX86_BUILTIN_RNDSCALENEPBF16128_MASK, UNKNOWN, (int) V8BF_FTYPE_V8BF_INT_V8BF_UQI)
+BDESC (0, OPTION_MASK_ISA2_AVX10_2_512, CODE_FOR_avx10_2_reducenepbf16_v32bf_mask, "__builtin_ia32_reducenepbf16512_mask", IX86_BUILTIN_REDUCENEPBF16512_MASK, UNKNOWN, (int) V32BF_FTYPE_V32BF_INT_V32BF_USI)
+BDESC (0, OPTION_MASK_ISA2_AVX10_2_256, CODE_FOR_avx10_2_reducenepbf16_v16bf_mask, "__builtin_ia32_reducenepbf16256_mask", IX86_BUILTIN_REDUCENEPBF16256_MASK, UNKNOWN, (int) V16BF_FTYPE_V16BF_INT_V16BF_UHI)
+BDESC (0, OPTION_MASK_ISA2_AVX10_2_256, CODE_FOR_avx10_2_reducenepbf16_v8bf_mask, "__builtin_ia32_reducenepbf16128_mask", IX86_BUILTIN_REDUCENEPBF16128_MASK, UNKNOWN, (int) V8BF_FTYPE_V8BF_INT_V8BF_UQI)
+BDESC (0, OPTION_MASK_ISA2_AVX10_2_512, CODE_FOR_avx10_2_getmantpbf16_v32bf_mask, "__builtin_ia32_getmantpbf16512_mask", IX86_BUILTIN_GETMANTPBF16512_MASK, UNKNOWN, (int) V32BF_FTYPE_V32BF_INT_V32BF_USI)
+BDESC (0, OPTION_MASK_ISA2_AVX10_2_256, CODE_FOR_avx10_2_getmantpbf16_v16bf_mask, "__builtin_ia32_getmantpbf16256_mask", IX86_BUILTIN_GETMANTPBF16256_MASK, UNKNOWN, (int) V16BF_FTYPE_V16BF_INT_V16BF_UHI)
+BDESC (0, OPTION_MASK_ISA2_AVX10_2_256, CODE_FOR_avx10_2_getmantpbf16_v8bf_mask, "__builtin_ia32_getmantpbf16128_mask", IX86_BUILTIN_GETMANTPBF16128_MASK, UNKNOWN, (int) V8BF_FTYPE_V8BF_INT_V8BF_UQI)
+BDESC (0, OPTION_MASK_ISA2_AVX10_2_512, CODE_FOR_avx10_2_fpclasspbf16_v32bf_mask, "__builtin_ia32_fpclasspbf16512_mask", IX86_BUILTIN_FPCLASSPBF16512_MASK, UNKNOWN, (int) SI_FTYPE_V32BF_INT_USI)
+BDESC (0, OPTION_MASK_ISA2_AVX10_2_256, CODE_FOR_avx10_2_fpclasspbf16_v16bf_mask, "__builtin_ia32_fpclasspbf16256_mask", IX86_BUILTIN_FPCLASSPBF16256_MASK, UNKNOWN, (int) HI_FTYPE_V16BF_INT_UHI)
+BDESC (0, OPTION_MASK_ISA2_AVX10_2_256, CODE_FOR_avx10_2_fpclasspbf16_v8bf_mask, "__builtin_ia32_fpclasspbf16128_mask", IX86_BUILTIN_FPCLASSPBF16128_MASK, UNKNOWN, (int) QI_FTYPE_V8BF_INT_UQI)
+BDESC (0, OPTION_MASK_ISA2_AVX10_2_512, CODE_FOR_avx10_2_cmppbf16_v32bf_mask, "__builtin_ia32_cmppbf16512_mask", IX86_BUILTIN_CMPPBF16512_MASK, UNKNOWN, (int) USI_FTYPE_V32BF_V32BF_INT_USI)
+BDESC (0, OPTION_MASK_ISA2_AVX10_2_256, CODE_FOR_avx10_2_cmppbf16_v16bf_mask, "__builtin_ia32_cmppbf16256_mask", IX86_BUILTIN_CMPPBF16256_MASK, UNKNOWN, (int) UHI_FTYPE_V16BF_V16BF_INT_UHI)
+BDESC (0, OPTION_MASK_ISA2_AVX10_2_256, CODE_FOR_avx10_2_cmppbf16_v8bf_mask, "__builtin_ia32_cmppbf16128_mask", IX86_BUILTIN_CMPPBF16128_MASK, UNKNOWN, (int) UQI_FTYPE_V8BF_V8BF_INT_UQI)
+BDESC (0, OPTION_MASK_ISA2_AVX10_2_256, CODE_FOR_avx10_2_comsbf16_v8bf, "__builtin_ia32_vcomsbf16eq", IX86_BUILTIN_VCOMSBF16EQ, EQ, (int) INT_FTYPE_V8BF_V8BF)
+BDESC (0, OPTION_MASK_ISA2_AVX10_2_256, CODE_FOR_avx10_2_comsbf16_v8bf, "__builtin_ia32_vcomsbf16gt", IX86_BUILTIN_VCOMSBF16GT, GT, (int) INT_FTYPE_V8BF_V8BF)
+BDESC (0, OPTION_MASK_ISA2_AVX10_2_256, CODE_FOR_avx10_2_comsbf16_v8bf, "__builtin_ia32_vcomsbf16ge", IX86_BUILTIN_VCOMSBF16GE, GE, (int) INT_FTYPE_V8BF_V8BF)
+BDESC (0, OPTION_MASK_ISA2_AVX10_2_256, CODE_FOR_avx10_2_comsbf16_v8bf, "__builtin_ia32_vcomsbf16le", IX86_BUILTIN_VCOMSBF16LE, LE, (int) INT_FTYPE_V8BF_V8BF)
+BDESC (0, OPTION_MASK_ISA2_AVX10_2_256, CODE_FOR_avx10_2_comsbf16_v8bf, "__builtin_ia32_vcomsbf16lt", IX86_BUILTIN_VCOMSBF16LT, LT, (int) INT_FTYPE_V8BF_V8BF)
+BDESC (0, OPTION_MASK_ISA2_AVX10_2_256, CODE_FOR_avx10_2_comsbf16_v8bf, "__builtin_ia32_vcomsbf16neq", IX86_BUILTIN_VCOMSBF16NE, NE, (int) INT_FTYPE_V8BF_V8BF)
/* Builtins with rounding support. */
BDESC_END (ARGS, ROUND_ARGS)
@@ -11712,6 +11712,9 @@ ix86_expand_args_builtin (const struct builtin_description *d,
case QI_FTYPE_V8HF_INT_UQI:
case HI_FTYPE_V16HF_INT_UHI:
case SI_FTYPE_V32HF_INT_USI:
+ case QI_FTYPE_V8BF_INT_UQI:
+ case HI_FTYPE_V16BF_INT_UHI:
+ case SI_FTYPE_V32BF_INT_USI:
case V4SI_FTYPE_V4SI_V4SI_UHI:
case V8SI_FTYPE_V8SI_V8SI_UHI:
nargs = 3;
@@ -11825,9 +11828,12 @@ ix86_expand_args_builtin (const struct builtin_description *d,
case USI_FTYPE_V32QI_V32QI_INT_USI:
case UHI_FTYPE_V16QI_V16QI_INT_UHI:
case USI_FTYPE_V32HI_V32HI_INT_USI:
+ case USI_FTYPE_V32BF_V32BF_INT_USI:
case USI_FTYPE_V32HF_V32HF_INT_USI:
case UHI_FTYPE_V16HI_V16HI_INT_UHI:
+ case UHI_FTYPE_V16BF_V16BF_INT_UHI:
case UQI_FTYPE_V8HI_V8HI_INT_UQI:
+ case UQI_FTYPE_V8BF_V8BF_INT_UQI:
nargs = 4;
mask_pos = 1;
nargs_constant = 1;
@@ -11864,6 +11870,9 @@ ix86_expand_args_builtin (const struct builtin_description *d,
case V32HI_FTYPE_V32HI_INT_V32HI_USI:
case V16HI_FTYPE_V16HI_INT_V16HI_UHI:
case V8HI_FTYPE_V8HI_INT_V8HI_UQI:
+ case V32BF_FTYPE_V32BF_INT_V32BF_USI:
+ case V16BF_FTYPE_V16BF_INT_V16BF_UHI:
+ case V8BF_FTYPE_V8BF_INT_V8BF_UQI:
case V4DI_FTYPE_V4DI_INT_V4DI_UQI:
case V2DI_FTYPE_V2DI_INT_V2DI_UQI:
case V8SI_FTYPE_V8SI_INT_V8SI_UQI:
@@ -15662,6 +15671,13 @@ rdseed_step:
case IX86_BUILTIN_RDPID:
return ix86_expand_special_args_builtin (bdesc_args + i, exp,
target);
+ case IX86_BUILTIN_VCOMSBF16EQ:
+ case IX86_BUILTIN_VCOMSBF16NE:
+ case IX86_BUILTIN_VCOMSBF16GT:
+ case IX86_BUILTIN_VCOMSBF16GE:
+ case IX86_BUILTIN_VCOMSBF16LT:
+ case IX86_BUILTIN_VCOMSBF16LE:
+ return ix86_expand_sse_comi (bdesc_args + i, exp, target);
case IX86_BUILTIN_FABSQ:
case IX86_BUILTIN_COPYSIGNQ:
if (!TARGET_SSE)
@@ -230,6 +230,11 @@
UNSPEC_VCVTNEPH2HF8S
UNSPEC_VCVTHF82PH
UNSPEC_VSCALEFPBF16
+ UNSPEC_VRNDSCALENEPBF16
+ UNSPEC_VREDUCENEPBF16
+ UNSPEC_VGETMANTPBF16
+ UNSPEC_VFPCLASSPBF16
+ UNSPEC_VCOMSBF16
])
(define_c_enum "unspecv" [
@@ -835,6 +840,7 @@
(define_mode_attr vecmemsuffix
[(V32HF "{z}") (V16HF "{y}") (V8HF "{x}")
+ (V32BF "{z}") (V16BF "{y}") (V8BF "{x}")
(V16SF "{z}") (V8SF "{y}") (V4SF "{x}")
(V8DF "{z}") (V4DF "{y}") (V2DF "{x}")])
@@ -32105,3 +32111,89 @@
[(set_attr "prefix" "evex")
(set_attr "type" "ssemuladd")
(set_attr "mode" "<sseinsnmode>")])
+
+(define_insn "avx10_2_rsqrtpbf16_<mode><mask_name>"
+ [(set (match_operand:VBF_AVX10_2 0 "register_operand" "=v")
+ (unspec:VBF_AVX10_2
+ [(match_operand:VBF_AVX10_2 1 "nonimmediate_operand" "vm")]
+ UNSPEC_RSQRT))]
+ "TARGET_AVX10_2_256"
+ "vrsqrtpbf16\t{%1, %0<mask_operand2>|%0<mask_operand2>, %1}"
+ [(set_attr "prefix" "evex")])
+
+(define_insn "avx10_2_sqrtnepbf16_<mode><mask_name>"
+ [(set (match_operand:VBF_AVX10_2 0 "register_operand" "=v")
+ (sqrt:VBF_AVX10_2
+ (match_operand:VBF_AVX10_2 1 "nonimmediate_operand" "vm")))]
+ "TARGET_AVX10_2_256"
+ "vsqrtnepbf16\t{%1, %0<mask_operand2>|%0<mask_operand2>, %1}"
+ [(set_attr "prefix" "evex")])
+
+(define_insn "avx10_2_rcppbf16_<mode><mask_name>"
+ [(set (match_operand:VBF_AVX10_2 0 "register_operand" "=v")
+ (unspec:VBF_AVX10_2
+ [(match_operand:VBF_AVX10_2 1 "nonimmediate_operand" "vm")]
+ UNSPEC_RCP))]
+ "TARGET_AVX10_2_256"
+ "vrcppbf16\t{%1, %0<mask_operand2>|%0<mask_operand2>, %1}"
+ [(set_attr "prefix" "evex")])
+
+(define_insn "avx10_2_getexppbf16_<mode><mask_name>"
+ [(set (match_operand:VBF_AVX10_2 0 "register_operand" "=v")
+ (unspec:VBF_AVX10_2
+ [(match_operand:VBF_AVX10_2 1 "nonimmediate_operand" "vm")]
+ UNSPEC_GETEXP))]
+ "TARGET_AVX10_2_256"
+ "vgetexppbf16\t{%1, %0<mask_operand2>|%0<mask_operand2>, %1}"
+ [(set_attr "prefix" "evex")])
+
+(define_int_iterator BF16IMMOP
+ [UNSPEC_VRNDSCALENEPBF16
+ UNSPEC_VREDUCENEPBF16
+ UNSPEC_VGETMANTPBF16])
+
+(define_int_attr bf16immop
+ [(UNSPEC_VRNDSCALENEPBF16 "rndscalene")
+ (UNSPEC_VREDUCENEPBF16 "reducene")
+ (UNSPEC_VGETMANTPBF16 "getmant")])
+
+(define_insn "avx10_2_<bf16immop>pbf16_<mode><mask_name>"
+ [(set (match_operand:VBF_AVX10_2 0 "register_operand" "=v")
+ (unspec:VBF_AVX10_2
+ [(match_operand:VBF_AVX10_2 1 "nonimmediate_operand" "vm")
+ (match_operand:SI 2 "const_0_to_255_operand")]
+ BF16IMMOP))]
+ "TARGET_AVX10_2_256"
+ "v<bf16immop>pbf16\t{%2, %1, %0<mask_operand3>|%0<mask_operand3>, %1, %2}"
+ [(set_attr "prefix" "evex")])
+
+(define_insn "avx10_2_fpclasspbf16_<mode><mask_scalar_merge_name>"
+ [(set (match_operand:<avx512fmaskmode> 0 "register_operand" "=k")
+ (unspec:<avx512fmaskmode>
+ [(match_operand:VBF_AVX10_2 1 "nonimmediate_operand" "vm")
+ (match_operand 2 "const_0_to_255_operand")]
+ UNSPEC_VFPCLASSPBF16))]
+ "TARGET_AVX10_2_256"
+ "vfpclasspbf16<vecmemsuffix>\t{%2, %1, %0<mask_scalar_merge_operand3>|%0<mask_scalar_merge_operand3>, %1, %2}"
+ [(set_attr "prefix" "evex")])
+
+(define_insn "avx10_2_cmppbf16_<mode><mask_scalar_merge_name>"
+ [(set (match_operand:<avx512fmaskmode> 0 "register_operand" "=k")
+ (unspec:<avx512fmaskmode>
+ [(match_operand:VBF_AVX10_2 1 "register_operand" "v")
+ (match_operand:VBF_AVX10_2 2 "nonimmediate_operand" "vm")
+ (match_operand 3 "const_0_to_31_operand" "n")]
+ UNSPEC_PCMP))]
+ "TARGET_AVX10_2_256"
+ "vcmppbf16\t{%3, %2, %1, %0<mask_scalar_merge_operand4>|%0<mask_scalar_merge_operand4>, %1, %2, %3}"
+ [(set_attr "prefix" "evex")])
+
+(define_insn "avx10_2_comsbf16_v8bf"
+ [(set (reg:CCFP FLAGS_REG)
+ (unspec:CCFP
+ [(match_operand:V8BF 0 "register_operand" "v")
+ (match_operand:V8BF 1 "nonimmediate_operand" "vm")]
+ UNSPEC_VCOMSBF16))]
+ "TARGET_AVX10_2_256"
+ "vcomsbf16\t{%1, %0|%0, %1}"
+ [(set_attr "prefix" "evex")])
@@ -1016,6 +1016,25 @@
/* avx10_2-512convertintrin.h */
#define __builtin_ia32_vcvt2ps2phx512_mask_round(A, B, C, D, E) __builtin_ia32_vcvt2ps2phx512_mask_round(A, B, C, D, 8)
+/* avx10_2-512bf16intrin.h */
+#define __builtin_ia32_rndscalenepbf16512_mask(A, B, C, D) __builtin_ia32_rndscalenepbf16512_mask(A, 123, C, D)
+#define __builtin_ia32_reducenepbf16512_mask(A, B, C, D) __builtin_ia32_reducenepbf16512_mask(A, 123, C, D)
+#define __builtin_ia32_getmantpbf16512_mask(A, B, C, D) __builtin_ia32_getmantpbf16512_mask(A, 1, C, D)
+#define __builtin_ia32_fpclasspbf16512_mask(A, B, C) __builtin_ia32_fpclasspbf16512_mask(A, 1, C)
+#define __builtin_ia32_cmppbf16512_mask(A, B, C, D) __builtin_ia32_cmppbf16512_mask(A, B, 1, D)
+
+/* avx10_2bf16intrin.h */
+#define __builtin_ia32_rndscalenepbf16256_mask(A, B, C, D) __builtin_ia32_rndscalenepbf16256_mask(A, 123, C, D)
+#define __builtin_ia32_rndscalenepbf16128_mask(A, B, C, D) __builtin_ia32_rndscalenepbf16128_mask(A, 123, C, D)
+#define __builtin_ia32_reducenepbf16256_mask(A, B, C, D) __builtin_ia32_reducenepbf16256_mask(A, 123, C, D)
+#define __builtin_ia32_reducenepbf16128_mask(A, B, C, D) __builtin_ia32_reducenepbf16128_mask(A, 123, C, D)
+#define __builtin_ia32_getmantpbf16256_mask(A, B, C, D) __builtin_ia32_getmantpbf16256_mask(A, 1, C, D)
+#define __builtin_ia32_getmantpbf16128_mask(A, B, C, D) __builtin_ia32_getmantpbf16128_mask(A, 1, C, D)
+#define __builtin_ia32_fpclasspbf16256_mask(A, B, C) __builtin_ia32_fpclasspbf16256_mask(A, 1, C)
+#define __builtin_ia32_fpclasspbf16128_mask(A, B, C) __builtin_ia32_fpclasspbf16128_mask(A, 1, C)
+#define __builtin_ia32_cmppbf16256_mask(A, B, C, D) __builtin_ia32_cmppbf16256_mask(A, B, 1, D)
+#define __builtin_ia32_cmppbf16128_mask(A, B, C, D) __builtin_ia32_cmppbf16128_mask(A, B, 1, D)
+
#include <wmmintrin.h>
#include <immintrin.h>
#include <mm3dnow.h>
@@ -5,7 +5,7 @@
#ifndef DO_TEST
#define DO_TEST do_test
-#if defined(AVX10_512BIT)
+#if defined(AVX10_512BIT) || defined(AVX10_SCALAR)
static void test_512 (void);
#else
static void test_256 (void);
@@ -16,7 +16,7 @@ __attribute__ ((noinline))
static void
do_test (void)
{
-#if defined(AVX10_512BIT)
+#if defined(AVX10_512BIT) || defined(AVX10_SCALAR)
test_512 ();
#else
test_256 ();
@@ -53,6 +53,34 @@ scalef (float x, float y)
return _mm_cvtss_f32 (out);
}
+float NOINLINE
+getexp (float val)
+{
+ float res;
+ __m128 px = _mm_load_ss (&val);
+ __m128 mx = _mm_broadcastss_ps (px);
+ __m128 out = _mm_getexp_ps (mx);
+ return _mm_cvtss_f32 (out);
+}
+
+float NOINLINE
+rndscale (float val)
+{
+ __m128 px = _mm_load_ss (&val);
+ __m128 mx = _mm_broadcastss_ps (px);
+ __m128 out = _mm_roundscale_ps (mx, 0x10);
+ return _mm_cvtss_f32 (out);
+}
+
+float NOINLINE
+getmant(float val)
+{
+ __m128 px = _mm_load_ss (&val);
+ __m128 mx = _mm_broadcastss_ps (px);
+ __m128 out = _mm_getmant_ps (mx, 0, 0);
+ return _mm_cvtss_f32 (out);
+}
+
#endif /* AVX10_HELPER_INCLUDED */
/* Intrinsic being tested. It has different deffinitions,
@@ -37,9 +37,36 @@
/* { dg-final { scan-assembler-times "vfnmsub132nepbf16\[ \\t\]+%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\{%k\[0-9\]\}(?:\n|\[ \\t\]+#)" 1 } } */
/* { dg-final { scan-assembler-times "vfnmsub231nepbf16\[ \\t\]+%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\{%k\[0-9\]\}(?:\n|\[ \\t\]+#)" 1 } } */
/* { dg-final { scan-assembler-times "vfnmsub132nepbf16\[ \\t\]+%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\{%k\[0-9\]\}\{z\}\[^\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vrsqrtpbf16\[ \\t\]+\[^\{\n\]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vrsqrtpbf16\[ \\t\]+\[^\{\n\]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\{%k\[0-9\]\}\[^\{\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vrsqrtpbf16\[ \\t\]+\[^\{\n\]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\{%k\[0-9\]\}\{z\}\[^\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vsqrtnepbf16\[ \\t\]+\[^\{\n\]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vsqrtnepbf16\[ \\t\]+\[^\{\n\]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\{%k\[0-9\]\}\[^\{\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vsqrtnepbf16\[ \\t\]+\[^\{\n\]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\{%k\[0-9\]\}\{z\}\[^\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vrcppbf16\[ \\t\]+\[^\{\n\]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vrcppbf16\[ \\t\]+\[^\{\n\]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\{%k\[0-9\]\}\[^\{\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vrcppbf16\[ \\t\]+\[^\{\n\]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\{%k\[0-9\]\}\{z\}\[^\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vgetexppbf16\[ \\t\]+\[^\{\n\]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vgetexppbf16\[ \\t\]+\[^\{\n\]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\{%k\[0-9\]\}\[^\{\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vgetexppbf16\[ \\t\]+\[^\{\n\]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\{%k\[0-9\]\}\{z\}\[^\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vrndscalenepbf16\[ \\t\]+\[^\{\n\]*%zmm\[0-9\]+(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vrndscalenepbf16\[ \\t\]+\[^\{\n\]*%zmm\[0-9\]+\{%k\[1-7\]\}(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vrndscalenepbf16\[ \\t\]+\[^\{\n\]*%zmm\[0-9\]+\{%k\[1-7\]\}\{z\}(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vreducenepbf16\[ \\t\]+\[^\{\n\]*%zmm\[0-9\]+(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vreducenepbf16\[ \\t\]+\[^\{\n\]*%zmm\[0-9\]+\{%k\[1-7\]\}(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vreducenepbf16\[ \\t\]+\[^\{\n\]*%zmm\[0-9\]+\{%k\[1-7\]\}\{z\}(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vgetmantpbf16\[ \\t\]+\[^\{\n\]*%zmm\[0-9\]+(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vgetmantpbf16\[ \\t\]+\[^\{\n\]*%zmm\[0-9\]+\{%k\[1-7\]\}(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vgetmantpbf16\[ \\t\]+\[^\{\n\]*%zmm\[0-9\]+\{%k\[1-7\]\}\{z\}(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vfpclasspbf16z\[ \\t\]+\[^\{\n\]*%zmm\[0-9\]+\[^\n^k\]*%k\[0-7\](?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vfpclasspbf16z\[ \\t\]+\[^\{\n\]*%zmm\[0-9\]+\[^\n^k\]*%k\[0-7\]\{%k\[1-7\]\}(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vcmppbf16\[ \\t\]+\\\$1\[^\n\r]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\[^\n\r]*%k\[0-9\](?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vcmppbf16\[ \\t\]+\\\$2\[^\n\r]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\[^\n\r]*%k\[0-9\]\{%k\[0-9\]\}\[^\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */
#include <immintrin.h>
+#define IMM 123
+
volatile __m512bh res, x1, x2;
volatile __mmask32 m32;
@@ -84,4 +111,35 @@ avx10_2_512_test (void)
res = _mm512_mask_fnmsubne_pbh (res, m32, x1, x2);
res = _mm512_mask3_fnmsubne_pbh (res, x1, x2, m32);
res = _mm512_maskz_fnmsubne_pbh (m32,res, x1, x2);
+
+ res = _mm512_rsqrt_pbh (x1);
+ res = _mm512_mask_rsqrt_pbh (res, m32, x1);
+ res = _mm512_maskz_rsqrt_pbh (m32, x1);
+ res = _mm512_sqrtne_pbh (x1);
+ res = _mm512_mask_sqrtne_pbh (res, m32, x1);
+ res = _mm512_maskz_sqrtne_pbh (m32, x1);
+ res = _mm512_rcp_pbh (x1);
+ res = _mm512_mask_rcp_pbh (res, m32, x1);
+ res = _mm512_maskz_rcp_pbh (m32, x1);
+ res = _mm512_getexp_pbh (x1);
+ res = _mm512_mask_getexp_pbh (res, m32, x1);
+ res = _mm512_maskz_getexp_pbh (m32, x1);
+
+ res = _mm512_roundscalene_pbh (x1, IMM);
+ res = _mm512_mask_roundscalene_pbh (res, m32, x1, IMM);
+ res = _mm512_maskz_roundscalene_pbh (m32, x1, IMM);
+ res = _mm512_reducene_pbh (x1, IMM);
+ res = _mm512_mask_reducene_pbh (res, m32, x1, IMM);
+ res = _mm512_maskz_reducene_pbh (m32, x1, IMM);
+ res = _mm512_getmant_pbh (x1, _MM_MANT_NORM_p75_1p5, _MM_MANT_SIGN_src);
+ res = _mm512_mask_getmant_pbh (res, m32, x1, _MM_MANT_NORM_p75_1p5,
+ _MM_MANT_SIGN_src);
+ res = _mm512_maskz_getmant_pbh (m32, x1, _MM_MANT_NORM_p75_1p5,
+ _MM_MANT_SIGN_src);
+
+ m32 = _mm512_fpclass_pbh_mask (x1, 13);
+ m32 = _mm512_mask_fpclass_pbh_mask (2, x1, 13);
+
+ m32 = _mm512_cmp_pbh_mask (x1, x2, 1);
+ m32 = _mm512_mask_cmp_pbh_mask (m32, x1, x2, 2);
}
new file mode 100644
@@ -0,0 +1,36 @@
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx10.2-512" } */
+/* { dg-require-effective-target avx10_2_512 } */
+
+#ifndef AVX10_2
+#define AVX10_2
+#define AVX10_2_512
+#define AVX10_512BIT
+#endif
+#include "avx10-helper.h"
+#define SIZE_RES (AVX512F_LEN / 16)
+
+void
+TEST (void)
+{
+ int i;
+ __mmask32 res1, res2, exp = 0;
+ UNION_TYPE (AVX512F_LEN, bf16_uw) src1, src2;
+ MASK_TYPE mask = MASK_VALUE;
+
+ for (i = 0; i < SIZE_RES; i++)
+ {
+ float x = 0.5;
+ float y = 0.25;
+ src2.a[i] = convert_fp32_to_bf16 (y);
+ src1.a[i] = convert_fp32_to_bf16 (x);
+ if (src1.a[i] == src2.a[i])
+ exp |= 1 << i;
+ }
+
+ res1 = INTRINSIC (_cmp_pbh_mask) (src1.x, src2.x, 0);
+ res2 = INTRINSIC (_mask_cmp_pbh_mask) (mask, src1.x, src2.x, 0);
+
+ if (exp != res1 || exp != res2)
+ abort ();
+}
new file mode 100644
@@ -0,0 +1,44 @@
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx10.2-512" } */
+/* { dg-require-effective-target avx10_2_512 } */
+
+#ifndef AVX10_2
+#define AVX10_2
+#define AVX10_2_512
+#define AVX10_512BIT
+#endif
+#include "avx10-helper.h"
+#define SIZE_RES (AVX512F_LEN / 16)
+
+void
+TEST (void)
+{
+ int i;
+ MASK_TYPE res1 = 0, res2 = 0;
+ __mmask16 exp = 0;
+ UNION_TYPE (AVX512F_LEN, bf16_uw) src1;
+ UNION_TYPE (AVX512F_LEN, ) src2;
+ MASK_TYPE mask = MASK_VALUE;
+
+ for (i = 0; i < SIZE_RES / 2; i++)
+ {
+ src1.a[i] = 0;
+ src2.a[i] = (uint32_t) (src1.a[i]) << 16;
+ }
+
+ for (i = SIZE_RES / 2; i < SIZE_RES; i++)
+ src1.a[i] = 0;
+
+ src1.a[0] = 0x7FC0;
+ src2.a[0] = convert_bf16_to_fp32 (src1.a[0]);
+
+ _mm_setcsr (0x9FC0);
+ exp = INTRINSIC (_fpclass_ps_mask) (src2.x, 0x01);
+
+ _mm_setcsr (0x1f80);
+ res1 = INTRINSIC (_fpclass_pbh_mask) (src1.x, 0x01);
+ res2 = INTRINSIC (_mask_fpclass_pbh_mask) (mask, src1.x, 1);
+
+ if (exp != res1 || exp != res2)
+ abort ();
+}
new file mode 100644
@@ -0,0 +1,47 @@
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx10.2-512" } */
+/* { dg-require-effective-target avx10_2_512 } */
+
+#ifndef AVX10_2
+#define AVX10_2
+#define AVX10_2_512
+#define AVX10_512BIT
+#endif
+#include "avx10-helper.h"
+#define SIZE_RES (AVX512F_LEN / 16)
+
+void
+TEST (void)
+{
+ int i;
+ UNION_TYPE (AVX512F_LEN, bf16_uw) res1, res2, res3, src1;
+ MASK_TYPE mask = MASK_VALUE;
+ unsigned short res_ref[SIZE_RES], res_ref2[SIZE_RES];
+
+ for (i = 0; i < SIZE_RES; i++)
+ {
+ res1.a[i] = 0;
+ res2.a[i] = DEFAULT_VALUE;
+ res3.a[i] = DEFAULT_VALUE;
+ float f, s;
+ f = 28 * i + 1;
+ src1.a[i] = convert_fp32_to_bf16 (f);
+ s = convert_bf16_to_fp32 (src1.a[i]);
+ res_ref[i] = res_ref2[i] = convert_fp32_to_bf16 (getexp (s));
+ }
+
+ res1.x = INTRINSIC (_getexp_pbh) (src1.x);
+ res2.x = INTRINSIC (_mask_getexp_pbh) (res2.x, mask, src1.x);
+ res3.x = INTRINSIC (_maskz_getexp_pbh) (mask, src1.x);
+
+ if (UNION_CHECK (AVX512F_LEN, bf16_uw) (res1, res_ref))
+ abort ();
+
+ MASK_MERGE (bf16_uw) (res_ref2, mask, SIZE_RES);
+ if (UNION_CHECK (AVX512F_LEN, bf16_uw) (res2, res_ref2))
+ abort ();
+
+ MASK_ZERO (bf16_uw) (res_ref2, mask, SIZE_RES);
+ if (UNION_CHECK (AVX512F_LEN, bf16_uw) (res3, res_ref2))
+ abort ();
+}
new file mode 100644
@@ -0,0 +1,50 @@
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx10.2-512" } */
+/* { dg-require-effective-target avx10_2_512 } */
+
+#ifndef AVX10_2
+#define AVX10_2
+#define AVX10_2_512
+#define AVX10_512BIT
+#endif
+#include "avx10-helper.h"
+#define SIZE_RES (AVX512F_LEN / 16)
+
+void
+TEST (void)
+{
+ int i;
+ UNION_TYPE (AVX512F_LEN, bf16_uw) res1, res2, res3, src1;
+ MASK_TYPE mask = MASK_VALUE;
+ unsigned short res_ref[SIZE_RES], res_ref2[SIZE_RES];
+
+ for (i = 0; i < SIZE_RES; i++)
+ {
+ res1.a[i] = 5.0;
+ res2.a[i] = DEFAULT_VALUE;
+ res3.a[i] = DEFAULT_VALUE;
+ src1.a[i] = 0.5;
+ float x = convert_bf16_to_fp32 (src1.a[i]);
+ res_ref[i] = res_ref2[i] = convert_fp32_to_bf16 (getmant (x));
+ }
+
+ res1.x = INTRINSIC (_getmant_pbh) (src1.x, _MM_MANT_NORM_1_2,
+ _MM_MANT_SIGN_src);
+ res2.x = INTRINSIC (_mask_getmant_pbh) (res2.x, mask, src1.x,
+ _MM_MANT_NORM_1_2,
+ _MM_MANT_SIGN_src);
+ res3.x = INTRINSIC (_maskz_getmant_pbh) (mask, src1.x,
+ _MM_MANT_NORM_1_2,
+ _MM_MANT_SIGN_src);
+
+ if (UNION_CHECK (AVX512F_LEN, bf16_uw) (res1, res_ref))
+ abort ();
+
+ MASK_MERGE (bf16_uw) (res_ref2, mask, SIZE_RES);
+ if (UNION_CHECK (AVX512F_LEN, bf16_uw) (res2, res_ref2))
+ abort ();
+
+ MASK_ZERO (bf16_uw) (res_ref2, mask, SIZE_RES);
+ if (UNION_CHECK (AVX512F_LEN, bf16_uw) (res3, res_ref2))
+ abort ();
+}
new file mode 100644
@@ -0,0 +1,45 @@
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx10.2-512" } */
+/* { dg-require-effective-target avx10_2_512 } */
+
+#ifndef AVX10_2
+#define AVX10_2
+#define AVX10_2_512
+#define AVX10_512BIT
+#endif
+#include "avx10-helper.h"
+#define SIZE_RES (AVX512F_LEN / 16)
+
+void
+TEST (void)
+{
+ int i;
+ UNION_TYPE (AVX512F_LEN, bf16_uw) res1, res2, res3, src1;
+ MASK_TYPE mask = MASK_VALUE;
+ unsigned short res_ref[SIZE_RES], res_ref2[SIZE_RES];
+
+ for (i = 0; i < SIZE_RES; i++)
+ {
+ res1.a[i] = 0;
+ res2.a[i] = DEFAULT_VALUE;
+ res3.a[i] = DEFAULT_VALUE;
+ float s1 = 2.0;
+ src1.a[i] = convert_fp32_to_bf16 (s1);
+ res_ref[i] = res_ref2[i] = convert_fp32_to_bf16 (1.0 / s1);
+ }
+
+ res1.x = INTRINSIC (_rcp_pbh) (src1.x);
+ res2.x = INTRINSIC (_mask_rcp_pbh) (res2.x, mask, src1.x);
+ res3.x = INTRINSIC (_maskz_rcp_pbh) (mask, src1.x);
+
+ if (UNION_CHECK (AVX512F_LEN, bf16_uw) (res1, res_ref))
+ abort ();
+
+ MASK_MERGE (bf16_uw) (res_ref2, mask, SIZE_RES);
+ if (UNION_CHECK (AVX512F_LEN, bf16_uw) (res2, res_ref2))
+ abort ();
+
+ MASK_ZERO (bf16_uw) (res_ref2, mask, SIZE_RES);
+ if (UNION_CHECK (AVX512F_LEN, bf16_uw) (res3, res_ref2))
+ abort ();
+}
new file mode 100644
@@ -0,0 +1,50 @@
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx10.2-512" } */
+/* { dg-require-effective-target avx10_2_512 } */
+
+#ifndef AVX10_2
+#define AVX10_2
+#define AVX10_2_512
+#define AVX10_512BIT
+#endif
+#include "avx10-helper.h"
+#define SIZE_RES (AVX512F_LEN / 16)
+
+void
+TEST (void)
+{
+ int i;
+ UNION_TYPE (AVX512F_LEN, bf16_uw) res1, res2, res3, src1;
+ MASK_TYPE mask = MASK_VALUE;
+ unsigned short res_ref[SIZE_RES], res_ref2[SIZE_RES];
+
+ for (i = 0; i < SIZE_RES; i++)
+ {
+ res1.a[i] = 5.0;
+ res2.a[i] = DEFAULT_VALUE;
+ res3.a[i] = DEFAULT_VALUE;
+ float s = (float) (SIZE_RES - 1) / (float) i;
+ src1.a[i] = convert_fp32_to_bf16 (s);
+ float x = convert_bf16_to_fp32 (src1.a[i]);
+ __m128 px = _mm_load_ss (&x);
+ __m128 mx = _mm_broadcastss_ps (px);
+ __m128 out = _mm_reduce_ps (mx, 0x10);
+ float res = _mm_cvtss_f32 (out);
+ res_ref[i] = res_ref2[i] = convert_fp32_to_bf16_ne (res);
+ }
+
+ res1.x = INTRINSIC (_reducene_pbh) (src1.x, 0x10);
+ res2.x = INTRINSIC (_mask_reducene_pbh) (res2.x, mask, src1.x, 0x10);
+ res3.x = INTRINSIC (_maskz_reducene_pbh) (mask, src1.x, 0x10);
+
+ if (UNION_CHECK (AVX512F_LEN, bf16_uw) (res1, res_ref))
+ abort ();
+
+ MASK_MERGE (bf16_uw) (res_ref2, mask, SIZE_RES);
+ if (UNION_CHECK (AVX512F_LEN, bf16_uw) (res2, res_ref2))
+ abort ();
+
+ MASK_ZERO (bf16_uw) (res_ref2, mask, SIZE_RES);
+ if (UNION_CHECK (AVX512F_LEN, bf16_uw) (res3, res_ref2))
+ abort ();
+}
new file mode 100644
@@ -0,0 +1,46 @@
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx10.2-512" } */
+/* { dg-require-effective-target avx10_2_512 } */
+
+#ifndef AVX10_2
+#define AVX10_2
+#define AVX10_2_512
+#define AVX10_512BIT
+#endif
+#include "avx10-helper.h"
+#define SIZE_RES (AVX512F_LEN / 16)
+
+void
+TEST (void)
+{
+ int i;
+ UNION_TYPE (AVX512F_LEN, bf16_uw) res1, res2, res3, src1;
+ MASK_TYPE mask = MASK_VALUE;
+ unsigned short res_ref[SIZE_RES], res_ref2[SIZE_RES];
+
+ for (i = 0; i < SIZE_RES; i++)
+ {
+ res1.a[i] = 5.0;
+ res2.a[i] = DEFAULT_VALUE;
+ res3.a[i] = DEFAULT_VALUE;
+ float s = (float) (SIZE_RES - 1) / (float) i;
+ src1.a[i] = convert_fp32_to_bf16 (s);
+ float x = convert_bf16_to_fp32 (src1.a[i]);
+ res_ref[i] = res_ref2[i] = convert_fp32_to_bf16_ne (rndscale (x));
+ }
+
+ res1.x = INTRINSIC (_roundscalene_pbh) (src1.x, 0x10);
+ res2.x = INTRINSIC (_mask_roundscalene_pbh) (res2.x, mask, src1.x, 0x10);
+ res3.x = INTRINSIC (_maskz_roundscalene_pbh) (mask, src1.x, 0x10);
+
+ if (UNION_CHECK (AVX512F_LEN, bf16_uw) (res1, res_ref))
+ abort ();
+
+ MASK_MERGE (bf16_uw) (res_ref2, mask, SIZE_RES);
+ if (UNION_CHECK (AVX512F_LEN, bf16_uw) (res2, res_ref2))
+ abort ();
+
+ MASK_ZERO (bf16_uw) (res_ref2, mask, SIZE_RES);
+ if (UNION_CHECK (AVX512F_LEN, bf16_uw) (res3, res_ref2))
+ abort ();
+}
new file mode 100644
@@ -0,0 +1,47 @@
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx10.2-512" } */
+/* { dg-require-effective-target avx10_2_512 } */
+
+#ifndef AVX10_2
+#define AVX10_2
+#define AVX10_2_512
+#define AVX10_512BIT
+#endif
+#include "avx10-helper.h"
+#include <math.h>
+#define SIZE_RES (AVX512F_LEN / 16)
+
+void
+TEST (void)
+{
+ int i;
+ UNION_TYPE (AVX512F_LEN, bf16_uw) res1, res2, res3, src1;
+ MASK_TYPE mask = MASK_VALUE;
+ unsigned short res_ref[SIZE_RES], res_ref2[SIZE_RES];
+
+ for (i = 0; i < SIZE_RES; i++)
+ {
+ res1.a[i] = 0;
+ res2.a[i] = DEFAULT_VALUE;
+ res3.a[i] = DEFAULT_VALUE;
+ float s1 = 2.0;
+ float rs = 1.0 / sqrtf (s1);
+ src1.a[i] = convert_fp32_to_bf16 (s1);
+ res_ref[i] = res_ref2[i] = convert_fp32_to_bf16 (rs);
+ }
+
+ res1.x = INTRINSIC (_rsqrt_pbh) (src1.x);
+ res2.x = INTRINSIC (_mask_rsqrt_pbh) (res2.x, mask, src1.x);
+ res3.x = INTRINSIC (_maskz_rsqrt_pbh) (mask, src1.x);
+
+ if (UNION_CHECK (AVX512F_LEN, bf16_uw) (res1, res_ref))
+ abort ();
+
+ MASK_MERGE (bf16_uw) (res_ref2, mask, SIZE_RES);
+ if (UNION_CHECK (AVX512F_LEN, bf16_uw) (res2, res_ref2))
+ abort ();
+
+ MASK_ZERO (bf16_uw) (res_ref2, mask, SIZE_RES);
+ if (UNION_CHECK (AVX512F_LEN, bf16_uw) (res3, res_ref2))
+ abort ();
+}
@@ -31,7 +31,7 @@ TEST (void)
xx = convert_bf16_to_fp32 (src1.a[i]);
yy = convert_bf16_to_fp32 (src2.a[i]);
res = scalef (xx, yy);
- res_ref[i] = res_ref2[i] = convert_fp32_to_bf16_ne(res);
+ res_ref[i] = res_ref2[i] = convert_fp32_to_bf16 (res);
}
res1.x = INTRINSIC (_scalef_pbh) (src1.x, src2.x);
new file mode 100644
@@ -0,0 +1,47 @@
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx10.2-512" } */
+/* { dg-require-effective-target avx10_2_512 } */
+
+#ifndef AVX10_2
+#define AVX10_2
+#define AVX10_2_512
+#define AVX10_512BIT
+#endif
+#include "avx10-helper.h"
+#include <math.h>
+#define SIZE_RES (AVX512F_LEN / 16)
+
+void
+TEST (void)
+{
+ int i;
+ UNION_TYPE (AVX512F_LEN, bf16_uw) res1, res2, res3, src1;
+ MASK_TYPE mask = MASK_VALUE;
+ unsigned short res_ref[SIZE_RES], res_ref2[SIZE_RES];
+
+ for (i = 0; i < SIZE_RES; i++)
+ {
+ res1.a[i] = 0;
+ res2.a[i] = DEFAULT_VALUE;
+ res3.a[i] = DEFAULT_VALUE;
+ float s1 = i + 1.0;
+ float rs = sqrtf (s1);
+ src1.a[i] = convert_fp32_to_bf16_ne (s1);
+ res_ref[i] = res_ref2[i] = convert_fp32_to_bf16_ne (rs);
+ }
+
+ res1.x = INTRINSIC (_sqrtne_pbh) (src1.x);
+ res2.x = INTRINSIC (_mask_sqrtne_pbh) (res2.x, mask, src1.x);
+ res3.x = INTRINSIC (_maskz_sqrtne_pbh) (mask, src1.x);
+
+ if (UNION_CHECK (AVX512F_LEN, bf16_uw) (res1, res_ref))
+ abort ();
+
+ MASK_MERGE (bf16_uw) (res_ref2, mask, SIZE_RES);
+ if (UNION_CHECK (AVX512F_LEN, bf16_uw) (res2, res_ref2))
+ abort ();
+
+ MASK_ZERO (bf16_uw) (res_ref2, mask, SIZE_RES);
+ if (UNION_CHECK (AVX512F_LEN, bf16_uw) (res3, res_ref2))
+ abort ();
+}
@@ -74,9 +74,60 @@
/* { dg-final { scan-assembler-times "vfnmsub132nepbf16\[ \\t\]+%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\{%k\[0-9\]\}(?:\n|\[ \\t\]+#)" 1 } } */
/* { dg-final { scan-assembler-times "vfnmsub231nepbf16\[ \\t\]+%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\{%k\[0-9\]\}(?:\n|\[ \\t\]+#)" 1 } } */
/* { dg-final { scan-assembler-times "vfnmsub132nepbf16\[ \\t\]+%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\{%k\[0-9\]\}\{z\}\[^\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vrsqrtpbf16\[ \\t\]+\[^\{\n\]*%ymm\[0-9\]+\[^\n\r]*%ymm\[0-9\]+(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vrsqrtpbf16\[ \\t\]+\[^\{\n\]*%ymm\[0-9\]+\[^\n\r]*%ymm\[0-9\]+\{%k\[0-9\]\}\[^\{\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vrsqrtpbf16\[ \\t\]+\[^\{\n\]*%ymm\[0-9\]+\[^\n\r]*%ymm\[0-9\]+\{%k\[0-9\]\}\{z\}\[^\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vrsqrtpbf16\[ \\t\]+\[^\{\n\]*%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vrsqrtpbf16\[ \\t\]+\[^\{\n\]*%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\{%k\[0-9\]\}\[^\{\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vrsqrtpbf16\[ \\t\]+\[^\{\n\]*%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\{%k\[0-9\]\}\{z\}\[^\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vsqrtnepbf16\[ \\t\]+\[^\{\n\]*%ymm\[0-9\]+\[^\n\r]*%ymm\[0-9\]+(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vsqrtnepbf16\[ \\t\]+\[^\{\n\]*%ymm\[0-9\]+\[^\n\r]*%ymm\[0-9\]+\{%k\[0-9\]\}\[^\{\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vsqrtnepbf16\[ \\t\]+\[^\{\n\]*%ymm\[0-9\]+\[^\n\r]*%ymm\[0-9\]+\{%k\[0-9\]\}\{z\}\[^\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vsqrtnepbf16\[ \\t\]+\[^\{\n\]*%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vsqrtnepbf16\[ \\t\]+\[^\{\n\]*%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\{%k\[0-9\]\}\[^\{\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vsqrtnepbf16\[ \\t\]+\[^\{\n\]*%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\{%k\[0-9\]\}\{z\}\[^\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vrcppbf16\[ \\t\]+\[^\{\n\]*%ymm\[0-9\]+\[^\n\r]*%ymm\[0-9\]+(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vrcppbf16\[ \\t\]+\[^\{\n\]*%ymm\[0-9\]+\[^\n\r]*%ymm\[0-9\]+\{%k\[0-9\]\}\[^\{\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vrcppbf16\[ \\t\]+\[^\{\n\]*%ymm\[0-9\]+\[^\n\r]*%ymm\[0-9\]+\{%k\[0-9\]\}\{z\}\[^\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vrcppbf16\[ \\t\]+\[^\{\n\]*%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vrcppbf16\[ \\t\]+\[^\{\n\]*%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\{%k\[0-9\]\}\[^\{\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vrcppbf16\[ \\t\]+\[^\{\n\]*%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\{%k\[0-9\]\}\{z\}\[^\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vgetexppbf16\[ \\t\]+\[^\{\n\]*%ymm\[0-9\]+\[^\n\r]*%ymm\[0-9\]+(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vgetexppbf16\[ \\t\]+\[^\{\n\]*%ymm\[0-9\]+\[^\n\r]*%ymm\[0-9\]+\{%k\[0-9\]\}\[^\{\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vgetexppbf16\[ \\t\]+\[^\{\n\]*%ymm\[0-9\]+\[^\n\r]*%ymm\[0-9\]+\{%k\[0-9\]\}\{z\}\[^\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vgetexppbf16\[ \\t\]+\[^\{\n\]*%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vgetexppbf16\[ \\t\]+\[^\{\n\]*%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\{%k\[0-9\]\}\[^\{\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vgetexppbf16\[ \\t\]+\[^\{\n\]*%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\{%k\[0-9\]\}\{z\}\[^\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vrndscalenepbf16\[ \\t\]+\[^\{\n\]*%ymm\[0-9\]+(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vrndscalenepbf16\[ \\t\]+\[^\{\n\]*%ymm\[0-9\]+\{%k\[1-7\]\}(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vrndscalenepbf16\[ \\t\]+\[^\{\n\]*%ymm\[0-9\]+\{%k\[1-7\]\}\{z\}(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vrndscalenepbf16\[ \\t\]+\[^\{\n\]*%xmm\[0-9\]+(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vrndscalenepbf16\[ \\t\]+\[^\{\n\]*%xmm\[0-9\]+\{%k\[1-7\]\}(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vrndscalenepbf16\[ \\t\]+\[^\{\n\]*%xmm\[0-9\]+\{%k\[1-7\]\}\{z\}(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vreducenepbf16\[ \\t\]+\[^\{\n\]*%ymm\[0-9\]+(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vreducenepbf16\[ \\t\]+\[^\{\n\]*%ymm\[0-9\]+\{%k\[1-7\]\}(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vreducenepbf16\[ \\t\]+\[^\{\n\]*%ymm\[0-9\]+\{%k\[1-7\]\}\{z\}(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vreducenepbf16\[ \\t\]+\[^\{\n\]*%xmm\[0-9\]+(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vreducenepbf16\[ \\t\]+\[^\{\n\]*%xmm\[0-9\]+\{%k\[1-7\]\}(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vreducenepbf16\[ \\t\]+\[^\{\n\]*%xmm\[0-9\]+\{%k\[1-7\]\}\{z\}(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vgetmantpbf16\[ \\t\]+\[^\{\n\]*%ymm\[0-9\]+(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vgetmantpbf16\[ \\t\]+\[^\{\n\]*%ymm\[0-9\]+\{%k\[1-7\]\}(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vgetmantpbf16\[ \\t\]+\[^\{\n\]*%ymm\[0-9\]+\{%k\[1-7\]\}\{z\}(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vgetmantpbf16\[ \\t\]+\[^\{\n\]*%xmm\[0-9\]+(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vgetmantpbf16\[ \\t\]+\[^\{\n\]*%xmm\[0-9\]+\{%k\[1-7\]\}(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vgetmantpbf16\[ \\t\]+\[^\{\n\]*%xmm\[0-9\]+\{%k\[1-7\]\}\{z\}(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vfpclasspbf16y\[ \\t\]+\[^\{\n\]*%ymm\[0-9\]+\[^\n^k\]*%k\[0-7\](?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vfpclasspbf16y\[ \\t\]+\[^\{\n\]*%ymm\[0-9\]+\[^\n^k\]*%k\[0-7\]\{%k\[1-7\]\}(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vfpclasspbf16x\[ \\t\]+\[^\{\n\]*%xmm\[0-9\]+\[^\n^k\]*%k\[0-7\](?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vfpclasspbf16x\[ \\t\]+\[^\{\n\]*%xmm\[0-9\]+\[^\n^k\]*%k\[0-7\]\{%k\[1-7\]\}(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vcmppbf16\[ \\t\]+\\\$1\[^\n\r]*%ymm\[0-9\]+\[^\n\r]*%ymm\[0-9\]+\[^\n\r]*%k\[0-9\](?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vcmppbf16\[ \\t\]+\\\$2\[^\n\r]*%ymm\[0-9\]+\[^\n\r]*%ymm\[0-9\]+\[^\n\r]*%k\[0-9\]\{%k\[0-9\]\}\[^\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vcmppbf16\[ \\t\]+\\\$1\[^\n\r]*%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\[^\n\r]*%k\[0-9\](?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "vcmppbf16\[ \\t\]+\\\$2\[^\n\r]*%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\[^\n\r]*%k\[0-9\]\{%k\[0-9\]\}\[^\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */
#include <immintrin.h>
+#define IMM 123
volatile __m256bh res, x1, x2;
volatile __m128bh res1, x3, x4;
volatile __mmask16 m16;
@@ -169,4 +220,67 @@ avx10_2_test (void)
res1 = _mm_mask_fnmsubne_pbh (res1, m8, x3, x4);
res1 = _mm_mask3_fnmsubne_pbh (res1, x3, x4, m8);
res1 = _mm_maskz_fnmsubne_pbh (m8,res1, x3, x4);
+
+ res = _mm256_rsqrt_pbh (x1);
+ res = _mm256_mask_rsqrt_pbh (res, m16, x1);
+ res = _mm256_maskz_rsqrt_pbh (m16, x1);
+ res1 = _mm_rsqrt_pbh (x3);
+ res1 = _mm_mask_rsqrt_pbh (res1, m8, x3);
+ res1 = _mm_maskz_rsqrt_pbh (m8, x3);
+
+ res = _mm256_sqrtne_pbh (x1);
+ res = _mm256_mask_sqrtne_pbh (res, m16, x1);
+ res = _mm256_maskz_sqrtne_pbh (m16, x1);
+ res1 = _mm_sqrtne_pbh (x3);
+ res1 = _mm_mask_sqrtne_pbh (res1, m8, x3);
+ res1 = _mm_maskz_sqrtne_pbh (m8, x3);
+
+ res = _mm256_rcp_pbh (x1);
+ res = _mm256_mask_rcp_pbh (res, m16, x1);
+ res = _mm256_maskz_rcp_pbh (m16, x1);
+ res1 = _mm_rcp_pbh (x3);
+ res1 = _mm_mask_rcp_pbh (res1, m8, x3);
+ res1 = _mm_maskz_rcp_pbh (m8, x3);
+
+ res = _mm256_getexp_pbh (x1);
+ res = _mm256_mask_getexp_pbh (res, m16, x1);
+ res = _mm256_maskz_getexp_pbh (m16, x1);
+ res1 = _mm_getexp_pbh (x3);
+ res1 = _mm_mask_getexp_pbh (res1, m8, x3);
+ res1 = _mm_maskz_getexp_pbh (m8, x3);
+
+ res = _mm256_roundscalene_pbh (x1, IMM);
+ res = _mm256_mask_roundscalene_pbh (res, m16, x1, IMM);
+ res = _mm256_maskz_roundscalene_pbh (m16, x1, IMM);
+ res1 = _mm_roundscalene_pbh (x3, IMM);
+ res1 = _mm_mask_roundscalene_pbh (res1, m8, x3, IMM);
+ res1 = _mm_maskz_roundscalene_pbh (m8, x3, IMM);
+
+ res = _mm256_reducene_pbh (x1, IMM);
+ res = _mm256_mask_reducene_pbh (res, m16, x1, IMM);
+ res = _mm256_maskz_reducene_pbh (m16, x1, IMM);
+ res1 = _mm_reducene_pbh (x3, IMM);
+ res1 = _mm_mask_reducene_pbh (res1, m8, x3, IMM);
+ res1 = _mm_maskz_reducene_pbh (m8, x3, IMM);
+
+ res = _mm256_getmant_pbh (x1, _MM_MANT_NORM_p75_1p5, _MM_MANT_SIGN_src);
+ res = _mm256_mask_getmant_pbh (res, m16, x1, _MM_MANT_NORM_p75_1p5,
+ _MM_MANT_SIGN_src);
+ res = _mm256_maskz_getmant_pbh (m16, x1, _MM_MANT_NORM_p75_1p5,
+ _MM_MANT_SIGN_src);
+ res1 = _mm_getmant_pbh (x3, _MM_MANT_NORM_p75_1p5, _MM_MANT_SIGN_src);
+ res1 = _mm_mask_getmant_pbh (res1, m8, x3, _MM_MANT_NORM_p75_1p5,
+ _MM_MANT_SIGN_src);
+ res1 = _mm_maskz_getmant_pbh (m8, x3, _MM_MANT_NORM_p75_1p5,
+ _MM_MANT_SIGN_src);
+
+ m16 = _mm256_fpclass_pbh_mask (x1, 13);
+ m16 = _mm256_mask_fpclass_pbh_mask (2, x1, 13);
+ m8 = _mm_fpclass_pbh_mask (x3, 13);
+ m8 = _mm_mask_fpclass_pbh_mask (2, x3, 13);
+
+ m16 = _mm256_cmp_pbh_mask (x1, x2, 1);
+ m16 = _mm256_mask_cmp_pbh_mask (m16, x1, x2, 2);
+ m8 = _mm_cmp_pbh_mask (x3, x4, 1);
+ m8 = _mm_mask_cmp_pbh_mask (m8, x3, x4, 2);
}
new file mode 100644
@@ -0,0 +1,16 @@
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx10.2" } */
+/* { dg-require-effective-target avx10_2 } */
+
+#define AVX10_2
+#define AVX512VL
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx10_2-512-vcmppbf16-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx10_2-512-vcmppbf16-2.c"
new file mode 100644
@@ -0,0 +1,19 @@
+/* { dg-do compile } */
+/* { dg-options "-mavx10.2 -O2" } */
+/* { dg-final { scan-assembler-times "vcomsbf16\[ \\t\]+\[^{}\n\]*%xmm\[0-9\]+(?:\n|\[ \\t\]+#)" 6 } } */
+/* { dg-final { scan-assembler-times "jp" 2 } } */
+#include <immintrin.h>
+
+volatile __m128bh x1, x2;
+volatile int res;
+
+void extern
+avx10_2_vcom_test (void)
+{
+ res = _mm_comeq_sbh (x1, x2);
+ res = _mm_comlt_sbh (x1, x2);
+ res = _mm_comle_sbh (x1, x2);
+ res = _mm_comgt_sbh (x1, x2);
+ res = _mm_comge_sbh (x1, x2);
+ res = _mm_comneq_sbh (x1, x2);
+}
new file mode 100644
@@ -0,0 +1,58 @@
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx10.2" } */
+/* { dg-require-effective-target avx10_2 } */
+
+#define AVX10_2
+#define AVX10_SCALAR
+#include "avx10-helper.h"
+#define SIZE_RES (128 / 16)
+
+#define CMP(PRED, IMM) \
+ exp = _mm_comi_round_ss (__A, __B, IMM, _MM_FROUND_NO_EXC); \
+ res1 = _mm_com##PRED##_sbh (src1.x, src2.x); \
+ if (exp != res1) \
+ abort ();
+
+void
+TEST (void)
+{
+ int i;
+ int res1, exp;
+ UNION_TYPE (128, bf16_uw) src1, src2;
+
+ struct
+ {
+ float x1;
+ float x2;
+ }
+ inputs[] =
+ {
+ { 4.3, 2.18 },
+ { -4.3, 3.18 },
+ { __builtin_nanf (""), -5.8 },
+ { -4.8, __builtin_nansf ("") },
+ { 3.8, __builtin_nansf ("") },
+ { 4.2, 4.2 },
+ { __builtin_nanf (""), __builtin_nansf ("") },
+ };
+
+ for (i = 0; i < sizeof (inputs) / sizeof (inputs[0]); i++)
+ {
+ float x = inputs[i].x1;
+ float y = inputs[i].x2;
+
+ __m128 __A = _mm_load_ss (&x);
+ __m128 __B = _mm_load_ss (&y);
+ for (int n = 0; n < SIZE_RES; n++)
+ {
+ src2.a[n] = convert_fp32_to_bf16(y);
+ src1.a[n] = convert_fp32_to_bf16(x);
+ }
+ CMP (eq, _CMP_EQ_OQ);
+ CMP (ge, _CMP_GE_OS);
+ CMP (gt, _CMP_GT_OS);
+ CMP (lt, _CMP_LT_OS);
+ CMP (le, _CMP_LE_OS);
+ CMP (neq, _CMP_NEQ_UQ);
+ }
+}
new file mode 100644
@@ -0,0 +1,16 @@
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx10.2" } */
+/* { dg-require-effective-target avx10_2 } */
+
+#define AVX10_2
+#define AVX512VL
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx10_2-512-vfpclasspbf16-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx10_2-512-vfpclasspbf16-2.c"
new file mode 100644
@@ -0,0 +1,16 @@
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx10.2" } */
+/* { dg-require-effective-target avx10_2 } */
+
+#define AVX10_2
+#define AVX512VL
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx10_2-512-vgetexppbf16-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx10_2-512-vgetexppbf16-2.c"
new file mode 100644
@@ -0,0 +1,16 @@
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx10.2" } */
+/* { dg-require-effective-target avx10_2 } */
+
+#define AVX10_2
+#define AVX512VL
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx10_2-512-vgetmantpbf16-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx10_2-512-vgetmantpbf16-2.c"
new file mode 100644
@@ -0,0 +1,16 @@
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx10.2" } */
+/* { dg-require-effective-target avx10_2 } */
+
+#define AVX10_2
+#define AVX512VL
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx10_2-512-vrcppbf16-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx10_2-512-vrcppbf16-2.c"
new file mode 100644
@@ -0,0 +1,16 @@
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx10.2" } */
+/* { dg-require-effective-target avx10_2 } */
+
+#define AVX10_2
+#define AVX512VL
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx10_2-512-vreducenepbf16-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx10_2-512-vreducenepbf16-2.c"
new file mode 100644
@@ -0,0 +1,16 @@
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx10.2" } */
+/* { dg-require-effective-target avx10_2 } */
+
+#define AVX10_2
+#define AVX512VL
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx10_2-512-vrndscalenepbf16-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx10_2-512-vrndscalenepbf16-2.c"
new file mode 100644
@@ -0,0 +1,16 @@
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx10.2" } */
+/* { dg-require-effective-target avx10_2 } */
+
+#define AVX10_2
+#define AVX512VL
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx10_2-512-vrsqrtpbf16-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx10_2-512-vrsqrtpbf16-2.c"
new file mode 100644
@@ -0,0 +1,16 @@
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx10.2" } */
+/* { dg-require-effective-target avx10_2 } */
+
+#define AVX10_2
+#define AVX512VL
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx10_2-512-vsqrtnepbf16-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx10_2-512-vsqrtnepbf16-2.c"
@@ -1024,4 +1024,23 @@
/* avx10_2-512convertintrin.h */
#define __builtin_ia32_vcvt2ps2phx512_mask_round(A, B, C, D, E) __builtin_ia32_vcvt2ps2phx512_mask_round(A, B, C, D, 8)
+/* avx10_2-512bf16intrin.h */
+#define __builtin_ia32_rndscalenepbf16512_mask(A, B, C, D) __builtin_ia32_rndscalenepbf16512_mask(A, 123, C, D)
+#define __builtin_ia32_reducenepbf16512_mask(A, B, C, D) __builtin_ia32_reducenepbf16512_mask(A, 123, C, D)
+#define __builtin_ia32_getmantpbf16512_mask(A, B, C, D) __builtin_ia32_getmantpbf16512_mask(A, 1, C, D)
+#define __builtin_ia32_fpclasspbf16512_mask(A, B, C) __builtin_ia32_fpclasspbf16512_mask(A, 1, C)
+#define __builtin_ia32_cmppbf16512_mask(A, B, C, D) __builtin_ia32_cmppbf16512_mask(A, B, 1, D)
+
+/* avx10_2bf16intrin.h */
+#define __builtin_ia32_rndscalenepbf16256_mask(A, B, C, D) __builtin_ia32_rndscalenepbf16256_mask(A, 123, C, D)
+#define __builtin_ia32_rndscalenepbf16128_mask(A, B, C, D) __builtin_ia32_rndscalenepbf16128_mask(A, 123, C, D)
+#define __builtin_ia32_reducenepbf16256_mask(A, B, C, D) __builtin_ia32_reducenepbf16256_mask(A, 123, C, D)
+#define __builtin_ia32_reducenepbf16128_mask(A, B, C, D) __builtin_ia32_reducenepbf16128_mask(A, 123, C, D)
+#define __builtin_ia32_getmantpbf16256_mask(A, B, C, D) __builtin_ia32_getmantpbf16256_mask(A, 1, C, D)
+#define __builtin_ia32_getmantpbf16128_mask(A, B, C, D) __builtin_ia32_getmantpbf16128_mask(A, 1, C, D)
+#define __builtin_ia32_fpclasspbf16256_mask(A, B, C) __builtin_ia32_fpclasspbf16256_mask(A, 1, C)
+#define __builtin_ia32_fpclasspbf16128_mask(A, B, C) __builtin_ia32_fpclasspbf16128_mask(A, 1, C)
+#define __builtin_ia32_cmppbf16256_mask(A, B, C, D) __builtin_ia32_cmppbf16256_mask(A, B, 1, D)
+#define __builtin_ia32_cmppbf16128_mask(A, B, C, D) __builtin_ia32_cmppbf16128_mask(A, B, 1, D)
+
#include <x86intrin.h>
@@ -1388,3 +1388,46 @@ test_2 (_mm256_cvtx_round2ps_ph, __m256h, __m256, __m256, 4)
/* avx10_2-512convertintrin.h */
test_2 (_mm512_cvtx_round2ps_ph, __m512h, __m512, __m512, 4)
+
+/* avx10_2-512bf16intrin.h */
+test_1 (_mm512_roundscalene_pbh, __m512bh, __m512bh, 123)
+test_2 (_mm512_maskz_roundscalene_pbh, __m512bh, __mmask32, __m512bh, 123)
+test_3 (_mm512_mask_roundscalene_pbh, __m512bh, __m512bh, __mmask32, __m512bh, 123)
+test_1 (_mm512_reducene_pbh, __m512bh, __m512bh, 123)
+test_2 (_mm512_maskz_reducene_pbh, __m512bh, __mmask32, __m512bh, 123)
+test_3 (_mm512_mask_reducene_pbh, __m512bh, __m512bh, __mmask32, __m512bh, 123)
+test_1x (_mm512_getmant_pbh, __m512bh, __m512bh, 1, 1)
+test_2x (_mm512_maskz_getmant_pbh, __m512bh, __mmask32,__m512bh, 1, 1)
+test_3x (_mm512_mask_getmant_pbh, __m512bh, __m512bh, __mmask32,__m512bh, 1, 1)
+test_1 (_mm512_fpclass_pbh_mask, __mmask32, __m512bh, 13)
+test_2 (_mm512_mask_fpclass_pbh_mask, __mmask32, __mmask32, __m512bh, 13)
+test_2 (_mm512_cmp_pbh_mask, __mmask32, __m512bh, __m512bh, 1)
+test_3 (_mm512_mask_cmp_pbh_mask, __mmask32, __mmask32,__m512bh, __m512bh, 1)
+
+/* avx10_2bf16intrin.h */
+test_1 (_mm256_roundscalene_pbh, __m256bh, __m256bh, 123)
+test_1 (_mm_roundscalene_pbh, __m128bh, __m128bh, 123)
+test_2 (_mm256_maskz_roundscalene_pbh, __m256bh, __mmask16, __m256bh, 123)
+test_2 (_mm_maskz_roundscalene_pbh, __m128bh, __mmask8, __m128bh, 123)
+test_3 (_mm256_mask_roundscalene_pbh, __m256bh, __m256bh, __mmask16, __m256bh, 123)
+test_3 (_mm_mask_roundscalene_pbh, __m128bh, __m128bh, __mmask8, __m128bh, 123)
+test_1 (_mm256_reducene_pbh, __m256bh, __m256bh, 123)
+test_1 (_mm_reducene_pbh, __m128bh, __m128bh, 123)
+test_2 (_mm256_maskz_reducene_pbh, __m256bh, __mmask16, __m256bh, 123)
+test_2 (_mm_maskz_reducene_pbh, __m128bh, __mmask8, __m128bh, 123)
+test_3 (_mm256_mask_reducene_pbh, __m256bh, __m256bh, __mmask16, __m256bh, 123)
+test_3 (_mm_mask_reducene_pbh, __m128bh, __m128bh, __mmask8, __m128bh, 123)
+test_1x (_mm256_getmant_pbh, __m256bh, __m256bh, 1, 1)
+test_1x (_mm_getmant_pbh, __m128bh, __m128bh, 1, 1)
+test_2x (_mm256_maskz_getmant_pbh, __m256bh, __mmask16,__m256bh, 1, 1)
+test_2x (_mm_maskz_getmant_pbh, __m128bh, __mmask8, __m128bh, 1, 1)
+test_3x (_mm256_mask_getmant_pbh, __m256bh, __m256bh, __mmask16,__m256bh, 1, 1)
+test_3x (_mm_mask_getmant_pbh, __m128bh, __m128bh, __mmask8, __m128bh, 1, 1)
+test_1 (_mm256_fpclass_pbh_mask, __mmask16, __m256bh, 13)
+test_1 (_mm_fpclass_pbh_mask, __mmask8, __m128bh, 13)
+test_2 (_mm256_mask_fpclass_pbh_mask, __mmask16, __mmask16, __m256bh, 13)
+test_2 (_mm_mask_fpclass_pbh_mask, __mmask8, __mmask8, __m128bh, 13)
+test_2 (_mm256_cmp_pbh_mask, __mmask16, __m256bh, __m256bh, 1)
+test_2 (_mm_cmp_pbh_mask, __mmask8, __m128bh, __m128bh, 1)
+test_3 (_mm256_mask_cmp_pbh_mask, __mmask16, __mmask16, __m256bh, __m256bh, 1)
+test_3 (_mm_mask_cmp_pbh_mask, __mmask8, __mmask8, __m128bh, __m128bh, 1)
@@ -1427,3 +1427,46 @@ test_2 (_mm256_cvtx_round2ps_ph, __m256h, __m256, __m256, 4)
/* avx10_2-512convertintrin.h */
test_2 (_mm512_cvtx_round2ps_ph, __m512h, __m512, __m512, 4)
+
+/* avx10_2-512bf16intrin.h */
+test_1 (_mm512_roundscalene_pbh, __m512bh, __m512bh, 123)
+test_2 (_mm512_maskz_roundscalene_pbh, __m512bh, __mmask32, __m512bh, 123)
+test_3 (_mm512_mask_roundscalene_pbh, __m512bh, __m512bh, __mmask32, __m512bh, 123)
+test_1 (_mm512_reducene_pbh, __m512bh, __m512bh, 123)
+test_2 (_mm512_maskz_reducene_pbh, __m512bh, __mmask32, __m512bh, 123)
+test_3 (_mm512_mask_reducene_pbh, __m512bh, __m512bh, __mmask32, __m512bh, 123)
+test_1x (_mm512_getmant_pbh, __m512bh, __m512bh, 1, 1)
+test_2x (_mm512_maskz_getmant_pbh, __m512bh, __mmask32,__m512bh, 1, 1)
+test_3x (_mm512_mask_getmant_pbh, __m512bh, __m512bh, __mmask32,__m512bh, 1, 1)
+test_1 (_mm512_fpclass_pbh_mask, __mmask32, __m512bh, 13)
+test_2 (_mm512_mask_fpclass_pbh_mask, __mmask32, __mmask32, __m512bh, 13)
+test_2 (_mm512_cmp_pbh_mask, __mmask32, __m512bh, __m512bh, 1)
+test_3 (_mm512_mask_cmp_pbh_mask, __mmask32, __mmask32,__m512bh, __m512bh, 1)
+
+/* avx10_2bf16intrin.h */
+test_1 (_mm256_roundscalene_pbh, __m256bh, __m256bh, 123)
+test_1 (_mm_roundscalene_pbh, __m128bh, __m128bh, 123)
+test_2 (_mm256_maskz_roundscalene_pbh, __m256bh, __mmask16, __m256bh, 123)
+test_2 (_mm_maskz_roundscalene_pbh, __m128bh, __mmask8, __m128bh, 123)
+test_3 (_mm256_mask_roundscalene_pbh, __m256bh, __m256bh, __mmask16, __m256bh, 123)
+test_3 (_mm_mask_roundscalene_pbh, __m128bh, __m128bh, __mmask8, __m128bh, 123)
+test_1 (_mm256_reducene_pbh, __m256bh, __m256bh, 123)
+test_1 (_mm_reducene_pbh, __m128bh, __m128bh, 123)
+test_2 (_mm256_maskz_reducene_pbh, __m256bh, __mmask16, __m256bh, 123)
+test_2 (_mm_maskz_reducene_pbh, __m128bh, __mmask8, __m128bh, 123)
+test_3 (_mm256_mask_reducene_pbh, __m256bh, __m256bh, __mmask16, __m256bh, 123)
+test_3 (_mm_mask_reducene_pbh, __m128bh, __m128bh, __mmask8, __m128bh, 123)
+test_1x (_mm256_getmant_pbh, __m256bh, __m256bh, 1, 1)
+test_1x (_mm_getmant_pbh, __m128bh, __m128bh, 1, 1)
+test_2x (_mm256_maskz_getmant_pbh, __m256bh, __mmask16,__m256bh, 1, 1)
+test_2x (_mm_maskz_getmant_pbh, __m128bh, __mmask8, __m128bh, 1, 1)
+test_3x (_mm256_mask_getmant_pbh, __m256bh, __m256bh, __mmask16,__m256bh, 1, 1)
+test_3x (_mm_mask_getmant_pbh, __m128bh, __m128bh, __mmask8, __m128bh, 1, 1)
+test_1 (_mm256_fpclass_pbh_mask, __mmask16, __m256bh, 13)
+test_1 (_mm_fpclass_pbh_mask, __mmask8, __m128bh, 13)
+test_2 (_mm256_mask_fpclass_pbh_mask, __mmask16, __mmask16, __m256bh, 13)
+test_2 (_mm_mask_fpclass_pbh_mask, __mmask8, __mmask8, __m128bh, 13)
+test_2 (_mm256_cmp_pbh_mask, __mmask16, __m256bh, __m256bh, 1)
+test_2 (_mm_cmp_pbh_mask, __mmask8, __m128bh, __m128bh, 1)
+test_3 (_mm256_mask_cmp_pbh_mask, __mmask16, __mmask16, __m256bh, __m256bh, 1)
+test_3 (_mm_mask_cmp_pbh_mask, __mmask8, __mmask8, __m128bh, __m128bh, 1)
@@ -998,6 +998,25 @@
/* avx10_2-512convertintrin.h */
#define __builtin_ia32_vcvt2ps2phx512_mask_round(A, B, C, D, E) __builtin_ia32_vcvt2ps2phx512_mask_round(A, B, C, D, 8)
+/* avx10_2-512bf16intrin.h */
+#define __builtin_ia32_rndscalenepbf16512_mask(A, B, C, D) __builtin_ia32_rndscalenepbf16512_mask(A, 123, C, D)
+#define __builtin_ia32_reducenepbf16512_mask(A, B, C, D) __builtin_ia32_reducenepbf16512_mask(A, 123, C, D)
+#define __builtin_ia32_getmantpbf16512_mask(A, B, C, D) __builtin_ia32_getmantpbf16512_mask(A, 1, C, D)
+#define __builtin_ia32_fpclasspbf16512_mask(A, B, C) __builtin_ia32_fpclasspbf16512_mask(A, 1, C)
+#define __builtin_ia32_cmppbf16512_mask(A, B, C, D) __builtin_ia32_cmppbf16512_mask(A, B, 1, D)
+
+/* avx10_2bf16intrin.h */
+#define __builtin_ia32_rndscalenepbf16256_mask(A, B, C, D) __builtin_ia32_rndscalenepbf16256_mask(A, 123, C, D)
+#define __builtin_ia32_rndscalenepbf16128_mask(A, B, C, D) __builtin_ia32_rndscalenepbf16128_mask(A, 123, C, D)
+#define __builtin_ia32_reducenepbf16256_mask(A, B, C, D) __builtin_ia32_reducenepbf16256_mask(A, 123, C, D)
+#define __builtin_ia32_reducenepbf16128_mask(A, B, C, D) __builtin_ia32_reducenepbf16128_mask(A, 123, C, D)
+#define __builtin_ia32_getmantpbf16256_mask(A, B, C, D) __builtin_ia32_getmantpbf16256_mask(A, 1, C, D)
+#define __builtin_ia32_getmantpbf16128_mask(A, B, C, D) __builtin_ia32_getmantpbf16128_mask(A, 1, C, D)
+#define __builtin_ia32_fpclasspbf16256_mask(A, B, C) __builtin_ia32_fpclasspbf16256_mask(A, 1, C)
+#define __builtin_ia32_fpclasspbf16128_mask(A, B, C) __builtin_ia32_fpclasspbf16128_mask(A, 1, C)
+#define __builtin_ia32_cmppbf16256_mask(A, B, C, D) __builtin_ia32_cmppbf16256_mask(A, B, 1, D)
+#define __builtin_ia32_cmppbf16128_mask(A, B, C, D) __builtin_ia32_cmppbf16128_mask(A, B, 1, D)
+
#pragma GCC target ("sse4a,3dnow,avx,avx2,fma4,xop,aes,pclmul,popcnt,abm,lzcnt,bmi,bmi2,tbm,lwp,fsgsbase,rdrnd,f16c,fma,rtm,rdseed,prfchw,adx,fxsr,xsaveopt,sha,xsavec,xsaves,clflushopt,clwb,mwaitx,clzero,pku,sgx,rdpid,gfni,vpclmulqdq,pconfig,wbnoinvd,enqcmd,avx512vp2intersect,serialize,tsxldtrk,amx-tile,amx-int8,amx-bf16,kl,widekl,avxvnni,avxifma,avxvnniint8,avxneconvert,cmpccxadd,amx-fp16,prefetchi,raoint,amx-complex,avxvnniint16,sm3,sha512,sm4,avx10.2-512")
#include <x86intrin.h>
From: konglin1 <lingling.kong@intel.com> gcc/ChangeLog: * config/i386/avx10_2-512bf16intrin.h: Add new intrinsics. * config/i386/avx10_2bf16intrin.h: Diito. * config/i386/i386-builtin-types.def : Add new DEF_FUNCTION_TYPE for new type. * config/i386/i386-builtin.def (BDESC): Add new buildin. * config/i386/i386-expand.cc (ix86_expand_args_builtin): Handle new type. * config/i386/sse.md (avx10_2_rsqrtpbf16_<mode><mask_name>): New define_insn. (avx10_2_sqrtnepbf16_<mode><mask_name>): Ditto. (avx10_2_rcppbf16_<mode><mask_name>): Ditto. (avx10_2_getexppbf16_<mode><mask_name>): Ditto. (BF16IMMOP): New iterator. (bf16immop): Ditto. (avx10_2_<bf16immop>pbf16_<mode><mask_name>): New define_insn. (avx10_2_fpclasspbf16_<mode><mask_scalar_merge_name>): Ditto. (avx10_2_cmppbf16_<mode><mask_scalar_merge_name>): Ditto. (avx10_2_comsbf16_v8bf): Ditto. gcc/testsuite/ChangeLog: * gcc.target/i386/avx10-helper.h: Add helper functions. * gcc.target/i386/avx10_2-512-bf16-1.c: Add new tests. * gcc.target/i386/avx10_2-bf16-1.c: Ditto. * gcc.target/i386/avx-1.c: Add macros. * gcc.target/i386/sse-13.c: Ditto. * gcc.target/i386/sse-14.c: Ditto. * gcc.target/i386/sse-22.c: Ditto. * gcc.target/i386/sse-23.c: Ditto. * gcc.target/i386/avx10_2-512-vcmppbf16-2.c: New test. * gcc.target/i386/avx10_2-512-vfpclasspbf16-2.c: Ditto. * gcc.target/i386/avx10_2-512-vgetexppbf16-2.c: Ditto. * gcc.target/i386/avx10_2-512-vgetmantpbf16-2.c: Ditto. * gcc.target/i386/avx10_2-512-vrcppbf16-2.c: Ditto. * gcc.target/i386/avx10_2-512-vreducenepbf16-2.c: Ditto. * gcc.target/i386/avx10_2-512-vrndscalenepbf16-2.c: Ditto. * gcc.target/i386/avx10_2-512-vrsqrtpbf16-2.c: Ditto. * gcc.target/i386/avx10_2-512-vsqrtnepbf16-2.c: Ditto. * gcc.target/i386/avx10_2-vcmppbf16-2.c: Ditto. * gcc.target/i386/avx10_2-vcomsbf16-1.c: Ditto. * gcc.target/i386/avx10_2-vcomsbf16-2.c: Ditto. * gcc.target/i386/avx10_2-vfpclasspbf16-2.c: Ditto. * gcc.target/i386/avx10_2-vgetexppbf16-2.c: Ditto. * gcc.target/i386/avx10_2-vgetmantpbf16-2.c: Ditto. * gcc.target/i386/avx10_2-vrcppbf16-2.c: Ditto. * gcc.target/i386/avx10_2-vreducenepbf16-2.c: Ditto. * gcc.target/i386/avx10_2-vrndscalenepbf16-2.c: Ditto. * gcc.target/i386/avx10_2-vrsqrtpbf16-2.c: Ditto. * gcc.target/i386/avx10_2-vsqrtnepbf16-2.c: Ditto. Co-authored-by: Levy Hsu <admin@levyhsu.com> --- gcc/config/i386/avx10_2-512bf16intrin.h | 317 +++++++++ gcc/config/i386/avx10_2bf16intrin.h | 650 ++++++++++++++++++ gcc/config/i386/i386-builtin-types.def | 10 + gcc/config/i386/i386-builtin.def | 33 + gcc/config/i386/i386-expand.cc | 16 + gcc/config/i386/sse.md | 92 +++ gcc/testsuite/gcc.target/i386/avx-1.c | 19 + gcc/testsuite/gcc.target/i386/avx10-check.h | 4 +- gcc/testsuite/gcc.target/i386/avx10-helper.h | 28 + .../gcc.target/i386/avx10_2-512-bf16-1.c | 58 ++ .../gcc.target/i386/avx10_2-512-vcmppbf16-2.c | 36 + .../i386/avx10_2-512-vfpclasspbf16-2.c | 44 ++ .../i386/avx10_2-512-vgetexppbf16-2.c | 47 ++ .../i386/avx10_2-512-vgetmantpbf16-2.c | 50 ++ .../gcc.target/i386/avx10_2-512-vrcppbf16-2.c | 45 ++ .../i386/avx10_2-512-vreducenepbf16-2.c | 50 ++ .../i386/avx10_2-512-vrndscalenepbf16-2.c | 46 ++ .../i386/avx10_2-512-vrsqrtpbf16-2.c | 47 ++ .../i386/avx10_2-512-vscalefpbf16-2.c | 2 +- .../i386/avx10_2-512-vsqrtnepbf16-2.c | 47 ++ .../gcc.target/i386/avx10_2-bf16-1.c | 114 +++ .../gcc.target/i386/avx10_2-vcmppbf16-2.c | 16 + .../gcc.target/i386/avx10_2-vcomsbf16-1.c | 19 + .../gcc.target/i386/avx10_2-vcomsbf16-2.c | 58 ++ .../gcc.target/i386/avx10_2-vfpclasspbf16-2.c | 16 + .../gcc.target/i386/avx10_2-vgetexppbf16-2.c | 16 + .../gcc.target/i386/avx10_2-vgetmantpbf16-2.c | 16 + .../gcc.target/i386/avx10_2-vrcppbf16-2.c | 16 + .../i386/avx10_2-vreducenepbf16-2.c | 16 + .../i386/avx10_2-vrndscalenepbf16-2.c | 16 + .../gcc.target/i386/avx10_2-vrsqrtpbf16-2.c | 16 + .../gcc.target/i386/avx10_2-vsqrtnepbf16-2.c | 16 + gcc/testsuite/gcc.target/i386/sse-13.c | 19 + gcc/testsuite/gcc.target/i386/sse-14.c | 43 ++ gcc/testsuite/gcc.target/i386/sse-22.c | 43 ++ gcc/testsuite/gcc.target/i386/sse-23.c | 19 + 36 files changed, 2097 insertions(+), 3 deletions(-) create mode 100644 gcc/testsuite/gcc.target/i386/avx10_2-512-vcmppbf16-2.c create mode 100644 gcc/testsuite/gcc.target/i386/avx10_2-512-vfpclasspbf16-2.c create mode 100644 gcc/testsuite/gcc.target/i386/avx10_2-512-vgetexppbf16-2.c create mode 100644 gcc/testsuite/gcc.target/i386/avx10_2-512-vgetmantpbf16-2.c create mode 100644 gcc/testsuite/gcc.target/i386/avx10_2-512-vrcppbf16-2.c create mode 100644 gcc/testsuite/gcc.target/i386/avx10_2-512-vreducenepbf16-2.c create mode 100644 gcc/testsuite/gcc.target/i386/avx10_2-512-vrndscalenepbf16-2.c create mode 100644 gcc/testsuite/gcc.target/i386/avx10_2-512-vrsqrtpbf16-2.c create mode 100644 gcc/testsuite/gcc.target/i386/avx10_2-512-vsqrtnepbf16-2.c create mode 100644 gcc/testsuite/gcc.target/i386/avx10_2-vcmppbf16-2.c create mode 100644 gcc/testsuite/gcc.target/i386/avx10_2-vcomsbf16-1.c create mode 100644 gcc/testsuite/gcc.target/i386/avx10_2-vcomsbf16-2.c create mode 100644 gcc/testsuite/gcc.target/i386/avx10_2-vfpclasspbf16-2.c create mode 100644 gcc/testsuite/gcc.target/i386/avx10_2-vgetexppbf16-2.c create mode 100644 gcc/testsuite/gcc.target/i386/avx10_2-vgetmantpbf16-2.c create mode 100644 gcc/testsuite/gcc.target/i386/avx10_2-vrcppbf16-2.c create mode 100644 gcc/testsuite/gcc.target/i386/avx10_2-vreducenepbf16-2.c create mode 100644 gcc/testsuite/gcc.target/i386/avx10_2-vrndscalenepbf16-2.c create mode 100644 gcc/testsuite/gcc.target/i386/avx10_2-vrsqrtpbf16-2.c create mode 100644 gcc/testsuite/gcc.target/i386/avx10_2-vsqrtnepbf16-2.c