From debe317dcf0802e8ce16ff56af45380ab7ce09d7 Mon Sep 17 00:00:00 2001 From: sayantn Date: Mon, 15 Jul 2024 16:01:06 +0530 Subject: [PATCH] AVX512FP16 Part 5: FP-Support `getexp`, `getmant`, `roundscale`, `scalef`, `reduce` --- .../stdarch/crates/core_arch/missing-x86.md | 90 - .../crates/core_arch/src/x86/avx512fp16.rs | 3309 ++++++++++++++++- 2 files changed, 3297 insertions(+), 102 deletions(-) diff --git a/library/stdarch/crates/core_arch/missing-x86.md b/library/stdarch/crates/core_arch/missing-x86.md index c0b8aa14572b..72fc8b840e27 100644 --- a/library/stdarch/crates/core_arch/missing-x86.md +++ b/library/stdarch/crates/core_arch/missing-x86.md @@ -103,10 +103,6 @@ * [ ] [`_mm512_cvtxph_ps`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm512_cvtxph_ps) * [ ] [`_mm512_cvtxps_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm512_cvtxps_ph) * [ ] [`_mm512_fpclass_ph_mask`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm512_fpclass_ph_mask) - * [ ] [`_mm512_getexp_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm512_getexp_ph) - * [ ] [`_mm512_getexp_round_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm512_getexp_round_ph) - * [ ] [`_mm512_getmant_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm512_getmant_ph) - * [ ] [`_mm512_getmant_round_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm512_getmant_round_ph) * [ ] [`_mm512_mask_blend_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm512_mask_blend_ph) * [ ] [`_mm512_mask_cmp_ph_mask`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm512_mask_cmp_ph_mask) * [ ] [`_mm512_mask_cmp_round_ph_mask`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm512_mask_cmp_round_ph_mask) @@ -155,16 +151,6 @@ * [ ] [`_mm512_mask_cvtxph_ps`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm512_mask_cvtxph_ps) * [ ] [`_mm512_mask_cvtxps_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm512_mask_cvtxps_ph) * [ ] [`_mm512_mask_fpclass_ph_mask`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm512_mask_fpclass_ph_mask) - * [ ] [`_mm512_mask_getexp_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm512_mask_getexp_ph) - * [ ] [`_mm512_mask_getexp_round_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm512_mask_getexp_round_ph) - * [ ] [`_mm512_mask_getmant_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm512_mask_getmant_ph) - * [ ] [`_mm512_mask_getmant_round_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm512_mask_getmant_round_ph) - * [ ] [`_mm512_mask_reduce_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm512_mask_reduce_ph) - * [ ] [`_mm512_mask_reduce_round_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm512_mask_reduce_round_ph) - * [ ] [`_mm512_mask_roundscale_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm512_mask_roundscale_ph) - * [ ] [`_mm512_mask_roundscale_round_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm512_mask_roundscale_round_ph) - * [ ] [`_mm512_mask_scalef_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm512_mask_scalef_ph) - * [ ] [`_mm512_mask_scalef_round_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm512_mask_scalef_round_ph) * [ ] [`_mm512_maskz_cvt_roundepi16_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm512_maskz_cvt_roundepi16_ph) * [ ] [`_mm512_maskz_cvt_roundepi32_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm512_maskz_cvt_roundepi32_ph) * [ ] [`_mm512_maskz_cvt_roundepi64_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm512_maskz_cvt_roundepi64_ph) @@ -209,28 +195,12 @@ * [ ] [`_mm512_maskz_cvtx_roundps_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm512_maskz_cvtx_roundps_ph) * [ ] [`_mm512_maskz_cvtxph_ps`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm512_maskz_cvtxph_ps) * [ ] [`_mm512_maskz_cvtxps_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm512_maskz_cvtxps_ph) - * [ ] [`_mm512_maskz_getexp_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm512_maskz_getexp_ph) - * [ ] [`_mm512_maskz_getexp_round_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm512_maskz_getexp_round_ph) - * [ ] [`_mm512_maskz_getmant_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm512_maskz_getmant_ph) - * [ ] [`_mm512_maskz_getmant_round_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm512_maskz_getmant_round_ph) - * [ ] [`_mm512_maskz_reduce_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm512_maskz_reduce_ph) - * [ ] [`_mm512_maskz_reduce_round_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm512_maskz_reduce_round_ph) - * [ ] [`_mm512_maskz_roundscale_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm512_maskz_roundscale_ph) - * [ ] [`_mm512_maskz_roundscale_round_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm512_maskz_roundscale_round_ph) - * [ ] [`_mm512_maskz_scalef_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm512_maskz_scalef_ph) - * [ ] [`_mm512_maskz_scalef_round_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm512_maskz_scalef_round_ph) * [ ] [`_mm512_permutex2var_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm512_permutex2var_ph) * [ ] [`_mm512_permutexvar_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm512_permutexvar_ph) * [ ] [`_mm512_reduce_add_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm512_reduce_add_ph) * [ ] [`_mm512_reduce_max_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm512_reduce_max_ph) * [ ] [`_mm512_reduce_min_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm512_reduce_min_ph) * [ ] [`_mm512_reduce_mul_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm512_reduce_mul_ph) - * [ ] [`_mm512_reduce_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm512_reduce_ph) - * [ ] [`_mm512_reduce_round_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm512_reduce_round_ph) - * [ ] [`_mm512_roundscale_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm512_roundscale_ph) - * [ ] [`_mm512_roundscale_round_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm512_roundscale_round_ph) - * [ ] [`_mm512_scalef_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm512_scalef_ph) - * [ ] [`_mm512_scalef_round_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm512_scalef_round_ph) * [ ] [`_mm512_set1_pch`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm512_set1_pch) * [ ] [`_mm_cvt_roundi32_sh`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_cvt_roundi32_sh) * [ ] [`_mm_cvt_roundi64_sh`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_cvt_roundi64_sh) @@ -268,10 +238,6 @@ * [ ] [`_mm_cvtu32_sh`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_cvtu32_sh) * [ ] [`_mm_cvtu64_sh`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_cvtu64_sh) * [ ] [`_mm_fpclass_sh_mask`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_fpclass_sh_mask) - * [ ] [`_mm_getexp_round_sh`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_getexp_round_sh) - * [ ] [`_mm_getexp_sh`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_getexp_sh) - * [ ] [`_mm_getmant_round_sh`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_getmant_round_sh) - * [ ] [`_mm_getmant_sh`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_getmant_sh) * [ ] [`_mm_mask_cvt_roundsd_sh`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_mask_cvt_roundsd_sh) * [ ] [`_mm_mask_cvt_roundsh_sd`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_mask_cvt_roundsh_sd) * [ ] [`_mm_mask_cvt_roundsh_ss`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_mask_cvt_roundsh_ss) @@ -281,16 +247,6 @@ * [ ] [`_mm_mask_cvtsh_ss`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_mask_cvtsh_ss) * [ ] [`_mm_mask_cvtss_sh`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_mask_cvtss_sh) * [ ] [`_mm_mask_fpclass_sh_mask`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_mask_fpclass_sh_mask) - * [ ] [`_mm_mask_getexp_round_sh`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_mask_getexp_round_sh) - * [ ] [`_mm_mask_getexp_sh`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_mask_getexp_sh) - * [ ] [`_mm_mask_getmant_round_sh`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_mask_getmant_round_sh) - * [ ] [`_mm_mask_getmant_sh`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_mask_getmant_sh) - * [ ] [`_mm_mask_reduce_round_sh`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_mask_reduce_round_sh) - * [ ] [`_mm_mask_reduce_sh`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_mask_reduce_sh) - * [ ] [`_mm_mask_roundscale_round_sh`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_mask_roundscale_round_sh) - * [ ] [`_mm_mask_roundscale_sh`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_mask_roundscale_sh) - * [ ] [`_mm_mask_scalef_round_sh`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_mask_scalef_round_sh) - * [ ] [`_mm_mask_scalef_sh`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_mask_scalef_sh) * [ ] [`_mm_maskz_cvt_roundsd_sh`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_maskz_cvt_roundsd_sh) * [ ] [`_mm_maskz_cvt_roundsh_sd`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_maskz_cvt_roundsh_sd) * [ ] [`_mm_maskz_cvt_roundsh_ss`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_maskz_cvt_roundsh_ss) @@ -299,22 +255,6 @@ * [ ] [`_mm_maskz_cvtsh_sd`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_maskz_cvtsh_sd) * [ ] [`_mm_maskz_cvtsh_ss`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_maskz_cvtsh_ss) * [ ] [`_mm_maskz_cvtss_sh`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_maskz_cvtss_sh) - * [ ] [`_mm_maskz_getexp_round_sh`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_maskz_getexp_round_sh) - * [ ] [`_mm_maskz_getexp_sh`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_maskz_getexp_sh) - * [ ] [`_mm_maskz_getmant_round_sh`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_maskz_getmant_round_sh) - * [ ] [`_mm_maskz_getmant_sh`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_maskz_getmant_sh) - * [ ] [`_mm_maskz_reduce_round_sh`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_maskz_reduce_round_sh) - * [ ] [`_mm_maskz_reduce_sh`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_maskz_reduce_sh) - * [ ] [`_mm_maskz_roundscale_round_sh`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_maskz_roundscale_round_sh) - * [ ] [`_mm_maskz_roundscale_sh`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_maskz_roundscale_sh) - * [ ] [`_mm_maskz_scalef_round_sh`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_maskz_scalef_round_sh) - * [ ] [`_mm_maskz_scalef_sh`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_maskz_scalef_sh) - * [ ] [`_mm_reduce_round_sh`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_reduce_round_sh) - * [ ] [`_mm_reduce_sh`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_reduce_sh) - * [ ] [`_mm_roundscale_round_sh`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_roundscale_round_sh) - * [ ] [`_mm_roundscale_sh`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_roundscale_sh) - * [ ] [`_mm_scalef_round_sh`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_scalef_round_sh) - * [ ] [`_mm_scalef_sh`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_scalef_sh) * [ ] [`_mm_set1_pch`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_set1_pch)

@@ -345,8 +285,6 @@ * [ ] [`_mm256_cvtxph_ps`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm256_cvtxph_ps) * [ ] [`_mm256_cvtxps_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm256_cvtxps_ph) * [ ] [`_mm256_fpclass_ph_mask`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm256_fpclass_ph_mask) - * [ ] [`_mm256_getexp_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm256_getexp_ph) - * [ ] [`_mm256_getmant_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm256_getmant_ph) * [ ] [`_mm256_mask_blend_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm256_mask_blend_ph) * [ ] [`_mm256_mask_cmp_ph_mask`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm256_mask_cmp_ph_mask) * [ ] [`_mm256_mask_cvtepi16_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm256_mask_cvtepi16_ph) @@ -372,11 +310,6 @@ * [ ] [`_mm256_mask_cvtxph_ps`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm256_mask_cvtxph_ps) * [ ] [`_mm256_mask_cvtxps_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm256_mask_cvtxps_ph) * [ ] [`_mm256_mask_fpclass_ph_mask`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm256_mask_fpclass_ph_mask) - * [ ] [`_mm256_mask_getexp_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm256_mask_getexp_ph) - * [ ] [`_mm256_mask_getmant_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm256_mask_getmant_ph) - * [ ] [`_mm256_mask_reduce_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm256_mask_reduce_ph) - * [ ] [`_mm256_mask_roundscale_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm256_mask_roundscale_ph) - * [ ] [`_mm256_mask_scalef_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm256_mask_scalef_ph) * [ ] [`_mm256_maskz_cvtepi16_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm256_maskz_cvtepi16_ph) * [ ] [`_mm256_maskz_cvtepi32_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm256_maskz_cvtepi32_ph) * [ ] [`_mm256_maskz_cvtepi64_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm256_maskz_cvtepi64_ph) @@ -399,20 +332,12 @@ * [ ] [`_mm256_maskz_cvttph_epu64`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm256_maskz_cvttph_epu64) * [ ] [`_mm256_maskz_cvtxph_ps`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm256_maskz_cvtxph_ps) * [ ] [`_mm256_maskz_cvtxps_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm256_maskz_cvtxps_ph) - * [ ] [`_mm256_maskz_getexp_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm256_maskz_getexp_ph) - * [ ] [`_mm256_maskz_getmant_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm256_maskz_getmant_ph) - * [ ] [`_mm256_maskz_reduce_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm256_maskz_reduce_ph) - * [ ] [`_mm256_maskz_roundscale_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm256_maskz_roundscale_ph) - * [ ] [`_mm256_maskz_scalef_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm256_maskz_scalef_ph) * [ ] [`_mm256_permutex2var_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm256_permutex2var_ph) * [ ] [`_mm256_permutexvar_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm256_permutexvar_ph) * [ ] [`_mm256_reduce_add_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm256_reduce_add_ph) * [ ] [`_mm256_reduce_max_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm256_reduce_max_ph) * [ ] [`_mm256_reduce_min_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm256_reduce_min_ph) * [ ] [`_mm256_reduce_mul_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm256_reduce_mul_ph) - * [ ] [`_mm256_reduce_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm256_reduce_ph) - * [ ] [`_mm256_roundscale_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm256_roundscale_ph) - * [ ] [`_mm256_scalef_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm256_scalef_ph) * [ ] [`_mm_cmp_ph_mask`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_cmp_ph_mask) * [ ] [`_mm_cvtepi16_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_cvtepi16_ph) * [ ] [`_mm_cvtepi32_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_cvtepi32_ph) @@ -437,8 +362,6 @@ * [ ] [`_mm_cvtxph_ps`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_cvtxph_ps) * [ ] [`_mm_cvtxps_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_cvtxps_ph) * [ ] [`_mm_fpclass_ph_mask`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_fpclass_ph_mask) - * [ ] [`_mm_getexp_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_getexp_ph) - * [ ] [`_mm_getmant_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_getmant_ph) * [ ] [`_mm_mask_blend_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_mask_blend_ph) * [ ] [`_mm_mask_cmp_ph_mask`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_mask_cmp_ph_mask) * [ ] [`_mm_mask_cvtepi16_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_mask_cvtepi16_ph) @@ -464,11 +387,6 @@ * [ ] [`_mm_mask_cvtxph_ps`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_mask_cvtxph_ps) * [ ] [`_mm_mask_cvtxps_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_mask_cvtxps_ph) * [ ] [`_mm_mask_fpclass_ph_mask`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_mask_fpclass_ph_mask) - * [ ] [`_mm_mask_getexp_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_mask_getexp_ph) - * [ ] [`_mm_mask_getmant_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_mask_getmant_ph) - * [ ] [`_mm_mask_reduce_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_mask_reduce_ph) - * [ ] [`_mm_mask_roundscale_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_mask_roundscale_ph) - * [ ] [`_mm_mask_scalef_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_mask_scalef_ph) * [ ] [`_mm_maskz_cvtepi16_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_maskz_cvtepi16_ph) * [ ] [`_mm_maskz_cvtepi32_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_maskz_cvtepi32_ph) * [ ] [`_mm_maskz_cvtepi64_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_maskz_cvtepi64_ph) @@ -491,20 +409,12 @@ * [ ] [`_mm_maskz_cvttph_epu64`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_maskz_cvttph_epu64) * [ ] [`_mm_maskz_cvtxph_ps`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_maskz_cvtxph_ps) * [ ] [`_mm_maskz_cvtxps_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_maskz_cvtxps_ph) - * [ ] [`_mm_maskz_getexp_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_maskz_getexp_ph) - * [ ] [`_mm_maskz_getmant_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_maskz_getmant_ph) - * [ ] [`_mm_maskz_reduce_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_maskz_reduce_ph) - * [ ] [`_mm_maskz_roundscale_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_maskz_roundscale_ph) - * [ ] [`_mm_maskz_scalef_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_maskz_scalef_ph) * [ ] [`_mm_permutex2var_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_permutex2var_ph) * [ ] [`_mm_permutexvar_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_permutexvar_ph) * [ ] [`_mm_reduce_add_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_reduce_add_ph) * [ ] [`_mm_reduce_max_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_reduce_max_ph) * [ ] [`_mm_reduce_min_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_reduce_min_ph) * [ ] [`_mm_reduce_mul_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_reduce_mul_ph) - * [ ] [`_mm_reduce_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_reduce_ph) - * [ ] [`_mm_roundscale_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_roundscale_ph) - * [ ] [`_mm_scalef_ph`](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_scalef_ph)

diff --git a/library/stdarch/crates/core_arch/src/x86/avx512fp16.rs b/library/stdarch/crates/core_arch/src/x86/avx512fp16.rs index b30bc63ed4d3..3c04d9ae9081 100644 --- a/library/stdarch/crates/core_arch/src/x86/avx512fp16.rs +++ b/library/stdarch/crates/core_arch/src/x86/avx512fp16.rs @@ -624,12 +624,13 @@ pub unsafe fn _mm512_zextph128_ph512(a: __m128h) -> __m512h { #[target_feature(enable = "avx512fp16")] #[rustc_legacy_const_generics(2, 3)] #[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] -pub unsafe fn _mm_cmp_round_sh_mask( +pub unsafe fn _mm_cmp_round_sh_mask( a: __m128h, b: __m128h, ) -> __mmask8 { + static_assert_uimm_bits!(IMM5, 5); static_assert_sae!(SAE); - _mm_mask_cmp_round_sh_mask::(0xff, a, b) + _mm_mask_cmp_round_sh_mask::(0xff, a, b) } /// Compare the lower half-precision (16-bit) floating-point elements in a and b based on the comparison @@ -641,13 +642,14 @@ pub unsafe fn _mm_cmp_round_sh_mask( #[target_feature(enable = "avx512fp16")] #[rustc_legacy_const_generics(3, 4)] #[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] -pub unsafe fn _mm_mask_cmp_round_sh_mask( +pub unsafe fn _mm_mask_cmp_round_sh_mask( k1: __mmask8, a: __m128h, b: __m128h, ) -> __mmask8 { + static_assert_uimm_bits!(IMM5, 5); static_assert_sae!(SAE); - vcmpsh(a, b, IMM8, k1, SAE) + vcmpsh(a, b, IMM5, k1, SAE) } /// Compare the lower half-precision (16-bit) floating-point elements in a and b based on the comparison @@ -658,8 +660,9 @@ pub unsafe fn _mm_mask_cmp_round_sh_mask( #[target_feature(enable = "avx512fp16")] #[rustc_legacy_const_generics(2)] #[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] -pub unsafe fn _mm_cmp_sh_mask(a: __m128h, b: __m128h) -> __mmask8 { - _mm_cmp_round_sh_mask::(a, b) +pub unsafe fn _mm_cmp_sh_mask(a: __m128h, b: __m128h) -> __mmask8 { + static_assert_uimm_bits!(IMM5, 5); + _mm_cmp_round_sh_mask::(a, b) } /// Compare the lower half-precision (16-bit) floating-point elements in a and b based on the comparison @@ -670,12 +673,13 @@ pub unsafe fn _mm_cmp_sh_mask(a: __m128h, b: __m128h) -> __mmas #[target_feature(enable = "avx512fp16")] #[rustc_legacy_const_generics(3)] #[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] -pub unsafe fn _mm_mask_cmp_sh_mask( +pub unsafe fn _mm_mask_cmp_sh_mask( k1: __mmask8, a: __m128h, b: __m128h, ) -> __mmask8 { - _mm_mask_cmp_round_sh_mask::(k1, a, b) + static_assert_uimm_bits!(IMM5, 5); + _mm_mask_cmp_round_sh_mask::(k1, a, b) } /// Cast vector of type `__m256h` to type `__m512h`. The upper 16 elements of the result are zeroed. @@ -706,9 +710,10 @@ pub unsafe fn _mm512_zextph256_ph512(a: __m256h) -> __m512h { #[target_feature(enable = "avx512fp16")] #[rustc_legacy_const_generics(2, 3)] #[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] -pub unsafe fn _mm_comi_round_sh(a: __m128h, b: __m128h) -> i32 { +pub unsafe fn _mm_comi_round_sh(a: __m128h, b: __m128h) -> i32 { + static_assert_uimm_bits!(IMM5, 5); static_assert_sae!(SAE); - vcomish(a, b, IMM8, SAE) + vcomish(a, b, IMM5, SAE) } /// Compare the lower half-precision (16-bit) floating-point elements in a and b based on the comparison @@ -719,8 +724,9 @@ pub unsafe fn _mm_comi_round_sh(a: __m128h, b: #[target_feature(enable = "avx512fp16")] #[rustc_legacy_const_generics(2)] #[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] -pub unsafe fn _mm_comi_sh(a: __m128h, b: __m128h) -> i32 { - _mm_comi_round_sh::(a, b) +pub unsafe fn _mm_comi_sh(a: __m128h, b: __m128h) -> i32 { + static_assert_uimm_bits!(IMM5, 5); + _mm_comi_round_sh::(a, b) } /// Compare the lower half-precision (16-bit) floating-point elements in a and b for equality, and return @@ -8440,6 +8446,2199 @@ pub unsafe fn _mm_maskz_min_round_sh( _mm_mask_min_round_sh::(_mm_setzero_ph(), k, a, b) } +/// Convert the exponent of each packed half-precision (16-bit) floating-point element in a to a half-precision +/// (16-bit) floating-point number representing the integer exponent, and store the results in dst. +/// This intrinsic essentially calculates `floor(log2(x))` for each element. +/// +/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_getexp_ph) +#[inline] +#[target_feature(enable = "avx512fp16,avx512vl")] +#[cfg_attr(test, assert_instr(vgetexpph))] +#[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] +pub unsafe fn _mm_getexp_ph(a: __m128h) -> __m128h { + _mm_mask_getexp_ph(_mm_undefined_ph(), 0xff, a) +} + +/// Convert the exponent of each packed half-precision (16-bit) floating-point element in a to a half-precision +/// (16-bit) floating-point number representing the integer exponent, and store the results in dst using writemask k +/// (elements are copied from src when the corresponding mask bit is not set). This intrinsic essentially calculates +/// `floor(log2(x))` for each element. +/// +/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_mask_getexp_ph) +#[inline] +#[target_feature(enable = "avx512fp16,avx512vl")] +#[cfg_attr(test, assert_instr(vgetexpph))] +#[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] +pub unsafe fn _mm_mask_getexp_ph(src: __m128h, k: __mmask8, a: __m128h) -> __m128h { + vgetexpph_128(a, src, k) +} + +/// Convert the exponent of each packed half-precision (16-bit) floating-point element in a to a half-precision +/// (16-bit) floating-point number representing the integer exponent, and store the results in dst using zeromask +/// k (elements are zeroed out when the corresponding mask bit is not set). This intrinsic essentially calculates +/// `floor(log2(x))` for each element. +/// +/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_maskz_getexp_ph) +#[inline] +#[target_feature(enable = "avx512fp16,avx512vl")] +#[cfg_attr(test, assert_instr(vgetexpph))] +#[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] +pub unsafe fn _mm_maskz_getexp_ph(k: __mmask8, a: __m128h) -> __m128h { + _mm_mask_getexp_ph(_mm_setzero_ph(), k, a) +} + +/// Convert the exponent of each packed half-precision (16-bit) floating-point element in a to a half-precision +/// (16-bit) floating-point number representing the integer exponent, and store the results in dst. +/// This intrinsic essentially calculates `floor(log2(x))` for each element. +/// +/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm256_getexp_ph) +#[inline] +#[target_feature(enable = "avx512fp16,avx512vl")] +#[cfg_attr(test, assert_instr(vgetexpph))] +#[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] +pub unsafe fn _mm256_getexp_ph(a: __m256h) -> __m256h { + _mm256_mask_getexp_ph(_mm256_undefined_ph(), 0xffff, a) +} + +/// Convert the exponent of each packed half-precision (16-bit) floating-point element in a to a half-precision +/// (16-bit) floating-point number representing the integer exponent, and store the results in dst using writemask k +/// (elements are copied from src when the corresponding mask bit is not set). This intrinsic essentially calculates +/// `floor(log2(x))` for each element. +/// +/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm256_mask_getexp_ph) +#[inline] +#[target_feature(enable = "avx512fp16,avx512vl")] +#[cfg_attr(test, assert_instr(vgetexpph))] +#[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] +pub unsafe fn _mm256_mask_getexp_ph(src: __m256h, k: __mmask16, a: __m256h) -> __m256h { + vgetexpph_256(a, src, k) +} + +/// Convert the exponent of each packed half-precision (16-bit) floating-point element in a to a half-precision +/// (16-bit) floating-point number representing the integer exponent, and store the results in dst using zeromask +/// k (elements are zeroed out when the corresponding mask bit is not set). This intrinsic essentially calculates +/// `floor(log2(x))` for each element. +/// +/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm256_maskz_getexp_ph) +#[inline] +#[target_feature(enable = "avx512fp16,avx512vl")] +#[cfg_attr(test, assert_instr(vgetexpph))] +#[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] +pub unsafe fn _mm256_maskz_getexp_ph(k: __mmask16, a: __m256h) -> __m256h { + _mm256_mask_getexp_ph(_mm256_setzero_ph(), k, a) +} + +/// Convert the exponent of each packed half-precision (16-bit) floating-point element in a to a half-precision +/// (16-bit) floating-point number representing the integer exponent, and store the results in dst. +/// This intrinsic essentially calculates `floor(log2(x))` for each element. +/// +/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm512_getexp_ph) +#[inline] +#[target_feature(enable = "avx512fp16")] +#[cfg_attr(test, assert_instr(vgetexpph))] +#[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] +pub unsafe fn _mm512_getexp_ph(a: __m512h) -> __m512h { + _mm512_mask_getexp_ph(_mm512_undefined_ph(), 0xffffffff, a) +} + +/// Convert the exponent of each packed half-precision (16-bit) floating-point element in a to a half-precision +/// (16-bit) floating-point number representing the integer exponent, and store the results in dst using writemask k +/// (elements are copied from src when the corresponding mask bit is not set). This intrinsic essentially calculates +/// `floor(log2(x))` for each element. +/// +/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm512_mask_getexp_ph) +#[inline] +#[target_feature(enable = "avx512fp16")] +#[cfg_attr(test, assert_instr(vgetexpph))] +#[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] +pub unsafe fn _mm512_mask_getexp_ph(src: __m512h, k: __mmask32, a: __m512h) -> __m512h { + _mm512_mask_getexp_round_ph::<_MM_FROUND_CUR_DIRECTION>(src, k, a) +} + +/// Convert the exponent of each packed half-precision (16-bit) floating-point element in a to a half-precision +/// (16-bit) floating-point number representing the integer exponent, and store the results in dst using zeromask +/// k (elements are zeroed out when the corresponding mask bit is not set). This intrinsic essentially calculates +/// `floor(log2(x))` for each element. +/// +/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm512_maskz_getexp_ph) +#[inline] +#[target_feature(enable = "avx512fp16")] +#[cfg_attr(test, assert_instr(vgetexpph))] +#[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] +pub unsafe fn _mm512_maskz_getexp_ph(k: __mmask32, a: __m512h) -> __m512h { + _mm512_mask_getexp_ph(_mm512_setzero_ph(), k, a) +} + +/// Convert the exponent of each packed half-precision (16-bit) floating-point element in a to a half-precision +/// (16-bit) floating-point number representing the integer exponent, and store the results in dst. +/// This intrinsic essentially calculates `floor(log2(x))` for each element. Exceptions can be suppressed +/// by passing _MM_FROUND_NO_EXC in the sae parameter +/// +/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm512_getexp_round_ph) +#[inline] +#[target_feature(enable = "avx512fp16")] +#[cfg_attr(test, assert_instr(vgetexpph, SAE = 8))] +#[rustc_legacy_const_generics(1)] +#[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] +pub unsafe fn _mm512_getexp_round_ph(a: __m512h) -> __m512h { + static_assert_sae!(SAE); + _mm512_mask_getexp_round_ph::(_mm512_undefined_ph(), 0xffffffff, a) +} + +/// Convert the exponent of each packed half-precision (16-bit) floating-point element in a to a half-precision +/// (16-bit) floating-point number representing the integer exponent, and store the results in dst using writemask k +/// (elements are copied from src when the corresponding mask bit is not set). This intrinsic essentially calculates +/// `floor(log2(x))` for each element. Exceptions can be suppressed by passing _MM_FROUND_NO_EXC in the sae parameter +/// +/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm512_mask_getexp_round_ph) +#[inline] +#[target_feature(enable = "avx512fp16")] +#[cfg_attr(test, assert_instr(vgetexpph, SAE = 8))] +#[rustc_legacy_const_generics(3)] +#[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] +pub unsafe fn _mm512_mask_getexp_round_ph( + src: __m512h, + k: __mmask32, + a: __m512h, +) -> __m512h { + static_assert_sae!(SAE); + vgetexpph_512(a, src, k, SAE) +} + +/// Convert the exponent of each packed half-precision (16-bit) floating-point element in a to a half-precision +/// (16-bit) floating-point number representing the integer exponent, and store the results in dst using zeromask +/// k (elements are zeroed out when the corresponding mask bit is not set). This intrinsic essentially calculates +/// `floor(log2(x))` for each element. Exceptions can be suppressed by passing _MM_FROUND_NO_EXC in the sae parameter +/// +/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm512_maskz_getexp_round_ph) +#[inline] +#[target_feature(enable = "avx512fp16")] +#[cfg_attr(test, assert_instr(vgetexpph, SAE = 8))] +#[rustc_legacy_const_generics(2)] +#[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] +pub unsafe fn _mm512_maskz_getexp_round_ph(k: __mmask32, a: __m512h) -> __m512h { + static_assert_sae!(SAE); + _mm512_mask_getexp_round_ph::(_mm512_setzero_ph(), k, a) +} + +/// Convert the exponent of the lower half-precision (16-bit) floating-point element in b to a half-precision +/// (16-bit) floating-point number representing the integer exponent, store the result in the lower element +/// of dst, and copy the upper 7 packed elements from a to the upper elements of dst. This intrinsic essentially +/// calculates `floor(log2(x))` for the lower element. +/// +/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_getexp_sh) +#[inline] +#[target_feature(enable = "avx512fp16")] +#[cfg_attr(test, assert_instr(vgetexpsh))] +#[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] +pub unsafe fn _mm_getexp_sh(a: __m128h, b: __m128h) -> __m128h { + _mm_mask_getexp_sh(_mm_undefined_ph(), 0xff, a, b) +} + +/// Convert the exponent of the lower half-precision (16-bit) floating-point element in b to a half-precision +/// (16-bit) floating-point number representing the integer exponent, store the result in the lower element +/// of dst using writemask k (the element is copied from src when mask bit 0 is not set), and copy the upper 7 +/// packed elements from a to the upper elements of dst. This intrinsic essentially calculates `floor(log2(x))` +/// for the lower element. +/// +/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_mask_getexp_sh) +#[inline] +#[target_feature(enable = "avx512fp16")] +#[cfg_attr(test, assert_instr(vgetexpsh))] +#[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] +pub unsafe fn _mm_mask_getexp_sh(src: __m128h, k: __mmask8, a: __m128h, b: __m128h) -> __m128h { + _mm_mask_getexp_round_sh::<_MM_FROUND_CUR_DIRECTION>(src, k, a, b) +} + +/// Convert the exponent of the lower half-precision (16-bit) floating-point element in b to a half-precision +/// (16-bit) floating-point number representing the integer exponent, store the result in the lower element +/// of dst using zeromask k (the element is zeroed out when mask bit 0 is not set), and copy the upper 7 packed +/// elements from a to the upper elements of dst. This intrinsic essentially calculates `floor(log2(x))` for the +/// lower element. +/// +/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_maskz_getexp_sh) +#[inline] +#[target_feature(enable = "avx512fp16")] +#[cfg_attr(test, assert_instr(vgetexpsh))] +#[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] +pub unsafe fn _mm_maskz_getexp_sh(k: __mmask8, a: __m128h, b: __m128h) -> __m128h { + _mm_mask_getexp_sh(_mm_setzero_ph(), k, a, b) +} + +/// Convert the exponent of the lower half-precision (16-bit) floating-point element in b to a half-precision +/// (16-bit) floating-point number representing the integer exponent, store the result in the lower element +/// of dst, and copy the upper 7 packed elements from a to the upper elements of dst. This intrinsic essentially +/// calculates `floor(log2(x))` for the lower element. Exceptions can be suppressed by passing _MM_FROUND_NO_EXC +/// in the sae parameter +/// +/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_getexp_round_sh) +#[inline] +#[target_feature(enable = "avx512fp16")] +#[cfg_attr(test, assert_instr(vgetexpsh, SAE = 8))] +#[rustc_legacy_const_generics(2)] +#[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] +pub unsafe fn _mm_getexp_round_sh(a: __m128h, b: __m128h) -> __m128h { + static_assert_sae!(SAE); + _mm_mask_getexp_round_sh::(_mm_undefined_ph(), 0xff, a, b) +} + +/// Convert the exponent of the lower half-precision (16-bit) floating-point element in b to a half-precision +/// (16-bit) floating-point number representing the integer exponent, store the result in the lower element +/// of dst using writemask k (the element is copied from src when mask bit 0 is not set), and copy the upper 7 +/// packed elements from a to the upper elements of dst. This intrinsic essentially calculates `floor(log2(x))` +/// for the lower element. Exceptions can be suppressed by passing _MM_FROUND_NO_EXC in the sae parameter +/// +/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_mask_getexp_round_sh) +#[inline] +#[target_feature(enable = "avx512fp16")] +#[cfg_attr(test, assert_instr(vgetexpsh, SAE = 8))] +#[rustc_legacy_const_generics(4)] +#[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] +pub unsafe fn _mm_mask_getexp_round_sh( + src: __m128h, + k: __mmask8, + a: __m128h, + b: __m128h, +) -> __m128h { + static_assert_sae!(SAE); + vgetexpsh(a, b, src, k, SAE) +} + +/// Convert the exponent of the lower half-precision (16-bit) floating-point element in b to a half-precision +/// (16-bit) floating-point number representing the integer exponent, store the result in the lower element +/// of dst using zeromask k (the element is zeroed out when mask bit 0 is not set), and copy the upper 7 packed +/// elements from a to the upper elements of dst. This intrinsic essentially calculates `floor(log2(x))` for the +/// lower element. Exceptions can be suppressed by passing _MM_FROUND_NO_EXC in the sae parameter +/// +/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_maskz_getexp_round_sh) +#[inline] +#[target_feature(enable = "avx512fp16")] +#[cfg_attr(test, assert_instr(vgetexpsh, SAE = 8))] +#[rustc_legacy_const_generics(3)] +#[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] +pub unsafe fn _mm_maskz_getexp_round_sh( + k: __mmask8, + a: __m128h, + b: __m128h, +) -> __m128h { + static_assert_sae!(SAE); + _mm_mask_getexp_round_sh::(_mm_setzero_ph(), k, a, b) +} + +/// Normalize the mantissas of packed half-precision (16-bit) floating-point elements in a, and store +/// the results in dst. This intrinsic essentially calculates `±(2^k)*|x.significand|`, where k depends +/// on the interval range defined by norm and the sign depends on sign and the source sign. +/// +/// The mantissa is normalized to the interval specified by interv, which can take the following values: +/// +/// _MM_MANT_NORM_1_2 // interval [1, 2) +/// _MM_MANT_NORM_p5_2 // interval [0.5, 2) +/// _MM_MANT_NORM_p5_1 // interval [0.5, 1) +/// _MM_MANT_NORM_p75_1p5 // interval [0.75, 1.5) +/// +/// The sign is determined by sc which can take the following values: +/// +/// _MM_MANT_SIGN_src // sign = sign(src) +/// _MM_MANT_SIGN_zero // sign = 0 +/// _MM_MANT_SIGN_nan // dst = NaN if sign(src) = 1 +/// +/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_getmant_ph) +#[inline] +#[target_feature(enable = "avx512fp16,avx512vl")] +#[cfg_attr(test, assert_instr(vgetmantph, NORM = 0, SIGN = 0))] +#[rustc_legacy_const_generics(1, 2)] +#[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] +pub unsafe fn _mm_getmant_ph< + const NORM: _MM_MANTISSA_NORM_ENUM, + const SIGN: _MM_MANTISSA_SIGN_ENUM, +>( + a: __m128h, +) -> __m128h { + static_assert_uimm_bits!(NORM, 4); + static_assert_uimm_bits!(SIGN, 2); + _mm_mask_getmant_ph::(_mm_undefined_ph(), 0xff, a) +} + +/// Normalize the mantissas of packed half-precision (16-bit) floating-point elements in a, and store +/// the results in dst using writemask k (elements are copied from src when the corresponding mask bit is not set). +/// This intrinsic essentially calculates `±(2^k)*|x.significand|`, where k depends on the interval range defined +/// by norm and the sign depends on sign and the source sign. +/// +/// The mantissa is normalized to the interval specified by interv, which can take the following values: +/// +/// _MM_MANT_NORM_1_2 // interval [1, 2) +/// _MM_MANT_NORM_p5_2 // interval [0.5, 2) +/// _MM_MANT_NORM_p5_1 // interval [0.5, 1) +/// _MM_MANT_NORM_p75_1p5 // interval [0.75, 1.5) +/// +/// The sign is determined by sc which can take the following values: +/// +/// _MM_MANT_SIGN_src // sign = sign(src) +/// _MM_MANT_SIGN_zero // sign = 0 +/// _MM_MANT_SIGN_nan // dst = NaN if sign(src) = 1 +/// +/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_mask_getmant_ph) +#[inline] +#[target_feature(enable = "avx512fp16,avx512vl")] +#[cfg_attr(test, assert_instr(vgetmantph, NORM = 0, SIGN = 0))] +#[rustc_legacy_const_generics(3, 4)] +#[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] +pub unsafe fn _mm_mask_getmant_ph< + const NORM: _MM_MANTISSA_NORM_ENUM, + const SIGN: _MM_MANTISSA_SIGN_ENUM, +>( + src: __m128h, + k: __mmask8, + a: __m128h, +) -> __m128h { + static_assert_uimm_bits!(NORM, 4); + static_assert_uimm_bits!(SIGN, 2); + vgetmantph_128(a, (SIGN << 2) | NORM, src, k) +} + +/// Normalize the mantissas of packed half-precision (16-bit) floating-point elements in a, and store +/// the results in dst using zeromask k (elements are zeroed out when the corresponding mask bit is not set). +/// This intrinsic essentially calculates `±(2^k)*|x.significand|`, where k depends on the interval range defined +/// by norm and the sign depends on sign and the source sign. +/// +/// The mantissa is normalized to the interval specified by interv, which can take the following values: +/// +/// _MM_MANT_NORM_1_2 // interval [1, 2) +/// _MM_MANT_NORM_p5_2 // interval [0.5, 2) +/// _MM_MANT_NORM_p5_1 // interval [0.5, 1) +/// _MM_MANT_NORM_p75_1p5 // interval [0.75, 1.5) +/// +/// The sign is determined by sc which can take the following values: +/// +/// _MM_MANT_SIGN_src // sign = sign(src) +/// _MM_MANT_SIGN_zero // sign = 0 +/// _MM_MANT_SIGN_nan // dst = NaN if sign(src) = 1 +/// +/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_maskz_getmant_ph) +#[inline] +#[target_feature(enable = "avx512fp16,avx512vl")] +#[cfg_attr(test, assert_instr(vgetmantph, NORM = 0, SIGN = 0))] +#[rustc_legacy_const_generics(2, 3)] +#[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] +pub unsafe fn _mm_maskz_getmant_ph< + const NORM: _MM_MANTISSA_NORM_ENUM, + const SIGN: _MM_MANTISSA_SIGN_ENUM, +>( + k: __mmask8, + a: __m128h, +) -> __m128h { + static_assert_uimm_bits!(NORM, 4); + static_assert_uimm_bits!(SIGN, 2); + _mm_mask_getmant_ph::(_mm_setzero_ph(), k, a) +} + +/// Normalize the mantissas of packed half-precision (16-bit) floating-point elements in a, and store +/// the results in dst. This intrinsic essentially calculates `±(2^k)*|x.significand|`, where k depends +/// on the interval range defined by norm and the sign depends on sign and the source sign. +/// +/// The mantissa is normalized to the interval specified by interv, which can take the following values: +/// +/// _MM_MANT_NORM_1_2 // interval [1, 2) +/// _MM_MANT_NORM_p5_2 // interval [0.5, 2) +/// _MM_MANT_NORM_p5_1 // interval [0.5, 1) +/// _MM_MANT_NORM_p75_1p5 // interval [0.75, 1.5) +/// +/// The sign is determined by sc which can take the following values: +/// +/// _MM_MANT_SIGN_src // sign = sign(src) +/// _MM_MANT_SIGN_zero // sign = 0 +/// _MM_MANT_SIGN_nan // dst = NaN if sign(src) = 1 +/// +/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm256_getmant_ph) +#[inline] +#[target_feature(enable = "avx512fp16,avx512vl")] +#[cfg_attr(test, assert_instr(vgetmantph, NORM = 0, SIGN = 0))] +#[rustc_legacy_const_generics(1, 2)] +#[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] +pub unsafe fn _mm256_getmant_ph< + const NORM: _MM_MANTISSA_NORM_ENUM, + const SIGN: _MM_MANTISSA_SIGN_ENUM, +>( + a: __m256h, +) -> __m256h { + static_assert_uimm_bits!(NORM, 4); + static_assert_uimm_bits!(SIGN, 2); + _mm256_mask_getmant_ph::(_mm256_undefined_ph(), 0xffff, a) +} + +/// Normalize the mantissas of packed half-precision (16-bit) floating-point elements in a, and store +/// the results in dst using writemask k (elements are copied from src when the corresponding mask bit is not set). +/// This intrinsic essentially calculates `±(2^k)*|x.significand|`, where k depends on the interval range defined +/// by norm and the sign depends on sign and the source sign. +/// +/// The mantissa is normalized to the interval specified by interv, which can take the following values: +/// +/// _MM_MANT_NORM_1_2 // interval [1, 2) +/// _MM_MANT_NORM_p5_2 // interval [0.5, 2) +/// _MM_MANT_NORM_p5_1 // interval [0.5, 1) +/// _MM_MANT_NORM_p75_1p5 // interval [0.75, 1.5) +/// +/// The sign is determined by sc which can take the following values: +/// +/// _MM_MANT_SIGN_src // sign = sign(src) +/// _MM_MANT_SIGN_zero // sign = 0 +/// _MM_MANT_SIGN_nan // dst = NaN if sign(src) = 1 +/// +/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm256_mask_getmant_ph) +#[inline] +#[target_feature(enable = "avx512fp16,avx512vl")] +#[cfg_attr(test, assert_instr(vgetmantph, NORM = 0, SIGN = 0))] +#[rustc_legacy_const_generics(3, 4)] +#[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] +pub unsafe fn _mm256_mask_getmant_ph< + const NORM: _MM_MANTISSA_NORM_ENUM, + const SIGN: _MM_MANTISSA_SIGN_ENUM, +>( + src: __m256h, + k: __mmask16, + a: __m256h, +) -> __m256h { + static_assert_uimm_bits!(NORM, 4); + static_assert_uimm_bits!(SIGN, 2); + vgetmantph_256(a, (SIGN << 2) | NORM, src, k) +} + +/// Normalize the mantissas of packed half-precision (16-bit) floating-point elements in a, and store +/// the results in dst using zeromask k (elements are zeroed out when the corresponding mask bit is not set). +/// This intrinsic essentially calculates `±(2^k)*|x.significand|`, where k depends on the interval range defined +/// by norm and the sign depends on sign and the source sign. +/// +/// The mantissa is normalized to the interval specified by interv, which can take the following values: +/// +/// _MM_MANT_NORM_1_2 // interval [1, 2) +/// _MM_MANT_NORM_p5_2 // interval [0.5, 2) +/// _MM_MANT_NORM_p5_1 // interval [0.5, 1) +/// _MM_MANT_NORM_p75_1p5 // interval [0.75, 1.5) +/// +/// The sign is determined by sc which can take the following values: +/// +/// _MM_MANT_SIGN_src // sign = sign(src) +/// _MM_MANT_SIGN_zero // sign = 0 +/// _MM_MANT_SIGN_nan // dst = NaN if sign(src) = 1 +/// +/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm256_maskz_getmant_ph) +#[inline] +#[target_feature(enable = "avx512fp16,avx512vl")] +#[cfg_attr(test, assert_instr(vgetmantph, NORM = 0, SIGN = 0))] +#[rustc_legacy_const_generics(2, 3)] +#[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] +pub unsafe fn _mm256_maskz_getmant_ph< + const NORM: _MM_MANTISSA_NORM_ENUM, + const SIGN: _MM_MANTISSA_SIGN_ENUM, +>( + k: __mmask16, + a: __m256h, +) -> __m256h { + static_assert_uimm_bits!(NORM, 4); + static_assert_uimm_bits!(SIGN, 2); + _mm256_mask_getmant_ph::(_mm256_setzero_ph(), k, a) +} + +/// Normalize the mantissas of packed half-precision (16-bit) floating-point elements in a, and store +/// the results in dst. This intrinsic essentially calculates `±(2^k)*|x.significand|`, where k depends +/// on the interval range defined by norm and the sign depends on sign and the source sign. +/// +/// The mantissa is normalized to the interval specified by interv, which can take the following values: +/// +/// _MM_MANT_NORM_1_2 // interval [1, 2) +/// _MM_MANT_NORM_p5_2 // interval [0.5, 2) +/// _MM_MANT_NORM_p5_1 // interval [0.5, 1) +/// _MM_MANT_NORM_p75_1p5 // interval [0.75, 1.5) +/// +/// The sign is determined by sc which can take the following values: +/// +/// _MM_MANT_SIGN_src // sign = sign(src) +/// _MM_MANT_SIGN_zero // sign = 0 +/// _MM_MANT_SIGN_nan // dst = NaN if sign(src) = 1 +/// +/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm512_getmant_ph) +#[inline] +#[target_feature(enable = "avx512fp16")] +#[cfg_attr(test, assert_instr(vgetmantph, NORM = 0, SIGN = 0))] +#[rustc_legacy_const_generics(1, 2)] +#[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] +pub unsafe fn _mm512_getmant_ph< + const NORM: _MM_MANTISSA_NORM_ENUM, + const SIGN: _MM_MANTISSA_SIGN_ENUM, +>( + a: __m512h, +) -> __m512h { + static_assert_uimm_bits!(NORM, 4); + static_assert_uimm_bits!(SIGN, 2); + _mm512_mask_getmant_ph::(_mm512_undefined_ph(), 0xffffffff, a) +} + +/// Normalize the mantissas of packed half-precision (16-bit) floating-point elements in a, and store +/// the results in dst using writemask k (elements are copied from src when the corresponding mask bit is not set). +/// This intrinsic essentially calculates `±(2^k)*|x.significand|`, where k depends on the interval range defined +/// by norm and the sign depends on sign and the source sign. +/// +/// The mantissa is normalized to the interval specified by interv, which can take the following values: +/// +/// _MM_MANT_NORM_1_2 // interval [1, 2) +/// _MM_MANT_NORM_p5_2 // interval [0.5, 2) +/// _MM_MANT_NORM_p5_1 // interval [0.5, 1) +/// _MM_MANT_NORM_p75_1p5 // interval [0.75, 1.5) +/// +/// The sign is determined by sc which can take the following values: +/// +/// _MM_MANT_SIGN_src // sign = sign(src) +/// _MM_MANT_SIGN_zero // sign = 0 +/// _MM_MANT_SIGN_nan // dst = NaN if sign(src) = 1 +/// +/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm512_mask_getmant_ph) +#[inline] +#[target_feature(enable = "avx512fp16")] +#[cfg_attr(test, assert_instr(vgetmantph, NORM = 0, SIGN = 0))] +#[rustc_legacy_const_generics(3, 4)] +#[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] +pub unsafe fn _mm512_mask_getmant_ph< + const NORM: _MM_MANTISSA_NORM_ENUM, + const SIGN: _MM_MANTISSA_SIGN_ENUM, +>( + src: __m512h, + k: __mmask32, + a: __m512h, +) -> __m512h { + static_assert_uimm_bits!(NORM, 4); + static_assert_uimm_bits!(SIGN, 2); + _mm512_mask_getmant_round_ph::(src, k, a) +} + +/// Normalize the mantissas of packed half-precision (16-bit) floating-point elements in a, and store +/// the results in dst using zeromask k (elements are zeroed out when the corresponding mask bit is not set). +/// This intrinsic essentially calculates `±(2^k)*|x.significand|`, where k depends on the interval range defined +/// by norm and the sign depends on sign and the source sign. +/// +/// The mantissa is normalized to the interval specified by interv, which can take the following values: +/// +/// _MM_MANT_NORM_1_2 // interval [1, 2) +/// _MM_MANT_NORM_p5_2 // interval [0.5, 2) +/// _MM_MANT_NORM_p5_1 // interval [0.5, 1) +/// _MM_MANT_NORM_p75_1p5 // interval [0.75, 1.5) +/// +/// The sign is determined by sc which can take the following values: +/// +/// _MM_MANT_SIGN_src // sign = sign(src) +/// _MM_MANT_SIGN_zero // sign = 0 +/// _MM_MANT_SIGN_nan // dst = NaN if sign(src) = 1 +/// +/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm512_maskz_getmant_ph) +#[inline] +#[target_feature(enable = "avx512fp16")] +#[cfg_attr(test, assert_instr(vgetmantph, NORM = 0, SIGN = 0))] +#[rustc_legacy_const_generics(2, 3)] +#[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] +pub unsafe fn _mm512_maskz_getmant_ph< + const NORM: _MM_MANTISSA_NORM_ENUM, + const SIGN: _MM_MANTISSA_SIGN_ENUM, +>( + k: __mmask32, + a: __m512h, +) -> __m512h { + static_assert_uimm_bits!(NORM, 4); + static_assert_uimm_bits!(SIGN, 2); + _mm512_mask_getmant_ph::(_mm512_setzero_ph(), k, a) +} + +/// Normalize the mantissas of packed half-precision (16-bit) floating-point elements in a, and store +/// the results in dst. This intrinsic essentially calculates `±(2^k)*|x.significand|`, where k depends +/// on the interval range defined by norm and the sign depends on sign and the source sign. Exceptions can +/// be suppressed by passing _MM_FROUND_NO_EXC in the sae parameter +/// +/// The mantissa is normalized to the interval specified by interv, which can take the following values: +/// +/// _MM_MANT_NORM_1_2 // interval [1, 2) +/// _MM_MANT_NORM_p5_2 // interval [0.5, 2) +/// _MM_MANT_NORM_p5_1 // interval [0.5, 1) +/// _MM_MANT_NORM_p75_1p5 // interval [0.75, 1.5) +/// +/// The sign is determined by sc which can take the following values: +/// +/// _MM_MANT_SIGN_src // sign = sign(src) +/// _MM_MANT_SIGN_zero // sign = 0 +/// _MM_MANT_SIGN_nan // dst = NaN if sign(src) = 1 +/// +/// Exceptions can be suppressed by passing _MM_FROUND_NO_EXC in the sae parameter +/// +/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm512_getmant_round_ph) +#[inline] +#[target_feature(enable = "avx512fp16")] +#[cfg_attr(test, assert_instr(vgetmantph, NORM = 0, SIGN = 0, SAE = 8))] +#[rustc_legacy_const_generics(1, 2, 3)] +#[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] +pub unsafe fn _mm512_getmant_round_ph< + const NORM: _MM_MANTISSA_NORM_ENUM, + const SIGN: _MM_MANTISSA_SIGN_ENUM, + const SAE: i32, +>( + a: __m512h, +) -> __m512h { + static_assert_uimm_bits!(NORM, 4); + static_assert_uimm_bits!(SIGN, 2); + static_assert_sae!(SAE); + _mm512_mask_getmant_round_ph::(_mm512_undefined_ph(), 0xffffffff, a) +} + +/// Normalize the mantissas of packed half-precision (16-bit) floating-point elements in a, and store +/// the results in dst using writemask k (elements are copied from src when the corresponding mask bit is not set). +/// This intrinsic essentially calculates `±(2^k)*|x.significand|`, where k depends on the interval range defined +/// by norm and the sign depends on sign and the source sign. Exceptions can be suppressed by passing _MM_FROUND_NO_EXC +/// in the sae parameter +/// +/// The mantissa is normalized to the interval specified by interv, which can take the following values: +/// +/// _MM_MANT_NORM_1_2 // interval [1, 2) +/// _MM_MANT_NORM_p5_2 // interval [0.5, 2) +/// _MM_MANT_NORM_p5_1 // interval [0.5, 1) +/// _MM_MANT_NORM_p75_1p5 // interval [0.75, 1.5) +/// +/// The sign is determined by sc which can take the following values: +/// +/// _MM_MANT_SIGN_src // sign = sign(src) +/// _MM_MANT_SIGN_zero // sign = 0 +/// _MM_MANT_SIGN_nan // dst = NaN if sign(src) = 1 +/// +/// Exceptions can be suppressed by passing _MM_FROUND_NO_EXC in the sae parameter +/// +/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm512_mask_getmant_round_ph) +#[inline] +#[target_feature(enable = "avx512fp16")] +#[cfg_attr(test, assert_instr(vgetmantph, NORM = 0, SIGN = 0, SAE = 8))] +#[rustc_legacy_const_generics(3, 4, 5)] +#[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] +pub unsafe fn _mm512_mask_getmant_round_ph< + const NORM: _MM_MANTISSA_NORM_ENUM, + const SIGN: _MM_MANTISSA_SIGN_ENUM, + const SAE: i32, +>( + src: __m512h, + k: __mmask32, + a: __m512h, +) -> __m512h { + static_assert_uimm_bits!(NORM, 4); + static_assert_uimm_bits!(SIGN, 2); + static_assert_sae!(SAE); + vgetmantph_512(a, (SIGN << 2) | NORM, src, k, SAE) +} + +/// Normalize the mantissas of packed half-precision (16-bit) floating-point elements in a, and store +/// the results in dst using zeromask k (elements are zeroed out when the corresponding mask bit is not set). +/// This intrinsic essentially calculates `±(2^k)*|x.significand|`, where k depends on the interval range defined +/// by norm and the sign depends on sign and the source sign. Exceptions can be suppressed by passing _MM_FROUND_NO_EXC +/// in the sae parameter +/// +/// The mantissa is normalized to the interval specified by interv, which can take the following values: +/// +/// _MM_MANT_NORM_1_2 // interval [1, 2) +/// _MM_MANT_NORM_p5_2 // interval [0.5, 2) +/// _MM_MANT_NORM_p5_1 // interval [0.5, 1) +/// _MM_MANT_NORM_p75_1p5 // interval [0.75, 1.5) +/// +/// The sign is determined by sc which can take the following values: +/// +/// _MM_MANT_SIGN_src // sign = sign(src) +/// _MM_MANT_SIGN_zero // sign = 0 +/// _MM_MANT_SIGN_nan // dst = NaN if sign(src) = 1 +/// +/// Exceptions can be suppressed by passing _MM_FROUND_NO_EXC in the sae parameter +/// +/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm512_maskz_getmant_round_ph) +#[inline] +#[target_feature(enable = "avx512fp16")] +#[cfg_attr(test, assert_instr(vgetmantph, NORM = 0, SIGN = 0, SAE = 8))] +#[rustc_legacy_const_generics(2, 3, 4)] +#[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] +pub unsafe fn _mm512_maskz_getmant_round_ph< + const NORM: _MM_MANTISSA_NORM_ENUM, + const SIGN: _MM_MANTISSA_SIGN_ENUM, + const SAE: i32, +>( + k: __mmask32, + a: __m512h, +) -> __m512h { + static_assert_uimm_bits!(NORM, 4); + static_assert_uimm_bits!(SIGN, 2); + static_assert_sae!(SAE); + _mm512_mask_getmant_round_ph::(_mm512_setzero_ph(), k, a) +} + +/// Normalize the mantissas of the lower half-precision (16-bit) floating-point element in b, store +/// the result in the lower element of dst, and copy the upper 7 packed elements from a to the upper +/// elements of dst. This intrinsic essentially calculates `±(2^k)*|x.significand|`, where k depends +/// on the interval range defined by norm and the sign depends on sign and the source sign. +/// +/// The mantissa is normalized to the interval specified by interv, which can take the following values: +/// +/// _MM_MANT_NORM_1_2 // interval [1, 2) +/// _MM_MANT_NORM_p5_2 // interval [0.5, 2) +/// _MM_MANT_NORM_p5_1 // interval [0.5, 1) +/// _MM_MANT_NORM_p75_1p5 // interval [0.75, 1.5) +/// +/// The sign is determined by sc which can take the following values: +/// +/// _MM_MANT_SIGN_src // sign = sign(src) +/// _MM_MANT_SIGN_zero // sign = 0 +/// _MM_MANT_SIGN_nan // dst = NaN if sign(src) = 1 +/// +/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_getmant_sh) +#[inline] +#[target_feature(enable = "avx512fp16")] +#[cfg_attr(test, assert_instr(vgetmantsh, NORM = 0, SIGN = 0))] +#[rustc_legacy_const_generics(2, 3)] +#[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] +pub unsafe fn _mm_getmant_sh< + const NORM: _MM_MANTISSA_NORM_ENUM, + const SIGN: _MM_MANTISSA_SIGN_ENUM, +>( + a: __m128h, + b: __m128h, +) -> __m128h { + static_assert_uimm_bits!(NORM, 4); + static_assert_uimm_bits!(SIGN, 2); + _mm_mask_getmant_sh::(_mm_undefined_ph(), 0xff, a, b) +} + +/// Normalize the mantissas of the lower half-precision (16-bit) floating-point element in b, store +/// the result in the lower element of dst using writemask k (the element is copied from src when mask bit 0 is not set), +/// and copy the upper 7 packed elements from a to the upper elements of dst. This intrinsic essentially calculates +/// `±(2^k)*|x.significand|`, where k depends on the interval range defined by norm and the sign depends on sign and +/// the source sign. +/// +/// The mantissa is normalized to the interval specified by interv, which can take the following values: +/// +/// _MM_MANT_NORM_1_2 // interval [1, 2) +/// _MM_MANT_NORM_p5_2 // interval [0.5, 2) +/// _MM_MANT_NORM_p5_1 // interval [0.5, 1) +/// _MM_MANT_NORM_p75_1p5 // interval [0.75, 1.5) +/// +/// The sign is determined by sc which can take the following values: +/// +/// _MM_MANT_SIGN_src // sign = sign(src) +/// _MM_MANT_SIGN_zero // sign = 0 +/// _MM_MANT_SIGN_nan // dst = NaN if sign(src) = 1 +/// +/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_mask_getmant_sh) +#[inline] +#[target_feature(enable = "avx512fp16")] +#[cfg_attr(test, assert_instr(vgetmantsh, NORM = 0, SIGN = 0))] +#[rustc_legacy_const_generics(4, 5)] +#[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] +pub unsafe fn _mm_mask_getmant_sh< + const NORM: _MM_MANTISSA_NORM_ENUM, + const SIGN: _MM_MANTISSA_SIGN_ENUM, +>( + src: __m128h, + k: __mmask8, + a: __m128h, + b: __m128h, +) -> __m128h { + static_assert_uimm_bits!(NORM, 4); + static_assert_uimm_bits!(SIGN, 2); + _mm_mask_getmant_round_sh::(src, k, a, b) +} + +/// Normalize the mantissas of the lower half-precision (16-bit) floating-point element in b, store +/// the result in the lower element of dst using zeromask k (the element is zeroed out when mask bit 0 is not set), +/// and copy the upper 7 packed elements from a to the upper elements of dst. This intrinsic essentially calculates +/// `±(2^k)*|x.significand|`, where k depends on the interval range defined by norm and the sign depends on sign and +/// the source sign. +/// +/// The mantissa is normalized to the interval specified by interv, which can take the following values: +/// +/// _MM_MANT_NORM_1_2 // interval [1, 2) +/// _MM_MANT_NORM_p5_2 // interval [0.5, 2) +/// _MM_MANT_NORM_p5_1 // interval [0.5, 1) +/// _MM_MANT_NORM_p75_1p5 // interval [0.75, 1.5) +/// +/// The sign is determined by sc which can take the following values: +/// +/// _MM_MANT_SIGN_src // sign = sign(src) +/// _MM_MANT_SIGN_zero // sign = 0 +/// _MM_MANT_SIGN_nan // dst = NaN if sign(src) = 1 +/// +/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_maskz_getmant_sh) +#[inline] +#[target_feature(enable = "avx512fp16")] +#[cfg_attr(test, assert_instr(vgetmantsh, NORM = 0, SIGN = 0))] +#[rustc_legacy_const_generics(3, 4)] +#[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] +pub unsafe fn _mm_maskz_getmant_sh< + const NORM: _MM_MANTISSA_NORM_ENUM, + const SIGN: _MM_MANTISSA_SIGN_ENUM, +>( + k: __mmask8, + a: __m128h, + b: __m128h, +) -> __m128h { + static_assert_uimm_bits!(NORM, 4); + static_assert_uimm_bits!(SIGN, 2); + _mm_mask_getmant_sh::(_mm_setzero_ph(), k, a, b) +} + +/// Normalize the mantissas of the lower half-precision (16-bit) floating-point element in b, store +/// the result in the lower element of dst, and copy the upper 7 packed elements from a to the upper +/// elements of dst. This intrinsic essentially calculates `±(2^k)*|x.significand|`, where k depends +/// on the interval range defined by norm and the sign depends on sign and the source sign. Exceptions can +/// be suppressed by passing _MM_FROUND_NO_EXC in the sae parameter +/// +/// The mantissa is normalized to the interval specified by interv, which can take the following values: +/// +/// _MM_MANT_NORM_1_2 // interval [1, 2) +/// _MM_MANT_NORM_p5_2 // interval [0.5, 2) +/// _MM_MANT_NORM_p5_1 // interval [0.5, 1) +/// _MM_MANT_NORM_p75_1p5 // interval [0.75, 1.5) +/// +/// The sign is determined by sc which can take the following values: +/// +/// _MM_MANT_SIGN_src // sign = sign(src) +/// _MM_MANT_SIGN_zero // sign = 0 +/// _MM_MANT_SIGN_nan // dst = NaN if sign(src) = 1 +/// +/// Exceptions can be suppressed by passing _MM_FROUND_NO_EXC in the sae parameter +/// +/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_getmant_round_sh) +#[inline] +#[target_feature(enable = "avx512fp16")] +#[cfg_attr(test, assert_instr(vgetmantsh, NORM = 0, SIGN = 0, SAE = 8))] +#[rustc_legacy_const_generics(2, 3, 4)] +#[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] +pub unsafe fn _mm_getmant_round_sh< + const NORM: _MM_MANTISSA_NORM_ENUM, + const SIGN: _MM_MANTISSA_SIGN_ENUM, + const SAE: i32, +>( + a: __m128h, + b: __m128h, +) -> __m128h { + static_assert_uimm_bits!(NORM, 4); + static_assert_uimm_bits!(SIGN, 2); + static_assert_sae!(SAE); + _mm_mask_getmant_round_sh::(_mm_undefined_ph(), 0xff, a, b) +} + +/// Normalize the mantissas of the lower half-precision (16-bit) floating-point element in b, store +/// the result in the lower element of dst using writemask k (the element is copied from src when mask bit 0 is not set), +/// and copy the upper 7 packed elements from a to the upper elements of dst. This intrinsic essentially calculates +/// `±(2^k)*|x.significand|`, where k depends on the interval range defined by norm and the sign depends on sign and +/// the source sign. Exceptions can be suppressed by passing _MM_FROUND_NO_EXC in the sae parameter +/// +/// The mantissa is normalized to the interval specified by interv, which can take the following values: +/// +/// _MM_MANT_NORM_1_2 // interval [1, 2) +/// _MM_MANT_NORM_p5_2 // interval [0.5, 2) +/// _MM_MANT_NORM_p5_1 // interval [0.5, 1) +/// _MM_MANT_NORM_p75_1p5 // interval [0.75, 1.5) +/// +/// The sign is determined by sc which can take the following values: +/// +/// _MM_MANT_SIGN_src // sign = sign(src) +/// _MM_MANT_SIGN_zero // sign = 0 +/// _MM_MANT_SIGN_nan // dst = NaN if sign(src) = 1 +/// +/// Exceptions can be suppressed by passing _MM_FROUND_NO_EXC in the sae parameter +/// +/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_mask_getmant_round_sh) +#[inline] +#[target_feature(enable = "avx512fp16")] +#[cfg_attr(test, assert_instr(vgetmantsh, NORM = 0, SIGN = 0, SAE = 8))] +#[rustc_legacy_const_generics(4, 5, 6)] +#[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] +pub unsafe fn _mm_mask_getmant_round_sh< + const NORM: _MM_MANTISSA_NORM_ENUM, + const SIGN: _MM_MANTISSA_SIGN_ENUM, + const SAE: i32, +>( + src: __m128h, + k: __mmask8, + a: __m128h, + b: __m128h, +) -> __m128h { + static_assert_uimm_bits!(NORM, 4); + static_assert_uimm_bits!(SIGN, 2); + static_assert_sae!(SAE); + vgetmantsh(a, b, (SIGN << 2) | NORM, src, k, SAE) +} + +/// Normalize the mantissas of the lower half-precision (16-bit) floating-point element in b, store +/// the result in the lower element of dst using zeromask k (the element is zeroed out when mask bit 0 is not set), +/// and copy the upper 7 packed elements from a to the upper elements of dst. This intrinsic essentially calculates +/// `±(2^k)*|x.significand|`, where k depends on the interval range defined by norm and the sign depends on sign and +/// the source sign. Exceptions can be suppressed by passing _MM_FROUND_NO_EXC in the sae parameter +/// +/// The mantissa is normalized to the interval specified by interv, which can take the following values: +/// +/// _MM_MANT_NORM_1_2 // interval [1, 2) +/// _MM_MANT_NORM_p5_2 // interval [0.5, 2) +/// _MM_MANT_NORM_p5_1 // interval [0.5, 1) +/// _MM_MANT_NORM_p75_1p5 // interval [0.75, 1.5) +/// +/// The sign is determined by sc which can take the following values: +/// +/// _MM_MANT_SIGN_src // sign = sign(src) +/// _MM_MANT_SIGN_zero // sign = 0 +/// _MM_MANT_SIGN_nan // dst = NaN if sign(src) = 1 +/// +/// Exceptions can be suppressed by passing _MM_FROUND_NO_EXC in the sae parameter +/// +/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_maskz_getmant_round_sh) +#[inline] +#[target_feature(enable = "avx512fp16")] +#[cfg_attr(test, assert_instr(vgetmantsh, NORM = 0, SIGN = 0, SAE = 8))] +#[rustc_legacy_const_generics(3, 4, 5)] +#[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] +pub unsafe fn _mm_maskz_getmant_round_sh< + const NORM: _MM_MANTISSA_NORM_ENUM, + const SIGN: _MM_MANTISSA_SIGN_ENUM, + const SAE: i32, +>( + k: __mmask8, + a: __m128h, + b: __m128h, +) -> __m128h { + static_assert_uimm_bits!(NORM, 4); + static_assert_uimm_bits!(SIGN, 2); + static_assert_sae!(SAE); + _mm_mask_getmant_round_sh::(_mm_setzero_ph(), k, a, b) +} + +/// Round packed half-precision (16-bit) floating-point elements in a to the number of fraction bits +/// specified by imm8, and store the results in dst. +/// +/// Rounding is done according to the imm8 parameter, which can be one of: +/// +/// _MM_FROUND_TO_NEAREST_INT // round to nearest +/// _MM_FROUND_TO_NEG_INF // round down +/// _MM_FROUND_TO_POS_INF // round up +/// _MM_FROUND_TO_ZERO // truncate +/// _MM_FROUND_CUR_DIRECTION // use MXCSR.RC; see _MM_SET_ROUNDING_MODE +/// +/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_roundscale_ph) +#[inline] +#[target_feature(enable = "avx512fp16,avx512vl")] +#[cfg_attr(test, assert_instr(vrndscaleph, IMM8 = 0))] +#[rustc_legacy_const_generics(1)] +#[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] +pub unsafe fn _mm_roundscale_ph(a: __m128h) -> __m128h { + static_assert_uimm_bits!(IMM8, 8); + _mm_mask_roundscale_ph::(_mm_undefined_ph(), 0xff, a) +} + +/// Round packed half-precision (16-bit) floating-point elements in a to the number of fraction bits +/// specified by imm8, and store the results in dst using writemask k (elements are copied from src when +/// the corresponding mask bit is not set). +/// +/// Rounding is done according to the imm8 parameter, which can be one of: +/// +/// _MM_FROUND_TO_NEAREST_INT // round to nearest +/// _MM_FROUND_TO_NEG_INF // round down +/// _MM_FROUND_TO_POS_INF // round up +/// _MM_FROUND_TO_ZERO // truncate +/// _MM_FROUND_CUR_DIRECTION // use MXCSR.RC; see _MM_SET_ROUNDING_MODE +/// +/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_mask_roundscale_ph) +#[inline] +#[target_feature(enable = "avx512fp16,avx512vl")] +#[cfg_attr(test, assert_instr(vrndscaleph, IMM8 = 0))] +#[rustc_legacy_const_generics(3)] +#[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] +pub unsafe fn _mm_mask_roundscale_ph( + src: __m128h, + k: __mmask8, + a: __m128h, +) -> __m128h { + static_assert_uimm_bits!(IMM8, 8); + vrndscaleph_128(a, IMM8, src, k) +} + +/// Round packed half-precision (16-bit) floating-point elements in a to the number of fraction bits +/// specified by imm8, and store the results in dst using zeromask k (elements are zeroed out when the corresponding +/// mask bit is not set). +/// +/// Rounding is done according to the imm8 parameter, which can be one of: +/// +/// _MM_FROUND_TO_NEAREST_INT // round to nearest +/// _MM_FROUND_TO_NEG_INF // round down +/// _MM_FROUND_TO_POS_INF // round up +/// _MM_FROUND_TO_ZERO // truncate +/// _MM_FROUND_CUR_DIRECTION // use MXCSR.RC; see _MM_SET_ROUNDING_MODE +/// +/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_maskz_roundscale_ph) +#[inline] +#[target_feature(enable = "avx512fp16,avx512vl")] +#[cfg_attr(test, assert_instr(vrndscaleph, IMM8 = 0))] +#[rustc_legacy_const_generics(2)] +#[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] +pub unsafe fn _mm_maskz_roundscale_ph(k: __mmask8, a: __m128h) -> __m128h { + static_assert_uimm_bits!(IMM8, 8); + _mm_mask_roundscale_ph::(_mm_setzero_ph(), k, a) +} + +/// Round packed half-precision (16-bit) floating-point elements in a to the number of fraction bits +/// specified by imm8, and store the results in dst. +/// +/// Rounding is done according to the imm8 parameter, which can be one of: +/// +/// _MM_FROUND_TO_NEAREST_INT // round to nearest +/// _MM_FROUND_TO_NEG_INF // round down +/// _MM_FROUND_TO_POS_INF // round up +/// _MM_FROUND_TO_ZERO // truncate +/// _MM_FROUND_CUR_DIRECTION // use MXCSR.RC; see _MM_SET_ROUNDING_MODE +/// +/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm256_roundscale_ph) +#[inline] +#[target_feature(enable = "avx512fp16,avx512vl")] +#[cfg_attr(test, assert_instr(vrndscaleph, IMM8 = 0))] +#[rustc_legacy_const_generics(1)] +#[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] +pub unsafe fn _mm256_roundscale_ph(a: __m256h) -> __m256h { + static_assert_uimm_bits!(IMM8, 8); + _mm256_mask_roundscale_ph::(_mm256_undefined_ph(), 0xffff, a) +} + +/// Round packed half-precision (16-bit) floating-point elements in a to the number of fraction bits +/// specified by imm8, and store the results in dst using writemask k (elements are copied from src when +/// the corresponding mask bit is not set). +/// +/// Rounding is done according to the imm8 parameter, which can be one of: +/// +/// _MM_FROUND_TO_NEAREST_INT // round to nearest +/// _MM_FROUND_TO_NEG_INF // round down +/// _MM_FROUND_TO_POS_INF // round up +/// _MM_FROUND_TO_ZERO // truncate +/// _MM_FROUND_CUR_DIRECTION // use MXCSR.RC; see _MM_SET_ROUNDING_MODE +/// +/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm256_mask_roundscale_ph) +#[inline] +#[target_feature(enable = "avx512fp16,avx512vl")] +#[cfg_attr(test, assert_instr(vrndscaleph, IMM8 = 0))] +#[rustc_legacy_const_generics(3)] +#[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] +pub unsafe fn _mm256_mask_roundscale_ph( + src: __m256h, + k: __mmask16, + a: __m256h, +) -> __m256h { + static_assert_uimm_bits!(IMM8, 8); + vrndscaleph_256(a, IMM8, src, k) +} + +/// Round packed half-precision (16-bit) floating-point elements in a to the number of fraction bits +/// specified by imm8, and store the results in dst using zeromask k (elements are zeroed out when the corresponding +/// mask bit is not set). +/// +/// Rounding is done according to the imm8 parameter, which can be one of: +/// +/// _MM_FROUND_TO_NEAREST_INT // round to nearest +/// _MM_FROUND_TO_NEG_INF // round down +/// _MM_FROUND_TO_POS_INF // round up +/// _MM_FROUND_TO_ZERO // truncate +/// _MM_FROUND_CUR_DIRECTION // use MXCSR.RC; see _MM_SET_ROUNDING_MODE +/// +/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm256_maskz_roundscale_ph) +#[inline] +#[target_feature(enable = "avx512fp16,avx512vl")] +#[cfg_attr(test, assert_instr(vrndscaleph, IMM8 = 0))] +#[rustc_legacy_const_generics(2)] +#[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] +pub unsafe fn _mm256_maskz_roundscale_ph(k: __mmask16, a: __m256h) -> __m256h { + static_assert_uimm_bits!(IMM8, 8); + _mm256_mask_roundscale_ph::(_mm256_setzero_ph(), k, a) +} + +/// Round packed half-precision (16-bit) floating-point elements in a to the number of fraction bits +/// specified by imm8, and store the results in dst. +/// +/// Rounding is done according to the imm8 parameter, which can be one of: +/// +/// _MM_FROUND_TO_NEAREST_INT // round to nearest +/// _MM_FROUND_TO_NEG_INF // round down +/// _MM_FROUND_TO_POS_INF // round up +/// _MM_FROUND_TO_ZERO // truncate +/// _MM_FROUND_CUR_DIRECTION // use MXCSR.RC; see _MM_SET_ROUNDING_MODE +/// +/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm512_roundscale_ph) +#[inline] +#[target_feature(enable = "avx512fp16")] +#[cfg_attr(test, assert_instr(vrndscaleph, IMM8 = 0))] +#[rustc_legacy_const_generics(1)] +#[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] +pub unsafe fn _mm512_roundscale_ph(a: __m512h) -> __m512h { + static_assert_uimm_bits!(IMM8, 8); + _mm512_mask_roundscale_ph::(_mm512_undefined_ph(), 0xffffffff, a) +} + +/// Round packed half-precision (16-bit) floating-point elements in a to the number of fraction bits +/// specified by imm8, and store the results in dst using writemask k (elements are copied from src when +/// the corresponding mask bit is not set). +/// +/// Rounding is done according to the imm8 parameter, which can be one of: +/// +/// _MM_FROUND_TO_NEAREST_INT // round to nearest +/// _MM_FROUND_TO_NEG_INF // round down +/// _MM_FROUND_TO_POS_INF // round up +/// _MM_FROUND_TO_ZERO // truncate +/// _MM_FROUND_CUR_DIRECTION // use MXCSR.RC; see _MM_SET_ROUNDING_MODE +/// +/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm512_mask_roundscale_ph) +#[inline] +#[target_feature(enable = "avx512fp16")] +#[cfg_attr(test, assert_instr(vrndscaleph, IMM8 = 0))] +#[rustc_legacy_const_generics(3)] +#[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] +pub unsafe fn _mm512_mask_roundscale_ph( + src: __m512h, + k: __mmask32, + a: __m512h, +) -> __m512h { + static_assert_uimm_bits!(IMM8, 8); + _mm512_mask_roundscale_round_ph::(src, k, a) +} + +/// Round packed half-precision (16-bit) floating-point elements in a to the number of fraction bits +/// specified by imm8, and store the results in dst using zeromask k (elements are zeroed out when the corresponding +/// mask bit is not set). +/// +/// Rounding is done according to the imm8 parameter, which can be one of: +/// +/// _MM_FROUND_TO_NEAREST_INT // round to nearest +/// _MM_FROUND_TO_NEG_INF // round down +/// _MM_FROUND_TO_POS_INF // round up +/// _MM_FROUND_TO_ZERO // truncate +/// _MM_FROUND_CUR_DIRECTION // use MXCSR.RC; see _MM_SET_ROUNDING_MODE +/// +/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm512_maskz_roundscale_ph) +#[inline] +#[target_feature(enable = "avx512fp16")] +#[cfg_attr(test, assert_instr(vrndscaleph, IMM8 = 0))] +#[rustc_legacy_const_generics(2)] +#[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] +pub unsafe fn _mm512_maskz_roundscale_ph(k: __mmask32, a: __m512h) -> __m512h { + static_assert_uimm_bits!(IMM8, 8); + _mm512_mask_roundscale_ph::(_mm512_setzero_ph(), k, a) +} + +/// Round packed half-precision (16-bit) floating-point elements in a to the number of fraction bits +/// specified by imm8, and store the results in dst. Exceptions can be suppressed by passing _MM_FROUND_NO_EXC +/// in the sae parameter +/// +/// Rounding is done according to the imm8 parameter, which can be one of: +/// +/// _MM_FROUND_TO_NEAREST_INT // round to nearest +/// _MM_FROUND_TO_NEG_INF // round down +/// _MM_FROUND_TO_POS_INF // round up +/// _MM_FROUND_TO_ZERO // truncate +/// _MM_FROUND_CUR_DIRECTION // use MXCSR.RC; see _MM_SET_ROUNDING_MODE +/// +/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm512_roundscale_round_ph) +#[inline] +#[target_feature(enable = "avx512fp16")] +#[cfg_attr(test, assert_instr(vrndscaleph, IMM8 = 0, SAE = 8))] +#[rustc_legacy_const_generics(1, 2)] +#[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] +pub unsafe fn _mm512_roundscale_round_ph(a: __m512h) -> __m512h { + static_assert_uimm_bits!(IMM8, 8); + static_assert_sae!(SAE); + _mm512_mask_roundscale_round_ph::(_mm512_undefined_ph(), 0xffffffff, a) +} + +/// Round packed half-precision (16-bit) floating-point elements in a to the number of fraction bits +/// specified by imm8, and store the results in dst using writemask k (elements are copied from src when +/// the corresponding mask bit is not set). Exceptions can be suppressed by passing _MM_FROUND_NO_EXC +/// in the sae parameter +/// +/// Rounding is done according to the imm8 parameter, which can be one of: +/// +/// _MM_FROUND_TO_NEAREST_INT // round to nearest +/// _MM_FROUND_TO_NEG_INF // round down +/// _MM_FROUND_TO_POS_INF // round up +/// _MM_FROUND_TO_ZERO // truncate +/// _MM_FROUND_CUR_DIRECTION // use MXCSR.RC; see _MM_SET_ROUNDING_MODE +/// +/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm512_mask_roundscale_round_ph) +#[inline] +#[target_feature(enable = "avx512fp16")] +#[cfg_attr(test, assert_instr(vrndscaleph, IMM8 = 0, SAE = 8))] +#[rustc_legacy_const_generics(3, 4)] +#[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] +pub unsafe fn _mm512_mask_roundscale_round_ph( + src: __m512h, + k: __mmask32, + a: __m512h, +) -> __m512h { + static_assert_uimm_bits!(IMM8, 8); + static_assert_sae!(SAE); + vrndscaleph_512(a, IMM8, src, k, SAE) +} + +/// Round packed half-precision (16-bit) floating-point elements in a to the number of fraction bits +/// specified by imm8, and store the results in dst using zeromask k (elements are zeroed out when the corresponding +/// mask bit is not set). Exceptions can be suppressed by passing _MM_FROUND_NO_EXC in the sae parameter +/// +/// Rounding is done according to the imm8 parameter, which can be one of: +/// +/// _MM_FROUND_TO_NEAREST_INT // round to nearest +/// _MM_FROUND_TO_NEG_INF // round down +/// _MM_FROUND_TO_POS_INF // round up +/// _MM_FROUND_TO_ZERO // truncate +/// _MM_FROUND_CUR_DIRECTION // use MXCSR.RC; see _MM_SET_ROUNDING_MODE +/// +/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm512_maskz_roundscale_round_ph) +#[inline] +#[target_feature(enable = "avx512fp16")] +#[cfg_attr(test, assert_instr(vrndscaleph, IMM8 = 0, SAE = 8))] +#[rustc_legacy_const_generics(2, 3)] +#[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] +pub unsafe fn _mm512_maskz_roundscale_round_ph( + k: __mmask32, + a: __m512h, +) -> __m512h { + static_assert_uimm_bits!(IMM8, 8); + static_assert_sae!(SAE); + _mm512_mask_roundscale_round_ph::(_mm512_setzero_ph(), k, a) +} + +/// Round the lower half-precision (16-bit) floating-point element in b to the number of fraction bits +/// specified by imm8, store the result in the lower element of dst, and copy the upper 7 packed elements +/// from a to the upper elements of dst. +/// +/// Rounding is done according to the imm8 parameter, which can be one of: +/// +/// _MM_FROUND_TO_NEAREST_INT // round to nearest +/// _MM_FROUND_TO_NEG_INF // round down +/// _MM_FROUND_TO_POS_INF // round up +/// _MM_FROUND_TO_ZERO // truncate +/// _MM_FROUND_CUR_DIRECTION // use MXCSR.RC; see _MM_SET_ROUNDING_MODE +/// +/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_roundscale_sh) +#[inline] +#[target_feature(enable = "avx512fp16")] +#[cfg_attr(test, assert_instr(vrndscalesh, IMM8 = 0))] +#[rustc_legacy_const_generics(2)] +#[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] +pub unsafe fn _mm_roundscale_sh(a: __m128h, b: __m128h) -> __m128h { + static_assert_uimm_bits!(IMM8, 8); + _mm_mask_roundscale_sh::(_mm_undefined_ph(), 0xff, a, b) +} + +/// Round the lower half-precision (16-bit) floating-point element in b to the number of fraction bits +/// specified by imm8, store the result in the lower element of dst using writemask k (the element is copied +/// from src when mask bit 0 is not set), and copy the upper 7 packed elements from a to the upper elements of dst. +/// +/// Rounding is done according to the imm8 parameter, which can be one of: +/// +/// _MM_FROUND_TO_NEAREST_INT // round to nearest +/// _MM_FROUND_TO_NEG_INF // round down +/// _MM_FROUND_TO_POS_INF // round up +/// _MM_FROUND_TO_ZERO // truncate +/// _MM_FROUND_CUR_DIRECTION // use MXCSR.RC; see _MM_SET_ROUNDING_MODE +/// +/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_mask_roundscale_sh) +#[inline] +#[target_feature(enable = "avx512fp16")] +#[cfg_attr(test, assert_instr(vrndscalesh, IMM8 = 0))] +#[rustc_legacy_const_generics(4)] +#[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] +pub unsafe fn _mm_mask_roundscale_sh( + src: __m128h, + k: __mmask8, + a: __m128h, + b: __m128h, +) -> __m128h { + static_assert_uimm_bits!(IMM8, 8); + _mm_mask_roundscale_round_sh::(src, k, a, b) +} + +/// Round the lower half-precision (16-bit) floating-point element in b to the number of fraction bits +/// specified by imm8, store the result in the lower element of dst using zeromask k (the element is zeroed +/// out when mask bit 0 is not set), and copy the upper 7 packed elements from a to the upper elements of dst. +/// +/// Rounding is done according to the imm8 parameter, which can be one of: +/// +/// _MM_FROUND_TO_NEAREST_INT // round to nearest +/// _MM_FROUND_TO_NEG_INF // round down +/// _MM_FROUND_TO_POS_INF // round up +/// _MM_FROUND_TO_ZERO // truncate +/// _MM_FROUND_CUR_DIRECTION // use MXCSR.RC; see _MM_SET_ROUNDING_MODE +/// +/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_maskz_roundscale_sh) +#[inline] +#[target_feature(enable = "avx512fp16")] +#[cfg_attr(test, assert_instr(vrndscalesh, IMM8 = 0))] +#[rustc_legacy_const_generics(3)] +#[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] +pub unsafe fn _mm_maskz_roundscale_sh( + k: __mmask8, + a: __m128h, + b: __m128h, +) -> __m128h { + static_assert_uimm_bits!(IMM8, 8); + _mm_mask_roundscale_sh::(_mm_setzero_ph(), k, a, b) +} + +/// Round the lower half-precision (16-bit) floating-point element in b to the number of fraction bits +/// specified by imm8, store the result in the lower element of dst, and copy the upper 7 packed elements +/// from a to the upper elements of dst. +/// +/// Rounding is done according to the imm8 parameter, which can be one of: +/// +/// _MM_FROUND_TO_NEAREST_INT // round to nearest +/// _MM_FROUND_TO_NEG_INF // round down +/// _MM_FROUND_TO_POS_INF // round up +/// _MM_FROUND_TO_ZERO // truncate +/// _MM_FROUND_CUR_DIRECTION // use MXCSR.RC; see _MM_SET_ROUNDING_MODE +/// +/// Exceptions can be suppressed by passing _MM_FROUND_NO_EXC in the sae parameter +/// +/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_roundscale_round_sh) +#[inline] +#[target_feature(enable = "avx512fp16")] +#[cfg_attr(test, assert_instr(vrndscalesh, IMM8 = 0, SAE = 8))] +#[rustc_legacy_const_generics(2, 3)] +#[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] +pub unsafe fn _mm_roundscale_round_sh( + a: __m128h, + b: __m128h, +) -> __m128h { + static_assert_uimm_bits!(IMM8, 8); + static_assert_sae!(SAE); + _mm_mask_roundscale_round_sh::(_mm_undefined_ph(), 0xff, a, b) +} + +/// Round the lower half-precision (16-bit) floating-point element in b to the number of fraction bits +/// specified by imm8, store the result in the lower element of dst using writemask k (the element is copied +/// from src when mask bit 0 is not set), and copy the upper 7 packed elements from a to the upper elements of dst. +/// +/// Rounding is done according to the imm8 parameter, which can be one of: +/// +/// _MM_FROUND_TO_NEAREST_INT // round to nearest +/// _MM_FROUND_TO_NEG_INF // round down +/// _MM_FROUND_TO_POS_INF // round up +/// _MM_FROUND_TO_ZERO // truncate +/// _MM_FROUND_CUR_DIRECTION // use MXCSR.RC; see _MM_SET_ROUNDING_MODE +/// +/// Exceptions can be suppressed by passing _MM_FROUND_NO_EXC in the sae parameter +/// +/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_mask_roundscale_round_sh) +#[inline] +#[target_feature(enable = "avx512fp16")] +#[cfg_attr(test, assert_instr(vrndscalesh, IMM8 = 0, SAE = 8))] +#[rustc_legacy_const_generics(4, 5)] +#[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] +pub unsafe fn _mm_mask_roundscale_round_sh( + src: __m128h, + k: __mmask8, + a: __m128h, + b: __m128h, +) -> __m128h { + static_assert_uimm_bits!(IMM8, 8); + static_assert_sae!(SAE); + vrndscalesh(a, b, src, k, IMM8, SAE) +} + +/// Round the lower half-precision (16-bit) floating-point element in b to the number of fraction bits +/// specified by imm8, store the result in the lower element of dst using zeromask k (the element is zeroed +/// out when mask bit 0 is not set), and copy the upper 7 packed elements from a to the upper elements of dst. +/// +/// Rounding is done according to the imm8 parameter, which can be one of: +/// +/// _MM_FROUND_TO_NEAREST_INT // round to nearest +/// _MM_FROUND_TO_NEG_INF // round down +/// _MM_FROUND_TO_POS_INF // round up +/// _MM_FROUND_TO_ZERO // truncate +/// _MM_FROUND_CUR_DIRECTION // use MXCSR.RC; see _MM_SET_ROUNDING_MODE +/// +/// Exceptions can be suppressed by passing _MM_FROUND_NO_EXC in the sae parameter +/// +/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_maskz_roundscale_round_sh) +#[inline] +#[target_feature(enable = "avx512fp16")] +#[cfg_attr(test, assert_instr(vrndscalesh, IMM8 = 0, SAE = 8))] +#[rustc_legacy_const_generics(3, 4)] +#[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] +pub unsafe fn _mm_maskz_roundscale_round_sh( + k: __mmask8, + a: __m128h, + b: __m128h, +) -> __m128h { + static_assert_uimm_bits!(IMM8, 8); + static_assert_sae!(SAE); + _mm_mask_roundscale_round_sh::(_mm_setzero_ph(), k, a, b) +} + +/// Scale the packed half-precision (16-bit) floating-point elements in a using values from b, and store +/// the results in dst. +/// +/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_scalef_ph) +#[inline] +#[target_feature(enable = "avx512fp16,avx512vl")] +#[cfg_attr(test, assert_instr(vscalefph))] +#[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] +pub unsafe fn _mm_scalef_ph(a: __m128h, b: __m128h) -> __m128h { + _mm_mask_scalef_ph(_mm_undefined_ph(), 0xff, a, b) +} + +/// Scale the packed half-precision (16-bit) floating-point elements in a using values from b, and store +/// the results in dst using writemask k (elements are copied from src when the corresponding mask bit is not set). +/// +/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_mask_scalef_ph) +#[inline] +#[target_feature(enable = "avx512fp16,avx512vl")] +#[cfg_attr(test, assert_instr(vscalefph))] +#[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] +pub unsafe fn _mm_mask_scalef_ph(src: __m128h, k: __mmask8, a: __m128h, b: __m128h) -> __m128h { + vscalefph_128(a, b, src, k) +} + +/// Scale the packed half-precision (16-bit) floating-point elements in a using values from b, and store +/// the results in dst using zeromask k (elements are zeroed out when the corresponding mask bit is not set). +/// +/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_maskz_scalef_ph) +#[inline] +#[target_feature(enable = "avx512fp16,avx512vl")] +#[cfg_attr(test, assert_instr(vscalefph))] +#[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] +pub unsafe fn _mm_maskz_scalef_ph(k: __mmask8, a: __m128h, b: __m128h) -> __m128h { + _mm_mask_scalef_ph(_mm_setzero_ph(), k, a, b) +} + +/// Scale the packed half-precision (16-bit) floating-point elements in a using values from b, and store +/// the results in dst. +/// +/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm256_scalef_ph) +#[inline] +#[target_feature(enable = "avx512fp16,avx512vl")] +#[cfg_attr(test, assert_instr(vscalefph))] +#[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] +pub unsafe fn _mm256_scalef_ph(a: __m256h, b: __m256h) -> __m256h { + _mm256_mask_scalef_ph(_mm256_undefined_ph(), 0xffff, a, b) +} + +/// Scale the packed half-precision (16-bit) floating-point elements in a using values from b, and store +/// the results in dst using writemask k (elements are copied from src when the corresponding mask bit is not set). +/// +/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm256_mask_scalef_ph) +#[inline] +#[target_feature(enable = "avx512fp16,avx512vl")] +#[cfg_attr(test, assert_instr(vscalefph))] +#[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] +pub unsafe fn _mm256_mask_scalef_ph(src: __m256h, k: __mmask16, a: __m256h, b: __m256h) -> __m256h { + vscalefph_256(a, b, src, k) +} + +/// Scale the packed half-precision (16-bit) floating-point elements in a using values from b, and store +/// the results in dst using zeromask k (elements are zeroed out when the corresponding mask bit is not set). +/// +/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm256_maskz_scalef_ph) +#[inline] +#[target_feature(enable = "avx512fp16,avx512vl")] +#[cfg_attr(test, assert_instr(vscalefph))] +#[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] +pub unsafe fn _mm256_maskz_scalef_ph(k: __mmask16, a: __m256h, b: __m256h) -> __m256h { + _mm256_mask_scalef_ph(_mm256_setzero_ph(), k, a, b) +} + +/// Scale the packed half-precision (16-bit) floating-point elements in a using values from b, and store +/// the results in dst. +/// +/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm512_scalef_ph) +#[inline] +#[target_feature(enable = "avx512fp16")] +#[cfg_attr(test, assert_instr(vscalefph))] +#[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] +pub unsafe fn _mm512_scalef_ph(a: __m512h, b: __m512h) -> __m512h { + _mm512_mask_scalef_ph(_mm512_undefined_ph(), 0xffffffff, a, b) +} + +/// Scale the packed half-precision (16-bit) floating-point elements in a using values from b, and store +/// the results in dst using writemask k (elements are copied from src when the corresponding mask bit is not set). +/// +/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm512_mask_scalef_ph) +#[inline] +#[target_feature(enable = "avx512fp16")] +#[cfg_attr(test, assert_instr(vscalefph))] +#[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] +pub unsafe fn _mm512_mask_scalef_ph(src: __m512h, k: __mmask32, a: __m512h, b: __m512h) -> __m512h { + _mm512_mask_scalef_round_ph::<_MM_FROUND_CUR_DIRECTION>(src, k, a, b) +} + +/// Scale the packed half-precision (16-bit) floating-point elements in a using values from b, and store +/// the results in dst using zeromask k (elements are zeroed out when the corresponding mask bit is not set). +/// +/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm512_maskz_scalef_ph) +#[inline] +#[target_feature(enable = "avx512fp16")] +#[cfg_attr(test, assert_instr(vscalefph))] +#[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] +pub unsafe fn _mm512_maskz_scalef_ph(k: __mmask32, a: __m512h, b: __m512h) -> __m512h { + _mm512_mask_scalef_ph(_mm512_setzero_ph(), k, a, b) +} + +/// Scale the packed half-precision (16-bit) floating-point elements in a using values from b, and store +/// the results in dst. +/// +/// Rounding is done according to the rounding parameter, which can be one of: +/// +/// (_MM_FROUND_TO_NEAREST_INT |_MM_FROUND_NO_EXC) // round to nearest, and suppress exceptions +/// (_MM_FROUND_TO_NEG_INF |_MM_FROUND_NO_EXC) // round down, and suppress exceptions +/// (_MM_FROUND_TO_POS_INF |_MM_FROUND_NO_EXC) // round up, and suppress exceptions +/// (_MM_FROUND_TO_ZERO |_MM_FROUND_NO_EXC) // truncate, and suppress exceptions +/// _MM_FROUND_CUR_DIRECTION // use MXCSR.RC; see _MM_SET_ROUNDING_MODE +/// +/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm512_scalef_round_ph) +#[inline] +#[target_feature(enable = "avx512fp16")] +#[cfg_attr(test, assert_instr(vscalefph, ROUNDING = 8))] +#[rustc_legacy_const_generics(2)] +#[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] +pub unsafe fn _mm512_scalef_round_ph(a: __m512h, b: __m512h) -> __m512h { + static_assert_rounding!(ROUNDING); + _mm512_mask_scalef_round_ph::(_mm512_undefined_ph(), 0xffffffff, a, b) +} + +/// Scale the packed half-precision (16-bit) floating-point elements in a using values from b, and store +/// the results in dst using writemask k (elements are copied from src when the corresponding mask bit is not set). +/// +/// Rounding is done according to the rounding parameter, which can be one of: +/// +/// (_MM_FROUND_TO_NEAREST_INT |_MM_FROUND_NO_EXC) // round to nearest, and suppress exceptions +/// (_MM_FROUND_TO_NEG_INF |_MM_FROUND_NO_EXC) // round down, and suppress exceptions +/// (_MM_FROUND_TO_POS_INF |_MM_FROUND_NO_EXC) // round up, and suppress exceptions +/// (_MM_FROUND_TO_ZERO |_MM_FROUND_NO_EXC) // truncate, and suppress exceptions +/// _MM_FROUND_CUR_DIRECTION // use MXCSR.RC; see _MM_SET_ROUNDING_MODE +/// +/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm512_mask_scalef_round_ph) +#[inline] +#[target_feature(enable = "avx512fp16")] +#[cfg_attr(test, assert_instr(vscalefph, ROUNDING = 8))] +#[rustc_legacy_const_generics(4)] +#[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] +pub unsafe fn _mm512_mask_scalef_round_ph( + src: __m512h, + k: __mmask32, + a: __m512h, + b: __m512h, +) -> __m512h { + static_assert_rounding!(ROUNDING); + vscalefph_512(a, b, src, k, ROUNDING) +} + +/// Scale the packed half-precision (16-bit) floating-point elements in a using values from b, and store +/// the results in dst using zeromask k (elements are zeroed out when the corresponding mask bit is not set). +/// +/// Rounding is done according to the rounding parameter, which can be one of: +/// +/// (_MM_FROUND_TO_NEAREST_INT |_MM_FROUND_NO_EXC) // round to nearest, and suppress exceptions +/// (_MM_FROUND_TO_NEG_INF |_MM_FROUND_NO_EXC) // round down, and suppress exceptions +/// (_MM_FROUND_TO_POS_INF |_MM_FROUND_NO_EXC) // round up, and suppress exceptions +/// (_MM_FROUND_TO_ZERO |_MM_FROUND_NO_EXC) // truncate, and suppress exceptions +/// _MM_FROUND_CUR_DIRECTION // use MXCSR.RC; see _MM_SET_ROUNDING_MODE +/// +/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm512_maskz_scalef_round_ph) +#[inline] +#[target_feature(enable = "avx512fp16")] +#[cfg_attr(test, assert_instr(vscalefph, ROUNDING = 8))] +#[rustc_legacy_const_generics(3)] +#[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] +pub unsafe fn _mm512_maskz_scalef_round_ph( + k: __mmask32, + a: __m512h, + b: __m512h, +) -> __m512h { + static_assert_rounding!(ROUNDING); + _mm512_mask_scalef_round_ph::(_mm512_setzero_ph(), k, a, b) +} + +/// Scale the packed single-precision (32-bit) floating-point elements in a using values from b, store +/// the result in the lower element of dst, and copy the upper 7 packed elements from a to the upper +/// elements of dst. +/// +/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_scalef_sh) +#[inline] +#[target_feature(enable = "avx512fp16")] +#[cfg_attr(test, assert_instr(vscalefsh))] +#[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] +pub unsafe fn _mm_scalef_sh(a: __m128h, b: __m128h) -> __m128h { + _mm_mask_scalef_sh(_mm_undefined_ph(), 0xff, a, b) +} + +/// Scale the packed single-precision (32-bit) floating-point elements in a using values from b, store +/// the result in the lower element of dst using writemask k (the element is copied from src when mask bit 0 is not set), +/// and copy the upper 7 packed elements from a to the upper elements of dst. +/// +/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_mask_scalef_sh) +#[inline] +#[target_feature(enable = "avx512fp16")] +#[cfg_attr(test, assert_instr(vscalefsh))] +#[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] +pub unsafe fn _mm_mask_scalef_sh(src: __m128h, k: __mmask8, a: __m128h, b: __m128h) -> __m128h { + _mm_mask_scalef_round_sh::<_MM_FROUND_CUR_DIRECTION>(src, k, a, b) +} + +/// Scale the packed single-precision (32-bit) floating-point elements in a using values from b, store +/// the result in the lower element of dst using zeromask k (the element is zeroed out when mask bit 0 is not set), +/// and copy the upper 7 packed elements from a to the upper elements of dst. +/// +/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_maskz_scalef_sh) +#[inline] +#[target_feature(enable = "avx512fp16")] +#[cfg_attr(test, assert_instr(vscalefsh))] +#[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] +pub unsafe fn _mm_maskz_scalef_sh(k: __mmask8, a: __m128h, b: __m128h) -> __m128h { + _mm_mask_scalef_sh(_mm_setzero_ph(), k, a, b) +} + +/// Scale the packed single-precision (32-bit) floating-point elements in a using values from b, store +/// the result in the lower element of dst, and copy the upper 7 packed elements from a to the upper +/// elements of dst. +/// +/// Rounding is done according to the rounding parameter, which can be one of: +/// +/// (_MM_FROUND_TO_NEAREST_INT |_MM_FROUND_NO_EXC) // round to nearest, and suppress exceptions +/// (_MM_FROUND_TO_NEG_INF |_MM_FROUND_NO_EXC) // round down, and suppress exceptions +/// (_MM_FROUND_TO_POS_INF |_MM_FROUND_NO_EXC) // round up, and suppress exceptions +/// (_MM_FROUND_TO_ZERO |_MM_FROUND_NO_EXC) // truncate, and suppress exceptions +/// _MM_FROUND_CUR_DIRECTION // use MXCSR.RC; see _MM_SET_ROUNDING_MODE +/// +/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_scalef_round_sh) +#[inline] +#[target_feature(enable = "avx512fp16")] +#[cfg_attr(test, assert_instr(vscalefsh, ROUNDING = 8))] +#[rustc_legacy_const_generics(2)] +#[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] +pub unsafe fn _mm_scalef_round_sh(a: __m128h, b: __m128h) -> __m128h { + static_assert_rounding!(ROUNDING); + _mm_mask_scalef_round_sh::(_mm_undefined_ph(), 0xff, a, b) +} + +/// Scale the packed single-precision (32-bit) floating-point elements in a using values from b, store +/// the result in the lower element of dst using writemask k (the element is copied from src when mask bit 0 is not set), +/// and copy the upper 7 packed elements from a to the upper elements of dst. +/// +/// Rounding is done according to the rounding parameter, which can be one of: +/// +/// (_MM_FROUND_TO_NEAREST_INT |_MM_FROUND_NO_EXC) // round to nearest, and suppress exceptions +/// (_MM_FROUND_TO_NEG_INF |_MM_FROUND_NO_EXC) // round down, and suppress exceptions +/// (_MM_FROUND_TO_POS_INF |_MM_FROUND_NO_EXC) // round up, and suppress exceptions +/// (_MM_FROUND_TO_ZERO |_MM_FROUND_NO_EXC) // truncate, and suppress exceptions +/// _MM_FROUND_CUR_DIRECTION // use MXCSR.RC; see _MM_SET_ROUNDING_MODE +/// +/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_mask_scalef_round_sh) +#[inline] +#[target_feature(enable = "avx512fp16")] +#[cfg_attr(test, assert_instr(vscalefsh, ROUNDING = 8))] +#[rustc_legacy_const_generics(4)] +#[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] +pub unsafe fn _mm_mask_scalef_round_sh( + src: __m128h, + k: __mmask8, + a: __m128h, + b: __m128h, +) -> __m128h { + static_assert_rounding!(ROUNDING); + vscalefsh(a, b, src, k, ROUNDING) +} + +/// Scale the packed single-precision (32-bit) floating-point elements in a using values from b, store +/// the result in the lower element of dst using zeromask k (the element is zeroed out when mask bit 0 is not set), +/// and copy the upper 7 packed elements from a to the upper elements of dst. +/// +/// Rounding is done according to the rounding parameter, which can be one of: +/// +/// (_MM_FROUND_TO_NEAREST_INT |_MM_FROUND_NO_EXC) // round to nearest, and suppress exceptions +/// (_MM_FROUND_TO_NEG_INF |_MM_FROUND_NO_EXC) // round down, and suppress exceptions +/// (_MM_FROUND_TO_POS_INF |_MM_FROUND_NO_EXC) // round up, and suppress exceptions +/// (_MM_FROUND_TO_ZERO |_MM_FROUND_NO_EXC) // truncate, and suppress exceptions +/// _MM_FROUND_CUR_DIRECTION // use MXCSR.RC; see _MM_SET_ROUNDING_MODE +/// +/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_maskz_scalef_round_sh) +#[inline] +#[target_feature(enable = "avx512fp16")] +#[cfg_attr(test, assert_instr(vscalefsh, ROUNDING = 8))] +#[rustc_legacy_const_generics(3)] +#[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] +pub unsafe fn _mm_maskz_scalef_round_sh( + k: __mmask8, + a: __m128h, + b: __m128h, +) -> __m128h { + static_assert_rounding!(ROUNDING); + _mm_mask_scalef_round_sh::(_mm_setzero_ph(), k, a, b) +} + +/// Extract the reduced argument of packed half-precision (16-bit) floating-point elements in a by the +/// number of bits specified by imm8, and store the results in dst. +/// +/// Rounding is done according to the imm8 parameter, which can be one of: +/// +/// _MM_FROUND_TO_NEAREST_INT // round to nearest +/// _MM_FROUND_TO_NEG_INF // round down +/// _MM_FROUND_TO_POS_INF // round up +/// _MM_FROUND_TO_ZERO // truncate +/// _MM_FROUND_CUR_DIRECTION // use MXCSR.RC; see _MM_SET_ROUNDING_MODE +/// +/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_reduce_ph) +#[inline] +#[target_feature(enable = "avx512fp16,avx512vl")] +#[cfg_attr(test, assert_instr(vreduceph, IMM8 = 0))] +#[rustc_legacy_const_generics(1)] +#[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] +pub unsafe fn _mm_reduce_ph(a: __m128h) -> __m128h { + static_assert_uimm_bits!(IMM8, 8); + _mm_mask_reduce_ph::(_mm_undefined_ph(), 0xff, a) +} + +/// Extract the reduced argument of packed half-precision (16-bit) floating-point elements in a by the +/// number of bits specified by imm8, and store the results in dst using writemask k (elements are copied +/// from src when the corresponding mask bit is not set). +/// +/// Rounding is done according to the imm8 parameter, which can be one of: +/// +/// _MM_FROUND_TO_NEAREST_INT // round to nearest +/// _MM_FROUND_TO_NEG_INF // round down +/// _MM_FROUND_TO_POS_INF // round up +/// _MM_FROUND_TO_ZERO // truncate +/// _MM_FROUND_CUR_DIRECTION // use MXCSR.RC; see _MM_SET_ROUNDING_MODE +/// +/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_mask_reduce_ph) +#[inline] +#[target_feature(enable = "avx512fp16,avx512vl")] +#[cfg_attr(test, assert_instr(vreduceph, IMM8 = 0))] +#[rustc_legacy_const_generics(3)] +#[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] +pub unsafe fn _mm_mask_reduce_ph( + src: __m128h, + k: __mmask8, + a: __m128h, +) -> __m128h { + static_assert_uimm_bits!(IMM8, 8); + vreduceph_128(a, IMM8, src, k) +} + +/// Extract the reduced argument of packed half-precision (16-bit) floating-point elements in a by the +/// number of bits specified by imm8, and store the results in dst using zeromask k (elements are zeroed +/// out when the corresponding mask bit is not set). +/// +/// Rounding is done according to the imm8 parameter, which can be one of: +/// +/// _MM_FROUND_TO_NEAREST_INT // round to nearest +/// _MM_FROUND_TO_NEG_INF // round down +/// _MM_FROUND_TO_POS_INF // round up +/// _MM_FROUND_TO_ZERO // truncate +/// _MM_FROUND_CUR_DIRECTION // use MXCSR.RC; see _MM_SET_ROUNDING_MODE +/// +/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_maskz_reduce_ph) +#[inline] +#[target_feature(enable = "avx512fp16,avx512vl")] +#[cfg_attr(test, assert_instr(vreduceph, IMM8 = 0))] +#[rustc_legacy_const_generics(2)] +#[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] +pub unsafe fn _mm_maskz_reduce_ph(k: __mmask8, a: __m128h) -> __m128h { + static_assert_uimm_bits!(IMM8, 8); + _mm_mask_reduce_ph::(_mm_setzero_ph(), k, a) +} + +/// Extract the reduced argument of packed half-precision (16-bit) floating-point elements in a by the +/// number of bits specified by imm8, and store the results in dst. +/// +/// Rounding is done according to the imm8 parameter, which can be one of: +/// +/// _MM_FROUND_TO_NEAREST_INT // round to nearest +/// _MM_FROUND_TO_NEG_INF // round down +/// _MM_FROUND_TO_POS_INF // round up +/// _MM_FROUND_TO_ZERO // truncate +/// _MM_FROUND_CUR_DIRECTION // use MXCSR.RC; see _MM_SET_ROUNDING_MODE +/// +/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm256_reduce_ph) +#[inline] +#[target_feature(enable = "avx512fp16,avx512vl")] +#[cfg_attr(test, assert_instr(vreduceph, IMM8 = 0))] +#[rustc_legacy_const_generics(1)] +#[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] +pub unsafe fn _mm256_reduce_ph(a: __m256h) -> __m256h { + static_assert_uimm_bits!(IMM8, 8); + _mm256_mask_reduce_ph::(_mm256_undefined_ph(), 0xffff, a) +} + +/// Extract the reduced argument of packed half-precision (16-bit) floating-point elements in a by the +/// number of bits specified by imm8, and store the results in dst using writemask k (elements are copied +/// from src when the corresponding mask bit is not set). +/// +/// Rounding is done according to the imm8 parameter, which can be one of: +/// +/// _MM_FROUND_TO_NEAREST_INT // round to nearest +/// _MM_FROUND_TO_NEG_INF // round down +/// _MM_FROUND_TO_POS_INF // round up +/// _MM_FROUND_TO_ZERO // truncate +/// _MM_FROUND_CUR_DIRECTION // use MXCSR.RC; see _MM_SET_ROUNDING_MODE +/// +/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm256_mask_reduce_ph) +#[inline] +#[target_feature(enable = "avx512fp16,avx512vl")] +#[cfg_attr(test, assert_instr(vreduceph, IMM8 = 0))] +#[rustc_legacy_const_generics(3)] +#[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] +pub unsafe fn _mm256_mask_reduce_ph( + src: __m256h, + k: __mmask16, + a: __m256h, +) -> __m256h { + static_assert_uimm_bits!(IMM8, 8); + vreduceph_256(a, IMM8, src, k) +} + +/// Extract the reduced argument of packed half-precision (16-bit) floating-point elements in a by the +/// number of bits specified by imm8, and store the results in dst using zeromask k (elements are zeroed +/// out when the corresponding mask bit is not set). +/// +/// Rounding is done according to the imm8 parameter, which can be one of: +/// +/// _MM_FROUND_TO_NEAREST_INT // round to nearest +/// _MM_FROUND_TO_NEG_INF // round down +/// _MM_FROUND_TO_POS_INF // round up +/// _MM_FROUND_TO_ZERO // truncate +/// _MM_FROUND_CUR_DIRECTION // use MXCSR.RC; see _MM_SET_ROUNDING_MODE +/// +/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm256_maskz_reduce_ph) +#[inline] +#[target_feature(enable = "avx512fp16,avx512vl")] +#[cfg_attr(test, assert_instr(vreduceph, IMM8 = 0))] +#[rustc_legacy_const_generics(2)] +#[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] +pub unsafe fn _mm256_maskz_reduce_ph(k: __mmask16, a: __m256h) -> __m256h { + static_assert_uimm_bits!(IMM8, 8); + _mm256_mask_reduce_ph::(_mm256_setzero_ph(), k, a) +} + +/// Extract the reduced argument of packed half-precision (16-bit) floating-point elements in a by the +/// number of bits specified by imm8, and store the results in dst. +/// +/// Rounding is done according to the imm8 parameter, which can be one of: +/// +/// _MM_FROUND_TO_NEAREST_INT // round to nearest +/// _MM_FROUND_TO_NEG_INF // round down +/// _MM_FROUND_TO_POS_INF // round up +/// _MM_FROUND_TO_ZERO // truncate +/// _MM_FROUND_CUR_DIRECTION // use MXCSR.RC; see _MM_SET_ROUNDING_MODE +/// +/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm512_reduce_ph) +#[inline] +#[target_feature(enable = "avx512fp16")] +#[cfg_attr(test, assert_instr(vreduceph, IMM8 = 0))] +#[rustc_legacy_const_generics(1)] +#[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] +pub unsafe fn _mm512_reduce_ph(a: __m512h) -> __m512h { + static_assert_uimm_bits!(IMM8, 8); + _mm512_mask_reduce_ph::(_mm512_undefined_ph(), 0xffffffff, a) +} + +/// Extract the reduced argument of packed half-precision (16-bit) floating-point elements in a by the +/// number of bits specified by imm8, and store the results in dst using writemask k (elements are copied +/// from src when the corresponding mask bit is not set). +/// +/// Rounding is done according to the imm8 parameter, which can be one of: +/// +/// _MM_FROUND_TO_NEAREST_INT // round to nearest +/// _MM_FROUND_TO_NEG_INF // round down +/// _MM_FROUND_TO_POS_INF // round up +/// _MM_FROUND_TO_ZERO // truncate +/// _MM_FROUND_CUR_DIRECTION // use MXCSR.RC; see _MM_SET_ROUNDING_MODE +/// +/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm512_mask_reduce_ph) +#[inline] +#[target_feature(enable = "avx512fp16")] +#[cfg_attr(test, assert_instr(vreduceph, IMM8 = 0))] +#[rustc_legacy_const_generics(3)] +#[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] +pub unsafe fn _mm512_mask_reduce_ph( + src: __m512h, + k: __mmask32, + a: __m512h, +) -> __m512h { + static_assert_uimm_bits!(IMM8, 8); + _mm512_mask_reduce_round_ph::(src, k, a) +} + +/// Extract the reduced argument of packed half-precision (16-bit) floating-point elements in a by the +/// number of bits specified by imm8, and store the results in dst using zeromask k (elements are zeroed +/// out when the corresponding mask bit is not set). +/// +/// Rounding is done according to the imm8 parameter, which can be one of: +/// +/// _MM_FROUND_TO_NEAREST_INT // round to nearest +/// _MM_FROUND_TO_NEG_INF // round down +/// _MM_FROUND_TO_POS_INF // round up +/// _MM_FROUND_TO_ZERO // truncate +/// _MM_FROUND_CUR_DIRECTION // use MXCSR.RC; see _MM_SET_ROUNDING_MODE +/// +/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm512_maskz_reduce_ph) +#[inline] +#[target_feature(enable = "avx512fp16")] +#[cfg_attr(test, assert_instr(vreduceph, IMM8 = 0))] +#[rustc_legacy_const_generics(2)] +#[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] +pub unsafe fn _mm512_maskz_reduce_ph(k: __mmask32, a: __m512h) -> __m512h { + static_assert_uimm_bits!(IMM8, 8); + _mm512_mask_reduce_ph::(_mm512_setzero_ph(), k, a) +} + +/// Extract the reduced argument of packed half-precision (16-bit) floating-point elements in a by the +/// number of bits specified by imm8, and store the results in dst. +/// +/// Rounding is done according to the imm8 parameter, which can be one of: +/// +/// _MM_FROUND_TO_NEAREST_INT // round to nearest +/// _MM_FROUND_TO_NEG_INF // round down +/// _MM_FROUND_TO_POS_INF // round up +/// _MM_FROUND_TO_ZERO // truncate +/// _MM_FROUND_CUR_DIRECTION // use MXCSR.RC; see _MM_SET_ROUNDING_MODE +/// +/// Exceptions can be suppressed by passing _MM_FROUND_NO_EXC in the sae parameter. +/// +/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm512_reduce_round_ph) +#[inline] +#[target_feature(enable = "avx512fp16")] +#[cfg_attr(test, assert_instr(vreduceph, IMM8 = 0, SAE = 8))] +#[rustc_legacy_const_generics(1, 2)] +#[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] +pub unsafe fn _mm512_reduce_round_ph(a: __m512h) -> __m512h { + static_assert_uimm_bits!(IMM8, 8); + static_assert_sae!(SAE); + _mm512_mask_reduce_round_ph::(_mm512_undefined_ph(), 0xffffffff, a) +} + +/// Extract the reduced argument of packed half-precision (16-bit) floating-point elements in a by the +/// number of bits specified by imm8, and store the results in dst using writemask k (elements are copied +/// from src when the corresponding mask bit is not set). +/// +/// Rounding is done according to the imm8 parameter, which can be one of: +/// +/// _MM_FROUND_TO_NEAREST_INT // round to nearest +/// _MM_FROUND_TO_NEG_INF // round down +/// _MM_FROUND_TO_POS_INF // round up +/// _MM_FROUND_TO_ZERO // truncate +/// _MM_FROUND_CUR_DIRECTION // use MXCSR.RC; see _MM_SET_ROUNDING_MODE +/// +/// Exceptions can be suppressed by passing _MM_FROUND_NO_EXC in the sae parameter. +/// +/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm512_mask_reduce_round_ph) +#[inline] +#[target_feature(enable = "avx512fp16")] +#[cfg_attr(test, assert_instr(vreduceph, IMM8 = 0, SAE = 8))] +#[rustc_legacy_const_generics(3, 4)] +#[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] +pub unsafe fn _mm512_mask_reduce_round_ph( + src: __m512h, + k: __mmask32, + a: __m512h, +) -> __m512h { + static_assert_uimm_bits!(IMM8, 8); + static_assert_sae!(SAE); + vreduceph_512(a, IMM8, src, k, SAE) +} + +/// Extract the reduced argument of packed half-precision (16-bit) floating-point elements in a by the +/// number of bits specified by imm8, and store the results in dst using zeromask k (elements are zeroed +/// out when the corresponding mask bit is not set). +/// +/// Rounding is done according to the imm8 parameter, which can be one of: +/// +/// _MM_FROUND_TO_NEAREST_INT // round to nearest +/// _MM_FROUND_TO_NEG_INF // round down +/// _MM_FROUND_TO_POS_INF // round up +/// _MM_FROUND_TO_ZERO // truncate +/// _MM_FROUND_CUR_DIRECTION // use MXCSR.RC; see _MM_SET_ROUNDING_MODE +/// +/// Exceptions can be suppressed by passing _MM_FROUND_NO_EXC in the sae parameter. +/// +/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm512_maskz_reduce_round_ph) +#[inline] +#[target_feature(enable = "avx512fp16")] +#[cfg_attr(test, assert_instr(vreduceph, IMM8 = 0, SAE = 8))] +#[rustc_legacy_const_generics(2, 3)] +#[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] +pub unsafe fn _mm512_maskz_reduce_round_ph( + k: __mmask32, + a: __m512h, +) -> __m512h { + static_assert_uimm_bits!(IMM8, 8); + static_assert_sae!(SAE); + _mm512_mask_reduce_round_ph::(_mm512_setzero_ph(), k, a) +} + +/// Extract the reduced argument of the lower half-precision (16-bit) floating-point element in b by +/// the number of bits specified by imm8, store the result in the lower element of dst, and copy the +/// upper 7 packed elements from a to the upper elements of dst. +/// +/// Rounding is done according to the imm8 parameter, which can be one of: +/// +/// _MM_FROUND_TO_NEAREST_INT // round to nearest +/// _MM_FROUND_TO_NEG_INF // round down +/// _MM_FROUND_TO_POS_INF // round up +/// _MM_FROUND_TO_ZERO // truncate +/// _MM_FROUND_CUR_DIRECTION // use MXCSR.RC; see _MM_SET_ROUNDING_MODE +/// +/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_reduce_sh) +#[inline] +#[target_feature(enable = "avx512fp16")] +#[cfg_attr(test, assert_instr(vreducesh, IMM8 = 0))] +#[rustc_legacy_const_generics(2)] +#[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] +pub unsafe fn _mm_reduce_sh(a: __m128h, b: __m128h) -> __m128h { + static_assert_uimm_bits!(IMM8, 8); + _mm_mask_reduce_sh::(_mm_undefined_ph(), 0xff, a, b) +} + +/// Extract the reduced argument of the lower half-precision (16-bit) floating-point element in b by +/// the number of bits specified by imm8, store the result in the lower element of dst using writemask k +/// (the element is copied from src when mask bit 0 is not set), and copy the upper 7 packed elements from +/// a to the upper elements of dst. +/// +/// Rounding is done according to the imm8 parameter, which can be one of: +/// +/// _MM_FROUND_TO_NEAREST_INT // round to nearest +/// _MM_FROUND_TO_NEG_INF // round down +/// _MM_FROUND_TO_POS_INF // round up +/// _MM_FROUND_TO_ZERO // truncate +/// _MM_FROUND_CUR_DIRECTION // use MXCSR.RC; see _MM_SET_ROUNDING_MODE +/// +/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_mask_reduce_sh) +#[inline] +#[target_feature(enable = "avx512fp16")] +#[cfg_attr(test, assert_instr(vreducesh, IMM8 = 0))] +#[rustc_legacy_const_generics(4)] +#[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] +pub unsafe fn _mm_mask_reduce_sh( + src: __m128h, + k: __mmask8, + a: __m128h, + b: __m128h, +) -> __m128h { + static_assert_uimm_bits!(IMM8, 8); + _mm_mask_reduce_round_sh::(src, k, a, b) +} + +/// Extract the reduced argument of the lower half-precision (16-bit) floating-point element in b by +/// the number of bits specified by imm8, store the result in the lower element of dst using zeromask k +/// (the element is zeroed out when mask bit 0 is not set), and copy the upper 7 packed elements from a +/// to the upper elements of dst. +/// +/// Rounding is done according to the imm8 parameter, which can be one of: +/// +/// _MM_FROUND_TO_NEAREST_INT // round to nearest +/// _MM_FROUND_TO_NEG_INF // round down +/// _MM_FROUND_TO_POS_INF // round up +/// _MM_FROUND_TO_ZERO // truncate +/// _MM_FROUND_CUR_DIRECTION // use MXCSR.RC; see _MM_SET_ROUNDING_MODE +/// +/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_maskz_reduce_sh) +#[inline] +#[target_feature(enable = "avx512fp16")] +#[cfg_attr(test, assert_instr(vreducesh, IMM8 = 0))] +#[rustc_legacy_const_generics(3)] +#[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] +pub unsafe fn _mm_maskz_reduce_sh(k: __mmask8, a: __m128h, b: __m128h) -> __m128h { + static_assert_uimm_bits!(IMM8, 8); + _mm_mask_reduce_sh::(_mm_setzero_ph(), k, a, b) +} + +/// Extract the reduced argument of the lower half-precision (16-bit) floating-point element in b by +/// the number of bits specified by imm8, store the result in the lower element of dst, and copy the upper +/// 7 packed elements from a to the upper elements of dst. +/// +/// Rounding is done according to the imm8 parameter, which can be one of: +/// +/// _MM_FROUND_TO_NEAREST_INT // round to nearest +/// _MM_FROUND_TO_NEG_INF // round down +/// _MM_FROUND_TO_POS_INF // round up +/// _MM_FROUND_TO_ZERO // truncate +/// _MM_FROUND_CUR_DIRECTION // use MXCSR.RC; see _MM_SET_ROUNDING_MODE +/// +/// Exceptions can be suppressed by passing _MM_FROUND_NO_EXC in the sae parameter. +/// +/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_reduce_round_sh) +#[inline] +#[target_feature(enable = "avx512fp16")] +#[cfg_attr(test, assert_instr(vreducesh, IMM8 = 0, SAE = 8))] +#[rustc_legacy_const_generics(2, 3)] +#[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] +pub unsafe fn _mm_reduce_round_sh( + a: __m128h, + b: __m128h, +) -> __m128h { + static_assert_uimm_bits!(IMM8, 8); + static_assert_sae!(SAE); + _mm_mask_reduce_round_sh::(_mm_undefined_ph(), 0xff, a, b) +} + +/// Extract the reduced argument of the lower half-precision (16-bit) floating-point element in b by +/// the number of bits specified by imm8, store the result in the lower element of dst using writemask k +/// (the element is copied from src when mask bit 0 is not set), and copy the upper 7 packed elements from a +/// to the upper elements of dst. +/// +/// Rounding is done according to the imm8 parameter, which can be one of: +/// +/// _MM_FROUND_TO_NEAREST_INT // round to nearest +/// _MM_FROUND_TO_NEG_INF // round down +/// _MM_FROUND_TO_POS_INF // round up +/// _MM_FROUND_TO_ZERO // truncate +/// _MM_FROUND_CUR_DIRECTION // use MXCSR.RC; see _MM_SET_ROUNDING_MODE +/// +/// Exceptions can be suppressed by passing _MM_FROUND_NO_EXC in the sae parameter. +/// +/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_mask_reduce_round_sh) +#[inline] +#[target_feature(enable = "avx512fp16")] +#[cfg_attr(test, assert_instr(vreducesh, IMM8 = 0, SAE = 8))] +#[rustc_legacy_const_generics(4, 5)] +#[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] +pub unsafe fn _mm_mask_reduce_round_sh( + src: __m128h, + k: __mmask8, + a: __m128h, + b: __m128h, +) -> __m128h { + static_assert_uimm_bits!(IMM8, 8); + static_assert_sae!(SAE); + vreducesh(a, b, src, k, IMM8, SAE) +} + +/// Extract the reduced argument of the lower half-precision (16-bit) floating-point element in b by +/// the number of bits specified by imm8, store the result in the lower element of dst using zeromask k +/// (the element is zeroed out when mask bit 0 is not set), and copy the upper 7 packed elements from a +/// to the upper elements of dst. +/// +/// Rounding is done according to the imm8 parameter, which can be one of: +/// +/// _MM_FROUND_TO_NEAREST_INT // round to nearest +/// _MM_FROUND_TO_NEG_INF // round down +/// _MM_FROUND_TO_POS_INF // round up +/// _MM_FROUND_TO_ZERO // truncate +/// _MM_FROUND_CUR_DIRECTION // use MXCSR.RC; see _MM_SET_ROUNDING_MODE +/// +/// Exceptions can be suppressed by passing _MM_FROUND_NO_EXC in the sae parameter. +/// +/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_maskz_reduce_round_sh) +#[inline] +#[target_feature(enable = "avx512fp16")] +#[cfg_attr(test, assert_instr(vreducesh, IMM8 = 0, SAE = 8))] +#[rustc_legacy_const_generics(3, 4)] +#[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")] +pub unsafe fn _mm_maskz_reduce_round_sh( + k: __mmask8, + a: __m128h, + b: __m128h, +) -> __m128h { + static_assert_uimm_bits!(IMM8, 8); + static_assert_sae!(SAE); + _mm_mask_reduce_round_sh::(_mm_setzero_ph(), k, a, b) +} + #[allow(improper_ctypes)] extern "C" { #[link_name = "llvm.x86.avx512fp16.mask.cmp.sh"] @@ -8574,6 +10773,65 @@ extern "C" { #[link_name = "llvm.x86.avx512fp16.mask.min.sh.round"] fn vminsh(a: __m128h, b: __m128h, src: __m128h, k: __mmask8, sae: i32) -> __m128h; + #[link_name = "llvm.x86.avx512fp16.mask.getexp.ph.128"] + fn vgetexpph_128(a: __m128h, src: __m128h, k: __mmask8) -> __m128h; + #[link_name = "llvm.x86.avx512fp16.mask.getexp.ph.256"] + fn vgetexpph_256(a: __m256h, src: __m256h, k: __mmask16) -> __m256h; + #[link_name = "llvm.x86.avx512fp16.mask.getexp.ph.512"] + fn vgetexpph_512(a: __m512h, src: __m512h, k: __mmask32, sae: i32) -> __m512h; + #[link_name = "llvm.x86.avx512fp16.mask.getexp.sh"] + fn vgetexpsh(a: __m128h, b: __m128h, src: __m128h, k: __mmask8, sae: i32) -> __m128h; + + #[link_name = "llvm.x86.avx512fp16.mask.getmant.ph.128"] + fn vgetmantph_128(a: __m128h, imm8: i32, src: __m128h, k: __mmask8) -> __m128h; + #[link_name = "llvm.x86.avx512fp16.mask.getmant.ph.256"] + fn vgetmantph_256(a: __m256h, imm8: i32, src: __m256h, k: __mmask16) -> __m256h; + #[link_name = "llvm.x86.avx512fp16.mask.getmant.ph.512"] + fn vgetmantph_512(a: __m512h, imm8: i32, src: __m512h, k: __mmask32, sae: i32) -> __m512h; + #[link_name = "llvm.x86.avx512fp16.mask.getmant.sh"] + fn vgetmantsh( + a: __m128h, + b: __m128h, + imm8: i32, + src: __m128h, + k: __mmask8, + sae: i32, + ) -> __m128h; + + #[link_name = "llvm.x86.avx512fp16.mask.rndscale.ph.128"] + fn vrndscaleph_128(a: __m128h, imm8: i32, src: __m128h, k: __mmask8) -> __m128h; + #[link_name = "llvm.x86.avx512fp16.mask.rndscale.ph.256"] + fn vrndscaleph_256(a: __m256h, imm8: i32, src: __m256h, k: __mmask16) -> __m256h; + #[link_name = "llvm.x86.avx512fp16.mask.rndscale.ph.512"] + fn vrndscaleph_512(a: __m512h, imm8: i32, src: __m512h, k: __mmask32, sae: i32) -> __m512h; + #[link_name = "llvm.x86.avx512fp16.mask.rndscale.sh"] + fn vrndscalesh( + a: __m128h, + b: __m128h, + src: __m128h, + k: __mmask8, + imm8: i32, + sae: i32, + ) -> __m128h; + + #[link_name = "llvm.x86.avx512fp16.mask.scalef.ph.128"] + fn vscalefph_128(a: __m128h, b: __m128h, src: __m128h, k: __mmask8) -> __m128h; + #[link_name = "llvm.x86.avx512fp16.mask.scalef.ph.256"] + fn vscalefph_256(a: __m256h, b: __m256h, src: __m256h, k: __mmask16) -> __m256h; + #[link_name = "llvm.x86.avx512fp16.mask.scalef.ph.512"] + fn vscalefph_512(a: __m512h, b: __m512h, src: __m512h, k: __mmask32, rounding: i32) -> __m512h; + #[link_name = "llvm.x86.avx512fp16.mask.scalef.sh"] + fn vscalefsh(a: __m128h, b: __m128h, src: __m128h, k: __mmask8, rounding: i32) -> __m128h; + + #[link_name = "llvm.x86.avx512fp16.mask.reduce.ph.128"] + fn vreduceph_128(a: __m128h, imm8: i32, src: __m128h, k: __mmask8) -> __m128h; + #[link_name = "llvm.x86.avx512fp16.mask.reduce.ph.256"] + fn vreduceph_256(a: __m256h, imm8: i32, src: __m256h, k: __mmask16) -> __m256h; + #[link_name = "llvm.x86.avx512fp16.mask.reduce.ph.512"] + fn vreduceph_512(a: __m512h, imm8: i32, src: __m512h, k: __mmask32, sae: i32) -> __m512h; + #[link_name = "llvm.x86.avx512fp16.mask.reduce.sh"] + fn vreducesh(a: __m128h, b: __m128h, src: __m128h, k: __mmask8, imm8: i32, sae: i32) + -> __m128h; } #[cfg(test)] @@ -14469,4 +16727,1031 @@ mod tests { let e = _mm_setr_ph(1.0, 10.0, 11.0, 12.0, 13.0, 14.0, 15.0, 16.0); assert_eq_m128h(r, e); } + + #[simd_test(enable = "avx512fp16,avx512vl")] + unsafe fn test_mm_getexp_ph() { + let a = _mm_set1_ph(3.0); + let r = _mm_getexp_ph(a); + let e = _mm_set1_ph(1.0); + assert_eq_m128h(r, e); + } + + #[simd_test(enable = "avx512fp16,avx512vl")] + unsafe fn test_mm_mask_getexp_ph() { + let a = _mm_set1_ph(3.0); + let src = _mm_set1_ph(4.0); + let r = _mm_mask_getexp_ph(src, 0b01010101, a); + let e = _mm_set_ph(4.0, 1.0, 4.0, 1.0, 4.0, 1.0, 4.0, 1.0); + assert_eq_m128h(r, e); + } + + #[simd_test(enable = "avx512fp16,avx512vl")] + unsafe fn test_mm_maskz_getexp_ph() { + let a = _mm_set1_ph(3.0); + let r = _mm_maskz_getexp_ph(0b01010101, a); + let e = _mm_set_ph(0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0); + assert_eq_m128h(r, e); + } + + #[simd_test(enable = "avx512fp16,avx512vl")] + unsafe fn test_mm256_getexp_ph() { + let a = _mm256_set1_ph(3.0); + let r = _mm256_getexp_ph(a); + let e = _mm256_set1_ph(1.0); + assert_eq_m256h(r, e); + } + + #[simd_test(enable = "avx512fp16,avx512vl")] + unsafe fn test_mm256_mask_getexp_ph() { + let a = _mm256_set1_ph(3.0); + let src = _mm256_set1_ph(4.0); + let r = _mm256_mask_getexp_ph(src, 0b0101010101010101, a); + let e = _mm256_set_ph( + 4.0, 1.0, 4.0, 1.0, 4.0, 1.0, 4.0, 1.0, 4.0, 1.0, 4.0, 1.0, 4.0, 1.0, 4.0, 1.0, + ); + assert_eq_m256h(r, e); + } + + #[simd_test(enable = "avx512fp16,avx512vl")] + unsafe fn test_mm256_maskz_getexp_ph() { + let a = _mm256_set1_ph(3.0); + let r = _mm256_maskz_getexp_ph(0b0101010101010101, a); + let e = _mm256_set_ph( + 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, + ); + assert_eq_m256h(r, e); + } + + #[simd_test(enable = "avx512fp16")] + unsafe fn test_mm512_getexp_ph() { + let a = _mm512_set1_ph(3.0); + let r = _mm512_getexp_ph(a); + let e = _mm512_set1_ph(1.0); + assert_eq_m512h(r, e); + } + + #[simd_test(enable = "avx512fp16")] + unsafe fn test_mm512_mask_getexp_ph() { + let a = _mm512_set1_ph(3.0); + let src = _mm512_set1_ph(4.0); + let r = _mm512_mask_getexp_ph(src, 0b01010101010101010101010101010101, a); + let e = _mm512_set_ph( + 4.0, 1.0, 4.0, 1.0, 4.0, 1.0, 4.0, 1.0, 4.0, 1.0, 4.0, 1.0, 4.0, 1.0, 4.0, 1.0, 4.0, + 1.0, 4.0, 1.0, 4.0, 1.0, 4.0, 1.0, 4.0, 1.0, 4.0, 1.0, 4.0, 1.0, 4.0, 1.0, + ); + assert_eq_m512h(r, e); + } + + #[simd_test(enable = "avx512fp16")] + unsafe fn test_mm512_maskz_getexp_ph() { + let a = _mm512_set1_ph(3.0); + let r = _mm512_maskz_getexp_ph(0b01010101010101010101010101010101, a); + let e = _mm512_set_ph( + 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, + 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, + ); + assert_eq_m512h(r, e); + } + + #[simd_test(enable = "avx512fp16")] + unsafe fn test_mm512_getexp_round_ph() { + let a = _mm512_set1_ph(3.0); + let r = _mm512_getexp_round_ph::<_MM_FROUND_NO_EXC>(a); + let e = _mm512_set1_ph(1.0); + assert_eq_m512h(r, e); + } + + #[simd_test(enable = "avx512fp16")] + unsafe fn test_mm512_mask_getexp_round_ph() { + let a = _mm512_set1_ph(3.0); + let src = _mm512_set1_ph(4.0); + let r = _mm512_mask_getexp_round_ph::<_MM_FROUND_NO_EXC>( + src, + 0b01010101010101010101010101010101, + a, + ); + let e = _mm512_set_ph( + 4.0, 1.0, 4.0, 1.0, 4.0, 1.0, 4.0, 1.0, 4.0, 1.0, 4.0, 1.0, 4.0, 1.0, 4.0, 1.0, 4.0, + 1.0, 4.0, 1.0, 4.0, 1.0, 4.0, 1.0, 4.0, 1.0, 4.0, 1.0, 4.0, 1.0, 4.0, 1.0, + ); + assert_eq_m512h(r, e); + } + + #[simd_test(enable = "avx512fp16")] + unsafe fn test_mm512_maskz_getexp_round_ph() { + let a = _mm512_set1_ph(3.0); + let r = _mm512_maskz_getexp_round_ph::<_MM_FROUND_NO_EXC>( + 0b01010101010101010101010101010101, + a, + ); + let e = _mm512_set_ph( + 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, + 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, + ); + assert_eq_m512h(r, e); + } + + #[simd_test(enable = "avx512fp16")] + unsafe fn test_mm_getexp_sh() { + let a = _mm_setr_ph(4.0, 10., 11., 12., 13., 14., 15., 16.); + let b = _mm_setr_ph(3.0, 20., 21., 22., 23., 24., 25., 26.); + let r = _mm_getexp_sh(a, b); + let e = _mm_setr_ph(1.0, 10., 11., 12., 13., 14., 15., 16.); + assert_eq_m128h(r, e); + } + + #[simd_test(enable = "avx512fp16")] + unsafe fn test_mm_mask_getexp_sh() { + let a = _mm_setr_ph(4.0, 10., 11., 12., 13., 14., 15., 16.); + let b = _mm_setr_ph(3.0, 20., 21., 22., 23., 24., 25., 26.); + let src = _mm_setr_ph(4.0, 30., 31., 32., 33., 34., 35., 36.); + let r = _mm_mask_getexp_sh(src, 0, a, b); + let e = _mm_setr_ph(4.0, 10., 11., 12., 13., 14., 15., 16.); + assert_eq_m128h(r, e); + let r = _mm_mask_getexp_sh(src, 1, a, b); + let e = _mm_setr_ph(1.0, 10., 11., 12., 13., 14., 15., 16.); + assert_eq_m128h(r, e); + } + + #[simd_test(enable = "avx512fp16")] + unsafe fn test_mm_maskz_getexp_sh() { + let a = _mm_setr_ph(4.0, 10., 11., 12., 13., 14., 15., 16.); + let b = _mm_setr_ph(3.0, 20., 21., 22., 23., 24., 25., 26.); + let r = _mm_maskz_getexp_sh(0, a, b); + let e = _mm_setr_ph(0.0, 10., 11., 12., 13., 14., 15., 16.); + assert_eq_m128h(r, e); + let r = _mm_maskz_getexp_sh(1, a, b); + let e = _mm_setr_ph(1.0, 10., 11., 12., 13., 14., 15., 16.); + assert_eq_m128h(r, e); + } + + #[simd_test(enable = "avx512fp16")] + unsafe fn test_mm_getexp_round_sh() { + let a = _mm_setr_ph(4.0, 10., 11., 12., 13., 14., 15., 16.); + let b = _mm_setr_ph(3.0, 20., 21., 22., 23., 24., 25., 26.); + let r = _mm_getexp_round_sh::<_MM_FROUND_NO_EXC>(a, b); + let e = _mm_setr_ph(1.0, 10., 11., 12., 13., 14., 15., 16.); + assert_eq_m128h(r, e); + } + + #[simd_test(enable = "avx512fp16")] + unsafe fn test_mm_mask_getexp_round_sh() { + let a = _mm_setr_ph(4.0, 10., 11., 12., 13., 14., 15., 16.); + let b = _mm_setr_ph(3.0, 20., 21., 22., 23., 24., 25., 26.); + let src = _mm_setr_ph(4.0, 30., 31., 32., 33., 34., 35., 36.); + let r = _mm_mask_getexp_round_sh::<_MM_FROUND_NO_EXC>(src, 0, a, b); + let e = _mm_setr_ph(4.0, 10., 11., 12., 13., 14., 15., 16.); + assert_eq_m128h(r, e); + let r = _mm_mask_getexp_round_sh::<_MM_FROUND_NO_EXC>(src, 1, a, b); + let e = _mm_setr_ph(1.0, 10., 11., 12., 13., 14., 15., 16.); + assert_eq_m128h(r, e); + } + + #[simd_test(enable = "avx512fp16")] + unsafe fn test_mm_maskz_getexp_round_sh() { + let a = _mm_setr_ph(4.0, 10., 11., 12., 13., 14., 15., 16.); + let b = _mm_setr_ph(3.0, 20., 21., 22., 23., 24., 25., 26.); + let r = _mm_maskz_getexp_round_sh::<_MM_FROUND_NO_EXC>(0, a, b); + let e = _mm_setr_ph(0.0, 10., 11., 12., 13., 14., 15., 16.); + assert_eq_m128h(r, e); + let r = _mm_maskz_getexp_round_sh::<_MM_FROUND_NO_EXC>(1, a, b); + let e = _mm_setr_ph(1.0, 10., 11., 12., 13., 14., 15., 16.); + assert_eq_m128h(r, e); + } + + #[simd_test(enable = "avx512fp16,avx512vl")] + unsafe fn test_mm_getmant_ph() { + let a = _mm_set1_ph(10.0); + let r = _mm_getmant_ph::<_MM_MANT_NORM_P75_1P5, _MM_MANT_SIGN_NAN>(a); + let e = _mm_set1_ph(1.25); + assert_eq_m128h(r, e); + } + + #[simd_test(enable = "avx512fp16,avx512vl")] + unsafe fn test_mm_mask_getmant_ph() { + let a = _mm_set1_ph(10.0); + let src = _mm_set1_ph(20.0); + let r = _mm_mask_getmant_ph::<_MM_MANT_NORM_P75_1P5, _MM_MANT_SIGN_NAN>(src, 0b01010101, a); + let e = _mm_set_ph(20.0, 1.25, 20.0, 1.25, 20.0, 1.25, 20.0, 1.25); + assert_eq_m128h(r, e); + } + + #[simd_test(enable = "avx512fp16,avx512vl")] + unsafe fn test_mm_maskz_getmant_ph() { + let a = _mm_set1_ph(10.0); + let r = _mm_maskz_getmant_ph::<_MM_MANT_NORM_P75_1P5, _MM_MANT_SIGN_NAN>(0b01010101, a); + let e = _mm_set_ph(0.0, 1.25, 0.0, 1.25, 0.0, 1.25, 0.0, 1.25); + assert_eq_m128h(r, e); + } + + #[simd_test(enable = "avx512fp16,avx512vl")] + unsafe fn test_mm256_getmant_ph() { + let a = _mm256_set1_ph(10.0); + let r = _mm256_getmant_ph::<_MM_MANT_NORM_P75_1P5, _MM_MANT_SIGN_NAN>(a); + let e = _mm256_set1_ph(1.25); + assert_eq_m256h(r, e); + } + + #[simd_test(enable = "avx512fp16,avx512vl")] + unsafe fn test_mm256_mask_getmant_ph() { + let a = _mm256_set1_ph(10.0); + let src = _mm256_set1_ph(20.0); + let r = _mm256_mask_getmant_ph::<_MM_MANT_NORM_P75_1P5, _MM_MANT_SIGN_NAN>( + src, + 0b0101010101010101, + a, + ); + let e = _mm256_set_ph( + 20.0, 1.25, 20.0, 1.25, 20.0, 1.25, 20.0, 1.25, 20.0, 1.25, 20.0, 1.25, 20.0, 1.25, + 20.0, 1.25, + ); + assert_eq_m256h(r, e); + } + + #[simd_test(enable = "avx512fp16,avx512vl")] + unsafe fn test_mm256_maskz_getmant_ph() { + let a = _mm256_set1_ph(10.0); + let r = _mm256_maskz_getmant_ph::<_MM_MANT_NORM_P75_1P5, _MM_MANT_SIGN_NAN>( + 0b0101010101010101, + a, + ); + let e = _mm256_set_ph( + 0.0, 1.25, 0.0, 1.25, 0.0, 1.25, 0.0, 1.25, 0.0, 1.25, 0.0, 1.25, 0.0, 1.25, 0.0, 1.25, + ); + assert_eq_m256h(r, e); + } + + #[simd_test(enable = "avx512fp16")] + unsafe fn test_mm512_getmant_ph() { + let a = _mm512_set1_ph(10.0); + let r = _mm512_getmant_ph::<_MM_MANT_NORM_P75_1P5, _MM_MANT_SIGN_NAN>(a); + let e = _mm512_set1_ph(1.25); + assert_eq_m512h(r, e); + } + + #[simd_test(enable = "avx512fp16")] + unsafe fn test_mm512_mask_getmant_ph() { + let a = _mm512_set1_ph(10.0); + let src = _mm512_set1_ph(20.0); + let r = _mm512_mask_getmant_ph::<_MM_MANT_NORM_P75_1P5, _MM_MANT_SIGN_NAN>( + src, + 0b01010101010101010101010101010101, + a, + ); + let e = _mm512_set_ph( + 20.0, 1.25, 20.0, 1.25, 20.0, 1.25, 20.0, 1.25, 20.0, 1.25, 20.0, 1.25, 20.0, 1.25, + 20.0, 1.25, 20.0, 1.25, 20.0, 1.25, 20.0, 1.25, 20.0, 1.25, 20.0, 1.25, 20.0, 1.25, + 20.0, 1.25, 20.0, 1.25, + ); + assert_eq_m512h(r, e); + } + + #[simd_test(enable = "avx512fp16")] + unsafe fn test_mm512_maskz_getmant_ph() { + let a = _mm512_set1_ph(10.0); + let r = _mm512_maskz_getmant_ph::<_MM_MANT_NORM_P75_1P5, _MM_MANT_SIGN_NAN>( + 0b01010101010101010101010101010101, + a, + ); + let e = _mm512_set_ph( + 0.0, 1.25, 0.0, 1.25, 0.0, 1.25, 0.0, 1.25, 0.0, 1.25, 0.0, 1.25, 0.0, 1.25, 0.0, 1.25, + 0.0, 1.25, 0.0, 1.25, 0.0, 1.25, 0.0, 1.25, 0.0, 1.25, 0.0, 1.25, 0.0, 1.25, 0.0, 1.25, + ); + assert_eq_m512h(r, e); + } + + #[simd_test(enable = "avx512fp16")] + unsafe fn test_mm512_getmant_round_ph() { + let a = _mm512_set1_ph(10.0); + let r = + _mm512_getmant_round_ph::<_MM_MANT_NORM_P75_1P5, _MM_MANT_SIGN_NAN, _MM_FROUND_NO_EXC>( + a, + ); + let e = _mm512_set1_ph(1.25); + assert_eq_m512h(r, e); + } + + #[simd_test(enable = "avx512fp16")] + unsafe fn test_mm512_mask_getmant_round_ph() { + let a = _mm512_set1_ph(10.0); + let src = _mm512_set1_ph(20.0); + let r = _mm512_mask_getmant_round_ph::< + _MM_MANT_NORM_P75_1P5, + _MM_MANT_SIGN_NAN, + _MM_FROUND_NO_EXC, + >(src, 0b01010101010101010101010101010101, a); + let e = _mm512_set_ph( + 20.0, 1.25, 20.0, 1.25, 20.0, 1.25, 20.0, 1.25, 20.0, 1.25, 20.0, 1.25, 20.0, 1.25, + 20.0, 1.25, 20.0, 1.25, 20.0, 1.25, 20.0, 1.25, 20.0, 1.25, 20.0, 1.25, 20.0, 1.25, + 20.0, 1.25, 20.0, 1.25, + ); + assert_eq_m512h(r, e); + } + + #[simd_test(enable = "avx512fp16")] + unsafe fn test_mm512_maskz_getmant_round_ph() { + let a = _mm512_set1_ph(10.0); + let r = _mm512_maskz_getmant_round_ph::< + _MM_MANT_NORM_P75_1P5, + _MM_MANT_SIGN_NAN, + _MM_FROUND_NO_EXC, + >(0b01010101010101010101010101010101, a); + let e = _mm512_set_ph( + 0.0, 1.25, 0.0, 1.25, 0.0, 1.25, 0.0, 1.25, 0.0, 1.25, 0.0, 1.25, 0.0, 1.25, 0.0, 1.25, + 0.0, 1.25, 0.0, 1.25, 0.0, 1.25, 0.0, 1.25, 0.0, 1.25, 0.0, 1.25, 0.0, 1.25, 0.0, 1.25, + ); + assert_eq_m512h(r, e); + } + + #[simd_test(enable = "avx512fp16")] + unsafe fn test_mm_getmant_sh() { + let a = _mm_setr_ph(15.0, 10., 11., 12., 13., 14., 15., 16.); + let b = _mm_setr_ph(10.0, 20., 21., 22., 23., 24., 25., 26.); + let r = _mm_getmant_sh::<_MM_MANT_NORM_P75_1P5, _MM_MANT_SIGN_NAN>(a, b); + let e = _mm_setr_ph(1.25, 10., 11., 12., 13., 14., 15., 16.); + assert_eq_m128h(r, e); + } + + #[simd_test(enable = "avx512fp16")] + unsafe fn test_mm_mask_getmant_sh() { + let a = _mm_setr_ph(15.0, 10., 11., 12., 13., 14., 15., 16.); + let b = _mm_setr_ph(10.0, 20., 21., 22., 23., 24., 25., 26.); + let src = _mm_setr_ph(20.0, 30., 31., 32., 33., 34., 35., 36.); + let r = _mm_mask_getmant_sh::<_MM_MANT_NORM_P75_1P5, _MM_MANT_SIGN_NAN>(src, 0, a, b); + let e = _mm_setr_ph(20.0, 10., 11., 12., 13., 14., 15., 16.); + assert_eq_m128h(r, e); + let r = _mm_mask_getmant_sh::<_MM_MANT_NORM_P75_1P5, _MM_MANT_SIGN_NAN>(src, 1, a, b); + let e = _mm_setr_ph(1.25, 10., 11., 12., 13., 14., 15., 16.); + assert_eq_m128h(r, e); + } + + #[simd_test(enable = "avx512fp16")] + unsafe fn test_mm_maskz_getmant_sh() { + let a = _mm_setr_ph(15.0, 10., 11., 12., 13., 14., 15., 16.); + let b = _mm_setr_ph(10.0, 20., 21., 22., 23., 24., 25., 26.); + let r = _mm_maskz_getmant_sh::<_MM_MANT_NORM_P75_1P5, _MM_MANT_SIGN_NAN>(0, a, b); + let e = _mm_setr_ph(0.0, 10., 11., 12., 13., 14., 15., 16.); + assert_eq_m128h(r, e); + let r = _mm_maskz_getmant_sh::<_MM_MANT_NORM_P75_1P5, _MM_MANT_SIGN_NAN>(1, a, b); + let e = _mm_setr_ph(1.25, 10., 11., 12., 13., 14., 15., 16.); + assert_eq_m128h(r, e); + } + + #[simd_test(enable = "avx512fp16")] + unsafe fn test_mm_getmant_round_sh() { + let a = _mm_setr_ph(15.0, 10., 11., 12., 13., 14., 15., 16.); + let b = _mm_setr_ph(10.0, 20., 21., 22., 23., 24., 25., 26.); + let r = _mm_getmant_round_sh::<_MM_MANT_NORM_P75_1P5, _MM_MANT_SIGN_NAN, _MM_FROUND_NO_EXC>( + a, b, + ); + let e = _mm_setr_ph(1.25, 10., 11., 12., 13., 14., 15., 16.); + assert_eq_m128h(r, e); + } + + #[simd_test(enable = "avx512fp16")] + unsafe fn test_mm_mask_getmant_round_sh() { + let a = _mm_setr_ph(15.0, 10., 11., 12., 13., 14., 15., 16.); + let b = _mm_setr_ph(10.0, 20., 21., 22., 23., 24., 25., 26.); + let src = _mm_setr_ph(20.0, 30., 31., 32., 33., 34., 35., 36.); + let r = _mm_mask_getmant_round_sh::< + _MM_MANT_NORM_P75_1P5, + _MM_MANT_SIGN_NAN, + _MM_FROUND_NO_EXC, + >(src, 0, a, b); + let e = _mm_setr_ph(20.0, 10., 11., 12., 13., 14., 15., 16.); + assert_eq_m128h(r, e); + let r = _mm_mask_getmant_round_sh::< + _MM_MANT_NORM_P75_1P5, + _MM_MANT_SIGN_NAN, + _MM_FROUND_NO_EXC, + >(src, 1, a, b); + let e = _mm_setr_ph(1.25, 10., 11., 12., 13., 14., 15., 16.); + assert_eq_m128h(r, e); + } + + #[simd_test(enable = "avx512fp16")] + unsafe fn test_mm_maskz_getmant_round_sh() { + let a = _mm_setr_ph(15.0, 10., 11., 12., 13., 14., 15., 16.); + let b = _mm_setr_ph(10.0, 20., 21., 22., 23., 24., 25., 26.); + let r = _mm_maskz_getmant_round_sh::< + _MM_MANT_NORM_P75_1P5, + _MM_MANT_SIGN_NAN, + _MM_FROUND_NO_EXC, + >(0, a, b); + let e = _mm_setr_ph(0.0, 10., 11., 12., 13., 14., 15., 16.); + assert_eq_m128h(r, e); + let r = _mm_maskz_getmant_round_sh::< + _MM_MANT_NORM_P75_1P5, + _MM_MANT_SIGN_NAN, + _MM_FROUND_NO_EXC, + >(1, a, b); + let e = _mm_setr_ph(1.25, 10., 11., 12., 13., 14., 15., 16.); + assert_eq_m128h(r, e); + } + + #[simd_test(enable = "avx512fp16,avx512vl")] + unsafe fn test_mm_roundscale_ph() { + let a = _mm_set1_ph(1.1); + let r = _mm_roundscale_ph::<0>(a); + let e = _mm_set1_ph(1.0); + assert_eq_m128h(r, e); + } + + #[simd_test(enable = "avx512fp16,avx512vl")] + unsafe fn test_mm_mask_roundscale_ph() { + let a = _mm_set1_ph(1.1); + let src = _mm_set1_ph(2.0); + let r = _mm_mask_roundscale_ph::<0>(src, 0b01010101, a); + let e = _mm_set_ph(2.0, 1.0, 2.0, 1.0, 2.0, 1.0, 2.0, 1.0); + assert_eq_m128h(r, e); + } + + #[simd_test(enable = "avx512fp16,avx512vl")] + unsafe fn test_mm_maskz_roundscale_ph() { + let a = _mm_set1_ph(1.1); + let r = _mm_maskz_roundscale_ph::<0>(0b01010101, a); + let e = _mm_set_ph(0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0); + assert_eq_m128h(r, e); + } + + #[simd_test(enable = "avx512fp16,avx512vl")] + unsafe fn test_mm256_roundscale_ph() { + let a = _mm256_set1_ph(1.1); + let r = _mm256_roundscale_ph::<0>(a); + let e = _mm256_set1_ph(1.0); + assert_eq_m256h(r, e); + } + + #[simd_test(enable = "avx512fp16,avx512vl")] + unsafe fn test_mm256_mask_roundscale_ph() { + let a = _mm256_set1_ph(1.1); + let src = _mm256_set1_ph(2.0); + let r = _mm256_mask_roundscale_ph::<0>(src, 0b0101010101010101, a); + let e = _mm256_set_ph( + 2.0, 1.0, 2.0, 1.0, 2.0, 1.0, 2.0, 1.0, 2.0, 1.0, 2.0, 1.0, 2.0, 1.0, 2.0, 1.0, + ); + assert_eq_m256h(r, e); + } + + #[simd_test(enable = "avx512fp16,avx512vl")] + unsafe fn test_mm256_maskz_roundscale_ph() { + let a = _mm256_set1_ph(1.1); + let r = _mm256_maskz_roundscale_ph::<0>(0b0101010101010101, a); + let e = _mm256_set_ph( + 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, + ); + assert_eq_m256h(r, e); + } + + #[simd_test(enable = "avx512fp16")] + unsafe fn test_mm512_roundscale_ph() { + let a = _mm512_set1_ph(1.1); + let r = _mm512_roundscale_ph::<0>(a); + let e = _mm512_set1_ph(1.0); + assert_eq_m512h(r, e); + } + + #[simd_test(enable = "avx512fp16")] + unsafe fn test_mm512_mask_roundscale_ph() { + let a = _mm512_set1_ph(1.1); + let src = _mm512_set1_ph(2.0); + let r = _mm512_mask_roundscale_ph::<0>(src, 0b01010101010101010101010101010101, a); + let e = _mm512_set_ph( + 2.0, 1.0, 2.0, 1.0, 2.0, 1.0, 2.0, 1.0, 2.0, 1.0, 2.0, 1.0, 2.0, 1.0, 2.0, 1.0, 2.0, + 1.0, 2.0, 1.0, 2.0, 1.0, 2.0, 1.0, 2.0, 1.0, 2.0, 1.0, 2.0, 1.0, 2.0, 1.0, + ); + assert_eq_m512h(r, e); + } + + #[simd_test(enable = "avx512fp16")] + unsafe fn test_mm512_maskz_roundscale_ph() { + let a = _mm512_set1_ph(1.1); + let r = _mm512_maskz_roundscale_ph::<0>(0b01010101010101010101010101010101, a); + let e = _mm512_set_ph( + 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, + 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, + ); + assert_eq_m512h(r, e); + } + + #[simd_test(enable = "avx512fp16")] + unsafe fn test_mm512_roundscale_round_ph() { + let a = _mm512_set1_ph(1.1); + let r = _mm512_roundscale_round_ph::<0, _MM_FROUND_NO_EXC>(a); + let e = _mm512_set1_ph(1.0); + assert_eq_m512h(r, e); + } + + #[simd_test(enable = "avx512fp16")] + unsafe fn test_mm512_mask_roundscale_round_ph() { + let a = _mm512_set1_ph(1.1); + let src = _mm512_set1_ph(2.0); + let r = _mm512_mask_roundscale_round_ph::<0, _MM_FROUND_NO_EXC>( + src, + 0b01010101010101010101010101010101, + a, + ); + let e = _mm512_set_ph( + 2.0, 1.0, 2.0, 1.0, 2.0, 1.0, 2.0, 1.0, 2.0, 1.0, 2.0, 1.0, 2.0, 1.0, 2.0, 1.0, 2.0, + 1.0, 2.0, 1.0, 2.0, 1.0, 2.0, 1.0, 2.0, 1.0, 2.0, 1.0, 2.0, 1.0, 2.0, 1.0, + ); + assert_eq_m512h(r, e); + } + + #[simd_test(enable = "avx512fp16")] + unsafe fn test_mm512_maskz_roundscale_round_ph() { + let a = _mm512_set1_ph(1.1); + let r = _mm512_maskz_roundscale_round_ph::<0, _MM_FROUND_NO_EXC>( + 0b01010101010101010101010101010101, + a, + ); + let e = _mm512_set_ph( + 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, + 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, + ); + assert_eq_m512h(r, e); + } + + #[simd_test(enable = "avx512fp16")] + unsafe fn test_mm_roundscale_sh() { + let a = _mm_setr_ph(2.0, 10., 11., 12., 13., 14., 15., 16.); + let b = _mm_setr_ph(1.1, 20., 21., 22., 23., 24., 25., 26.); + let r = _mm_roundscale_sh::<0>(a, b); + let e = _mm_setr_ph(1.0, 10., 11., 12., 13., 14., 15., 16.); + assert_eq_m128h(r, e); + } + + #[simd_test(enable = "avx512fp16")] + unsafe fn test_mm_mask_roundscale_sh() { + let a = _mm_setr_ph(2.0, 10., 11., 12., 13., 14., 15., 16.); + let b = _mm_setr_ph(1.1, 20., 21., 22., 23., 24., 25., 26.); + let src = _mm_setr_ph(3.0, 30., 31., 32., 33., 34., 35., 36.); + let r = _mm_mask_roundscale_sh::<0>(src, 0, a, b); + let e = _mm_setr_ph(3.0, 10., 11., 12., 13., 14., 15., 16.); + assert_eq_m128h(r, e); + let r = _mm_mask_roundscale_sh::<0>(src, 1, a, b); + let e = _mm_setr_ph(1.0, 10., 11., 12., 13., 14., 15., 16.); + assert_eq_m128h(r, e); + } + + #[simd_test(enable = "avx512fp16")] + unsafe fn test_mm_maskz_roundscale_sh() { + let a = _mm_setr_ph(2.0, 10., 11., 12., 13., 14., 15., 16.); + let b = _mm_setr_ph(1.1, 20., 21., 22., 23., 24., 25., 26.); + let r = _mm_maskz_roundscale_sh::<0>(0, a, b); + let e = _mm_setr_ph(0.0, 10., 11., 12., 13., 14., 15., 16.); + assert_eq_m128h(r, e); + let r = _mm_maskz_roundscale_sh::<0>(1, a, b); + let e = _mm_setr_ph(1.0, 10., 11., 12., 13., 14., 15., 16.); + assert_eq_m128h(r, e); + } + + #[simd_test(enable = "avx512fp16")] + unsafe fn test_mm_roundscale_round_sh() { + let a = _mm_setr_ph(2.0, 10., 11., 12., 13., 14., 15., 16.); + let b = _mm_setr_ph(1.1, 20., 21., 22., 23., 24., 25., 26.); + let r = _mm_roundscale_round_sh::<0, _MM_FROUND_NO_EXC>(a, b); + let e = _mm_setr_ph(1.0, 10., 11., 12., 13., 14., 15., 16.); + assert_eq_m128h(r, e); + } + + #[simd_test(enable = "avx512fp16")] + unsafe fn test_mm_mask_roundscale_round_sh() { + let a = _mm_setr_ph(2.0, 10., 11., 12., 13., 14., 15., 16.); + let b = _mm_setr_ph(1.1, 20., 21., 22., 23., 24., 25., 26.); + let src = _mm_setr_ph(3.0, 30., 31., 32., 33., 34., 35., 36.); + let r = _mm_mask_roundscale_round_sh::<0, _MM_FROUND_NO_EXC>(src, 0, a, b); + let e = _mm_setr_ph(3.0, 10., 11., 12., 13., 14., 15., 16.); + assert_eq_m128h(r, e); + let r = _mm_mask_roundscale_round_sh::<0, _MM_FROUND_NO_EXC>(src, 1, a, b); + let e = _mm_setr_ph(1.0, 10., 11., 12., 13., 14., 15., 16.); + assert_eq_m128h(r, e); + } + + #[simd_test(enable = "avx512fp16")] + unsafe fn test_mm_maskz_roundscale_round_sh() { + let a = _mm_setr_ph(2.0, 10., 11., 12., 13., 14., 15., 16.); + let b = _mm_setr_ph(1.1, 20., 21., 22., 23., 24., 25., 26.); + let r = _mm_maskz_roundscale_round_sh::<0, _MM_FROUND_NO_EXC>(0, a, b); + let e = _mm_setr_ph(0.0, 10., 11., 12., 13., 14., 15., 16.); + assert_eq_m128h(r, e); + let r = _mm_maskz_roundscale_round_sh::<0, _MM_FROUND_NO_EXC>(1, a, b); + let e = _mm_setr_ph(1.0, 10., 11., 12., 13., 14., 15., 16.); + assert_eq_m128h(r, e); + } + + #[simd_test(enable = "avx512fp16,avx512vl")] + unsafe fn test_mm_scalef_ph() { + let a = _mm_set1_ph(1.); + let b = _mm_set1_ph(3.); + let r = _mm_scalef_ph(a, b); + let e = _mm_set1_ph(8.0); + assert_eq_m128h(r, e); + } + + #[simd_test(enable = "avx512fp16,avx512vl")] + unsafe fn test_mm_mask_scalef_ph() { + let a = _mm_set1_ph(1.); + let b = _mm_set1_ph(3.); + let src = _mm_set1_ph(2.); + let r = _mm_mask_scalef_ph(src, 0b01010101, a, b); + let e = _mm_set_ph(2.0, 8.0, 2.0, 8.0, 2.0, 8.0, 2.0, 8.0); + assert_eq_m128h(r, e); + } + + #[simd_test(enable = "avx512fp16,avx512vl")] + unsafe fn test_mm_maskz_scalef_ph() { + let a = _mm_set1_ph(1.); + let b = _mm_set1_ph(3.); + let r = _mm_maskz_scalef_ph(0b01010101, a, b); + let e = _mm_set_ph(0.0, 8.0, 0.0, 8.0, 0.0, 8.0, 0.0, 8.0); + assert_eq_m128h(r, e); + } + + #[simd_test(enable = "avx512fp16,avx512vl")] + unsafe fn test_mm256_scalef_ph() { + let a = _mm256_set1_ph(1.); + let b = _mm256_set1_ph(3.); + let r = _mm256_scalef_ph(a, b); + let e = _mm256_set1_ph(8.0); + assert_eq_m256h(r, e); + } + + #[simd_test(enable = "avx512fp16,avx512vl")] + unsafe fn test_mm256_mask_scalef_ph() { + let a = _mm256_set1_ph(1.); + let b = _mm256_set1_ph(3.); + let src = _mm256_set1_ph(2.); + let r = _mm256_mask_scalef_ph(src, 0b0101010101010101, a, b); + let e = _mm256_set_ph( + 2.0, 8.0, 2.0, 8.0, 2.0, 8.0, 2.0, 8.0, 2.0, 8.0, 2.0, 8.0, 2.0, 8.0, 2.0, 8.0, + ); + assert_eq_m256h(r, e); + } + + #[simd_test(enable = "avx512fp16,avx512vl")] + unsafe fn test_mm256_maskz_scalef_ph() { + let a = _mm256_set1_ph(1.); + let b = _mm256_set1_ph(3.); + let r = _mm256_maskz_scalef_ph(0b0101010101010101, a, b); + let e = _mm256_set_ph( + 0.0, 8.0, 0.0, 8.0, 0.0, 8.0, 0.0, 8.0, 0.0, 8.0, 0.0, 8.0, 0.0, 8.0, 0.0, 8.0, + ); + assert_eq_m256h(r, e); + } + + #[simd_test(enable = "avx512fp16")] + unsafe fn test_mm512_scalef_ph() { + let a = _mm512_set1_ph(1.); + let b = _mm512_set1_ph(3.); + let r = _mm512_scalef_ph(a, b); + let e = _mm512_set1_ph(8.0); + assert_eq_m512h(r, e); + } + + #[simd_test(enable = "avx512fp16")] + unsafe fn test_mm512_mask_scalef_ph() { + let a = _mm512_set1_ph(1.); + let b = _mm512_set1_ph(3.); + let src = _mm512_set1_ph(2.); + let r = _mm512_mask_scalef_ph(src, 0b01010101010101010101010101010101, a, b); + let e = _mm512_set_ph( + 2.0, 8.0, 2.0, 8.0, 2.0, 8.0, 2.0, 8.0, 2.0, 8.0, 2.0, 8.0, 2.0, 8.0, 2.0, 8.0, 2.0, + 8.0, 2.0, 8.0, 2.0, 8.0, 2.0, 8.0, 2.0, 8.0, 2.0, 8.0, 2.0, 8.0, 2.0, 8.0, + ); + assert_eq_m512h(r, e); + } + + #[simd_test(enable = "avx512fp16")] + unsafe fn test_mm512_maskz_scalef_ph() { + let a = _mm512_set1_ph(1.); + let b = _mm512_set1_ph(3.); + let r = _mm512_maskz_scalef_ph(0b01010101010101010101010101010101, a, b); + let e = _mm512_set_ph( + 0.0, 8.0, 0.0, 8.0, 0.0, 8.0, 0.0, 8.0, 0.0, 8.0, 0.0, 8.0, 0.0, 8.0, 0.0, 8.0, 0.0, + 8.0, 0.0, 8.0, 0.0, 8.0, 0.0, 8.0, 0.0, 8.0, 0.0, 8.0, 0.0, 8.0, 0.0, 8.0, + ); + assert_eq_m512h(r, e); + } + + #[simd_test(enable = "avx512fp16")] + unsafe fn test_mm512_scalef_round_ph() { + let a = _mm512_set1_ph(1.); + let b = _mm512_set1_ph(3.); + let r = _mm512_scalef_round_ph::<{ _MM_FROUND_TO_NEAREST_INT | _MM_FROUND_NO_EXC }>(a, b); + let e = _mm512_set1_ph(8.0); + assert_eq_m512h(r, e); + } + + #[simd_test(enable = "avx512fp16")] + unsafe fn test_mm512_mask_scalef_round_ph() { + let a = _mm512_set1_ph(1.); + let b = _mm512_set1_ph(3.); + let src = _mm512_set1_ph(2.); + let r = _mm512_mask_scalef_round_ph::<{ _MM_FROUND_TO_NEAREST_INT | _MM_FROUND_NO_EXC }>( + src, + 0b01010101010101010101010101010101, + a, + b, + ); + let e = _mm512_set_ph( + 2.0, 8.0, 2.0, 8.0, 2.0, 8.0, 2.0, 8.0, 2.0, 8.0, 2.0, 8.0, 2.0, 8.0, 2.0, 8.0, 2.0, + 8.0, 2.0, 8.0, 2.0, 8.0, 2.0, 8.0, 2.0, 8.0, 2.0, 8.0, 2.0, 8.0, 2.0, 8.0, + ); + assert_eq_m512h(r, e); + } + + #[simd_test(enable = "avx512fp16")] + unsafe fn test_mm512_maskz_scalef_round_ph() { + let a = _mm512_set1_ph(1.); + let b = _mm512_set1_ph(3.); + let r = _mm512_maskz_scalef_round_ph::<{ _MM_FROUND_TO_NEAREST_INT | _MM_FROUND_NO_EXC }>( + 0b01010101010101010101010101010101, + a, + b, + ); + let e = _mm512_set_ph( + 0.0, 8.0, 0.0, 8.0, 0.0, 8.0, 0.0, 8.0, 0.0, 8.0, 0.0, 8.0, 0.0, 8.0, 0.0, 8.0, 0.0, + 8.0, 0.0, 8.0, 0.0, 8.0, 0.0, 8.0, 0.0, 8.0, 0.0, 8.0, 0.0, 8.0, 0.0, 8.0, + ); + assert_eq_m512h(r, e); + } + + #[simd_test(enable = "avx512fp16")] + unsafe fn test_mm_scalef_sh() { + let a = _mm_setr_ph(1.0, 10., 11., 12., 13., 14., 15., 16.); + let b = _mm_setr_ph(3.0, 20., 21., 22., 23., 24., 25., 26.); + let r = _mm_scalef_sh(a, b); + let e = _mm_setr_ph(8.0, 10., 11., 12., 13., 14., 15., 16.); + assert_eq_m128h(r, e); + } + + #[simd_test(enable = "avx512fp16")] + unsafe fn test_mm_mask_scalef_sh() { + let a = _mm_setr_ph(1.0, 10., 11., 12., 13., 14., 15., 16.); + let b = _mm_setr_ph(3.0, 20., 21., 22., 23., 24., 25., 26.); + let src = _mm_setr_ph(2.0, 30., 31., 32., 33., 34., 35., 36.); + let r = _mm_mask_scalef_sh(src, 0, a, b); + let e = _mm_setr_ph(2.0, 10., 11., 12., 13., 14., 15., 16.); + assert_eq_m128h(r, e); + let r = _mm_mask_scalef_sh(src, 1, a, b); + let e = _mm_setr_ph(8.0, 10., 11., 12., 13., 14., 15., 16.); + assert_eq_m128h(r, e); + } + + #[simd_test(enable = "avx512fp16")] + unsafe fn test_mm_maskz_scalef_sh() { + let a = _mm_setr_ph(1.0, 10., 11., 12., 13., 14., 15., 16.); + let b = _mm_setr_ph(3.0, 20., 21., 22., 23., 24., 25., 26.); + let r = _mm_maskz_scalef_sh(0, a, b); + let e = _mm_setr_ph(0.0, 10., 11., 12., 13., 14., 15., 16.); + assert_eq_m128h(r, e); + let r = _mm_maskz_scalef_sh(1, a, b); + let e = _mm_setr_ph(8.0, 10., 11., 12., 13., 14., 15., 16.); + assert_eq_m128h(r, e); + } + + #[simd_test(enable = "avx512fp16")] + unsafe fn test_mm_scalef_round_sh() { + let a = _mm_setr_ph(1.0, 10., 11., 12., 13., 14., 15., 16.); + let b = _mm_setr_ph(3.0, 20., 21., 22., 23., 24., 25., 26.); + let r = _mm_scalef_round_sh::<{ _MM_FROUND_TO_NEAREST_INT | _MM_FROUND_NO_EXC }>(a, b); + let e = _mm_setr_ph(8.0, 10., 11., 12., 13., 14., 15., 16.); + assert_eq_m128h(r, e); + } + + #[simd_test(enable = "avx512fp16")] + unsafe fn test_mm_mask_scalef_round_sh() { + let a = _mm_setr_ph(1.0, 10., 11., 12., 13., 14., 15., 16.); + let b = _mm_setr_ph(3.0, 20., 21., 22., 23., 24., 25., 26.); + let src = _mm_setr_ph(2.0, 30., 31., 32., 33., 34., 35., 36.); + let r = _mm_mask_scalef_round_sh::<{ _MM_FROUND_TO_NEAREST_INT | _MM_FROUND_NO_EXC }>( + src, 0, a, b, + ); + let e = _mm_setr_ph(2.0, 10., 11., 12., 13., 14., 15., 16.); + assert_eq_m128h(r, e); + let r = _mm_mask_scalef_round_sh::<{ _MM_FROUND_TO_NEAREST_INT | _MM_FROUND_NO_EXC }>( + src, 1, a, b, + ); + let e = _mm_setr_ph(8.0, 10., 11., 12., 13., 14., 15., 16.); + assert_eq_m128h(r, e); + } + + #[simd_test(enable = "avx512fp16")] + unsafe fn test_mm_maskz_scalef_round_sh() { + let a = _mm_setr_ph(1.0, 10., 11., 12., 13., 14., 15., 16.); + let b = _mm_setr_ph(3.0, 20., 21., 22., 23., 24., 25., 26.); + let r = + _mm_maskz_scalef_round_sh::<{ _MM_FROUND_TO_NEAREST_INT | _MM_FROUND_NO_EXC }>(0, a, b); + let e = _mm_setr_ph(0.0, 10., 11., 12., 13., 14., 15., 16.); + assert_eq_m128h(r, e); + let r = + _mm_maskz_scalef_round_sh::<{ _MM_FROUND_TO_NEAREST_INT | _MM_FROUND_NO_EXC }>(1, a, b); + let e = _mm_setr_ph(8.0, 10., 11., 12., 13., 14., 15., 16.); + assert_eq_m128h(r, e); + } + + #[simd_test(enable = "avx512fp16,avx512vl")] + unsafe fn test_mm_reduce_ph() { + let a = _mm_set1_ph(1.25); + let r = _mm_reduce_ph::<{ 16 | _MM_FROUND_TO_ZERO }>(a); + let e = _mm_set1_ph(0.25); + assert_eq_m128h(r, e); + } + + #[simd_test(enable = "avx512fp16,avx512vl")] + unsafe fn test_mm_mask_reduce_ph() { + let a = _mm_set1_ph(1.25); + let src = _mm_set1_ph(2.0); + let r = _mm_mask_reduce_ph::<{ 16 | _MM_FROUND_TO_ZERO }>(src, 0b01010101, a); + let e = _mm_set_ph(2.0, 0.25, 2.0, 0.25, 2.0, 0.25, 2.0, 0.25); + assert_eq_m128h(r, e); + } + + #[simd_test(enable = "avx512fp16,avx512vl")] + unsafe fn test_mm_maskz_reduce_ph() { + let a = _mm_set1_ph(1.25); + let r = _mm_maskz_reduce_ph::<{ 16 | _MM_FROUND_TO_ZERO }>(0b01010101, a); + let e = _mm_set_ph(0.0, 0.25, 0.0, 0.25, 0.0, 0.25, 0.0, 0.25); + assert_eq_m128h(r, e); + } + + #[simd_test(enable = "avx512fp16,avx512vl")] + unsafe fn test_mm256_reduce_ph() { + let a = _mm256_set1_ph(1.25); + let r = _mm256_reduce_ph::<{ 16 | _MM_FROUND_TO_ZERO }>(a); + let e = _mm256_set1_ph(0.25); + assert_eq_m256h(r, e); + } + + #[simd_test(enable = "avx512fp16,avx512vl")] + unsafe fn test_mm256_mask_reduce_ph() { + let a = _mm256_set1_ph(1.25); + let src = _mm256_set1_ph(2.0); + let r = _mm256_mask_reduce_ph::<{ 16 | _MM_FROUND_TO_ZERO }>(src, 0b0101010101010101, a); + let e = _mm256_set_ph( + 2.0, 0.25, 2.0, 0.25, 2.0, 0.25, 2.0, 0.25, 2.0, 0.25, 2.0, 0.25, 2.0, 0.25, 2.0, 0.25, + ); + assert_eq_m256h(r, e); + } + + #[simd_test(enable = "avx512fp16,avx512vl")] + unsafe fn test_mm256_maskz_reduce_ph() { + let a = _mm256_set1_ph(1.25); + let r = _mm256_maskz_reduce_ph::<{ 16 | _MM_FROUND_TO_ZERO }>(0b0101010101010101, a); + let e = _mm256_set_ph( + 0.0, 0.25, 0.0, 0.25, 0.0, 0.25, 0.0, 0.25, 0.0, 0.25, 0.0, 0.25, 0.0, 0.25, 0.0, 0.25, + ); + assert_eq_m256h(r, e); + } + + #[simd_test(enable = "avx512fp16")] + unsafe fn test_mm512_reduce_ph() { + let a = _mm512_set1_ph(1.25); + let r = _mm512_reduce_ph::<{ 16 | _MM_FROUND_TO_ZERO }>(a); + let e = _mm512_set1_ph(0.25); + assert_eq_m512h(r, e); + } + + #[simd_test(enable = "avx512fp16")] + unsafe fn test_mm512_mask_reduce_ph() { + let a = _mm512_set1_ph(1.25); + let src = _mm512_set1_ph(2.0); + let r = _mm512_mask_reduce_ph::<{ 16 | _MM_FROUND_TO_ZERO }>( + src, + 0b01010101010101010101010101010101, + a, + ); + let e = _mm512_set_ph( + 2.0, 0.25, 2.0, 0.25, 2.0, 0.25, 2.0, 0.25, 2.0, 0.25, 2.0, 0.25, 2.0, 0.25, 2.0, 0.25, + 2.0, 0.25, 2.0, 0.25, 2.0, 0.25, 2.0, 0.25, 2.0, 0.25, 2.0, 0.25, 2.0, 0.25, 2.0, 0.25, + ); + assert_eq_m512h(r, e); + } + + #[simd_test(enable = "avx512fp16")] + unsafe fn test_mm512_maskz_reduce_ph() { + let a = _mm512_set1_ph(1.25); + let r = _mm512_maskz_reduce_ph::<{ 16 | _MM_FROUND_TO_ZERO }>( + 0b01010101010101010101010101010101, + a, + ); + let e = _mm512_set_ph( + 0.0, 0.25, 0.0, 0.25, 0.0, 0.25, 0.0, 0.25, 0.0, 0.25, 0.0, 0.25, 0.0, 0.25, 0.0, 0.25, + 0.0, 0.25, 0.0, 0.25, 0.0, 0.25, 0.0, 0.25, 0.0, 0.25, 0.0, 0.25, 0.0, 0.25, 0.0, 0.25, + ); + assert_eq_m512h(r, e); + } + + #[simd_test(enable = "avx512fp16")] + unsafe fn test_mm512_reduce_round_ph() { + let a = _mm512_set1_ph(1.25); + let r = _mm512_reduce_round_ph::<{ 16 | _MM_FROUND_TO_ZERO }, _MM_FROUND_NO_EXC>(a); + let e = _mm512_set1_ph(0.25); + assert_eq_m512h(r, e); + } + + #[simd_test(enable = "avx512fp16")] + unsafe fn test_mm512_mask_reduce_round_ph() { + let a = _mm512_set1_ph(1.25); + let src = _mm512_set1_ph(2.0); + let r = _mm512_mask_reduce_round_ph::<{ 16 | _MM_FROUND_TO_ZERO }, _MM_FROUND_NO_EXC>( + src, + 0b01010101010101010101010101010101, + a, + ); + let e = _mm512_set_ph( + 2.0, 0.25, 2.0, 0.25, 2.0, 0.25, 2.0, 0.25, 2.0, 0.25, 2.0, 0.25, 2.0, 0.25, 2.0, 0.25, + 2.0, 0.25, 2.0, 0.25, 2.0, 0.25, 2.0, 0.25, 2.0, 0.25, 2.0, 0.25, 2.0, 0.25, 2.0, 0.25, + ); + assert_eq_m512h(r, e); + } + + #[simd_test(enable = "avx512fp16")] + unsafe fn test_mm512_maskz_reduce_round_ph() { + let a = _mm512_set1_ph(1.25); + let r = _mm512_maskz_reduce_round_ph::<{ 16 | _MM_FROUND_TO_ZERO }, _MM_FROUND_NO_EXC>( + 0b01010101010101010101010101010101, + a, + ); + let e = _mm512_set_ph( + 0.0, 0.25, 0.0, 0.25, 0.0, 0.25, 0.0, 0.25, 0.0, 0.25, 0.0, 0.25, 0.0, 0.25, 0.0, 0.25, + 0.0, 0.25, 0.0, 0.25, 0.0, 0.25, 0.0, 0.25, 0.0, 0.25, 0.0, 0.25, 0.0, 0.25, 0.0, 0.25, + ); + assert_eq_m512h(r, e); + } + + #[simd_test(enable = "avx512fp16")] + unsafe fn test_mm_reduce_sh() { + let a = _mm_setr_ph(3.0, 10., 11., 12., 13., 14., 15., 16.); + let b = _mm_setr_ph(1.25, 20., 21., 22., 23., 24., 25., 26.); + let r = _mm_reduce_sh::<{ 16 | _MM_FROUND_TO_ZERO }>(a, b); + let e = _mm_setr_ph(0.25, 10., 11., 12., 13., 14., 15., 16.); + assert_eq_m128h(r, e); + } + + #[simd_test(enable = "avx512fp16")] + unsafe fn test_mm_mask_reduce_sh() { + let a = _mm_setr_ph(3.0, 10., 11., 12., 13., 14., 15., 16.); + let b = _mm_setr_ph(1.25, 20., 21., 22., 23., 24., 25., 26.); + let src = _mm_setr_ph(2.0, 30., 31., 32., 33., 34., 35., 36.); + let r = _mm_mask_reduce_sh::<{ 16 | _MM_FROUND_TO_ZERO }>(src, 0, a, b); + let e = _mm_setr_ph(2.0, 10., 11., 12., 13., 14., 15., 16.); + assert_eq_m128h(r, e); + let r = _mm_mask_reduce_sh::<{ 16 | _MM_FROUND_TO_ZERO }>(src, 1, a, b); + let e = _mm_setr_ph(0.25, 10., 11., 12., 13., 14., 15., 16.); + assert_eq_m128h(r, e); + } + + #[simd_test(enable = "avx512fp16")] + unsafe fn test_mm_maskz_reduce_sh() { + let a = _mm_setr_ph(3.0, 10., 11., 12., 13., 14., 15., 16.); + let b = _mm_setr_ph(1.25, 20., 21., 22., 23., 24., 25., 26.); + let r = _mm_maskz_reduce_sh::<{ 16 | _MM_FROUND_TO_ZERO }>(0, a, b); + let e = _mm_setr_ph(0.0, 10., 11., 12., 13., 14., 15., 16.); + assert_eq_m128h(r, e); + let r = _mm_maskz_reduce_sh::<{ 16 | _MM_FROUND_TO_ZERO }>(1, a, b); + let e = _mm_setr_ph(0.25, 10., 11., 12., 13., 14., 15., 16.); + assert_eq_m128h(r, e); + } + + #[simd_test(enable = "avx512fp16")] + unsafe fn test_mm_reduce_round_sh() { + let a = _mm_setr_ph(3.0, 10., 11., 12., 13., 14., 15., 16.); + let b = _mm_setr_ph(1.25, 20., 21., 22., 23., 24., 25., 26.); + let r = _mm_reduce_round_sh::<{ 16 | _MM_FROUND_TO_ZERO }, _MM_FROUND_NO_EXC>(a, b); + let e = _mm_setr_ph(0.25, 10., 11., 12., 13., 14., 15., 16.); + assert_eq_m128h(r, e); + } + + #[simd_test(enable = "avx512fp16")] + unsafe fn test_mm_mask_reduce_round_sh() { + let a = _mm_setr_ph(3.0, 10., 11., 12., 13., 14., 15., 16.); + let b = _mm_setr_ph(1.25, 20., 21., 22., 23., 24., 25., 26.); + let src = _mm_setr_ph(2.0, 30., 31., 32., 33., 34., 35., 36.); + let r = _mm_mask_reduce_round_sh::<{ 16 | _MM_FROUND_TO_ZERO }, _MM_FROUND_NO_EXC>( + src, 0, a, b, + ); + let e = _mm_setr_ph(2.0, 10., 11., 12., 13., 14., 15., 16.); + assert_eq_m128h(r, e); + let r = _mm_mask_reduce_round_sh::<{ 16 | _MM_FROUND_TO_ZERO }, _MM_FROUND_NO_EXC>( + src, 1, a, b, + ); + let e = _mm_setr_ph(0.25, 10., 11., 12., 13., 14., 15., 16.); + assert_eq_m128h(r, e); + } + + #[simd_test(enable = "avx512fp16")] + unsafe fn test_mm_maskz_reduce_round_sh() { + let a = _mm_setr_ph(3.0, 10., 11., 12., 13., 14., 15., 16.); + let b = _mm_setr_ph(1.25, 20., 21., 22., 23., 24., 25., 26.); + let r = + _mm_maskz_reduce_round_sh::<{ 16 | _MM_FROUND_TO_ZERO }, _MM_FROUND_NO_EXC>(0, a, b); + let e = _mm_setr_ph(0.0, 10., 11., 12., 13., 14., 15., 16.); + assert_eq_m128h(r, e); + let r = + _mm_maskz_reduce_round_sh::<{ 16 | _MM_FROUND_TO_ZERO }, _MM_FROUND_NO_EXC>(1, a, b); + let e = _mm_setr_ph(0.25, 10., 11., 12., 13., 14., 15., 16.); + assert_eq_m128h(r, e); + } }