From 25a978a69f3cbb5a50687ed47ed20fc3e4eb0242 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Eduardo=20S=C3=A1nchez=20Mu=C3=B1oz?= Date: Sat, 12 Aug 2023 11:50:00 +0200 Subject: [PATCH] Implement AVX512F 64-bit shift by immediate (slli_epi64) with `simd_shl` instead of LLVM intrinsics --- .../crates/core_arch/src/x86/avx512f.rs | 76 +++++++++++-------- 1 file changed, 44 insertions(+), 32 deletions(-) diff --git a/library/stdarch/crates/core_arch/src/x86/avx512f.rs b/library/stdarch/crates/core_arch/src/x86/avx512f.rs index 5e567eb1112c..acfe6525337e 100644 --- a/library/stdarch/crates/core_arch/src/x86/avx512f.rs +++ b/library/stdarch/crates/core_arch/src/x86/avx512f.rs @@ -17387,9 +17387,11 @@ pub unsafe fn _mm_maskz_srli_epi32(k: __mmask8, a: __m128i) -> #[rustc_legacy_const_generics(1)] pub unsafe fn _mm512_slli_epi64(a: __m512i) -> __m512i { static_assert_uimm_bits!(IMM8, 8); - let a = a.as_i64x8(); - let r = vpslliq(a, IMM8); - transmute(r) + if IMM8 >= 64 { + _mm512_setzero_si512() + } else { + transmute(simd_shl(a.as_u64x8(), u64x8::splat(IMM8 as u64))) + } } /// Shift packed 64-bit integers in a left by imm8 while shifting in zeros, and store the results in dst using writemask k (elements are copied from src when the corresponding mask bit is not set). @@ -17405,9 +17407,12 @@ pub unsafe fn _mm512_mask_slli_epi64( a: __m512i, ) -> __m512i { static_assert_uimm_bits!(IMM8, 8); - let a = a.as_i64x8(); - let shf = vpslliq(a, IMM8); - transmute(simd_select_bitmask(k, shf, src.as_i64x8())) + let shf = if IMM8 >= 64 { + u64x8::splat(0) + } else { + simd_shl(a.as_u64x8(), u64x8::splat(IMM8 as u64)) + }; + transmute(simd_select_bitmask(k, shf, src.as_u64x8())) } /// Shift packed 64-bit integers in a left by imm8 while shifting in zeros, and store the results in dst using zeromask k (elements are zeroed out when the corresponding mask bit is not set). @@ -17419,10 +17424,13 @@ pub unsafe fn _mm512_mask_slli_epi64( #[rustc_legacy_const_generics(2)] pub unsafe fn _mm512_maskz_slli_epi64(k: __mmask8, a: __m512i) -> __m512i { static_assert_uimm_bits!(IMM8, 8); - let a = a.as_i64x8(); - let shf = vpslliq(a, IMM8); - let zero = _mm512_setzero_si512().as_i64x8(); - transmute(simd_select_bitmask(k, shf, zero)) + if IMM8 >= 64 { + _mm512_setzero_si512() + } else { + let shf = simd_shl(a.as_u64x8(), u64x8::splat(IMM8 as u64)); + let zero = u64x8::splat(0); + transmute(simd_select_bitmask(k, shf, zero)) + } } /// Shift packed 64-bit integers in a left by imm8 while shifting in zeros, and store the results in dst using writemask k (elements are copied from src when the corresponding mask bit is not set). @@ -17438,9 +17446,12 @@ pub unsafe fn _mm256_mask_slli_epi64( a: __m256i, ) -> __m256i { static_assert_uimm_bits!(IMM8, 8); - let imm8 = IMM8 as i32; - let r = pslliq256(a.as_i64x4(), imm8); - transmute(simd_select_bitmask(k, r, src.as_i64x4())) + let r = if IMM8 >= 64 { + u64x4::splat(0) + } else { + simd_shl(a.as_u64x4(), u64x4::splat(IMM8 as u64)) + }; + transmute(simd_select_bitmask(k, r, src.as_u64x4())) } /// Shift packed 64-bit integers in a left by imm8 while shifting in zeros, and store the results in dst using zeromask k (elements are zeroed out when the corresponding mask bit is not set). @@ -17452,10 +17463,13 @@ pub unsafe fn _mm256_mask_slli_epi64( #[rustc_legacy_const_generics(2)] pub unsafe fn _mm256_maskz_slli_epi64(k: __mmask8, a: __m256i) -> __m256i { static_assert_uimm_bits!(IMM8, 8); - let imm8 = IMM8 as i32; - let r = pslliq256(a.as_i64x4(), imm8); - let zero = _mm256_setzero_si256().as_i64x4(); - transmute(simd_select_bitmask(k, r, zero)) + if IMM8 >= 64 { + _mm256_setzero_si256() + } else { + let r = simd_shl(a.as_u64x4(), u64x4::splat(IMM8 as u64)); + let zero = u64x4::splat(0); + transmute(simd_select_bitmask(k, r, zero)) + } } /// Shift packed 64-bit integers in a left by imm8 while shifting in zeros, and store the results in dst using writemask k (elements are copied from src when the corresponding mask bit is not set). @@ -17471,9 +17485,12 @@ pub unsafe fn _mm_mask_slli_epi64( a: __m128i, ) -> __m128i { static_assert_uimm_bits!(IMM8, 8); - let imm8 = IMM8 as i32; - let r = pslliq128(a.as_i64x2(), imm8); - transmute(simd_select_bitmask(k, r, src.as_i64x2())) + let r = if IMM8 >= 64 { + u64x2::splat(0) + } else { + simd_shl(a.as_u64x2(), u64x2::splat(IMM8 as u64)) + }; + transmute(simd_select_bitmask(k, r, src.as_u64x2())) } /// Shift packed 64-bit integers in a left by imm8 while shifting in zeros, and store the results in dst using zeromask k (elements are zeroed out when the corresponding mask bit is not set). @@ -17485,10 +17502,13 @@ pub unsafe fn _mm_mask_slli_epi64( #[rustc_legacy_const_generics(2)] pub unsafe fn _mm_maskz_slli_epi64(k: __mmask8, a: __m128i) -> __m128i { static_assert_uimm_bits!(IMM8, 8); - let imm8 = IMM8 as i32; - let r = pslliq128(a.as_i64x2(), imm8); - let zero = _mm_setzero_si128().as_i64x2(); - transmute(simd_select_bitmask(k, r, zero)) + if IMM8 >= 64 { + _mm_setzero_si128() + } else { + let r = simd_shl(a.as_u64x2(), u64x2::splat(IMM8 as u64)); + let zero = u64x2::splat(0); + transmute(simd_select_bitmask(k, r, zero)) + } } /// Shift packed 64-bit integers in a right by imm8 while shifting in zeros, and store the results in dst. @@ -38477,14 +38497,6 @@ extern "C" { #[link_name = "llvm.x86.sse2.psrli.d"] fn psrlid128(a: i32x4, imm8: i32) -> i32x4; - #[link_name = "llvm.x86.avx512.pslli.q.512"] - fn vpslliq(a: i64x8, imm8: u32) -> i64x8; - - #[link_name = "llvm.x86.avx2.pslli.q"] - fn pslliq256(a: i64x4, imm8: i32) -> i64x4; - #[link_name = "llvm.x86.sse2.pslli.q"] - fn pslliq128(a: i64x2, imm8: i32) -> i64x2; - #[link_name = "llvm.x86.avx512.psrli.q.512"] fn vpsrliq(a: i64x8, imm8: u32) -> i64x8;