diff options
| -rw-r--r-- | llvm/include/llvm/IR/IntrinsicsX86.td | 260 | ||||
| -rw-r--r-- | llvm/lib/Target/X86/X86IntrinsicsInfo.h | 18 | ||||
| -rw-r--r-- | llvm/test/CodeGen/X86/avx512-intrinsics-fast-isel.ll | 333 | ||||
| -rw-r--r-- | llvm/test/CodeGen/X86/avx512bw-intrinsics-fast-isel.ll | 92 | ||||
| -rw-r--r-- | llvm/test/CodeGen/X86/avx512bwvl-intrinsics-fast-isel.ll | 186 | ||||
| -rw-r--r-- | llvm/test/CodeGen/X86/avx512vbmi-intrinsics-fast-isel.ll | 103 | ||||
| -rw-r--r-- | llvm/test/CodeGen/X86/avx512vbmivl-intrinsics-fast-isel.ll | 188 | ||||
| -rw-r--r-- | llvm/test/CodeGen/X86/avx512vl-intrinsics-fast-isel.ll | 687 |
8 files changed, 1759 insertions, 108 deletions
diff --git a/llvm/include/llvm/IR/IntrinsicsX86.td b/llvm/include/llvm/IR/IntrinsicsX86.td index 21bf565d10f..8e37afe93cb 100644 --- a/llvm/include/llvm/IR/IntrinsicsX86.td +++ b/llvm/include/llvm/IR/IntrinsicsX86.td @@ -1012,326 +1012,370 @@ let TargetPrefix = "x86" in { // All intrinsics start with "llvm.x86.". GCCBuiltin<"__builtin_ia32_vpermilvarps256">, Intrinsic<[llvm_v8f32_ty], [llvm_v8f32_ty, llvm_v8i32_ty], [IntrNoMem]>; - def int_x86_avx512_mask_vpermi2var_d_128 : - GCCBuiltin<"__builtin_ia32_vpermi2vard128_mask">, + def int_x86_avx512_mask_vpermi2var_d_128 : // FIXME: Remove Intrinsic<[llvm_v4i32_ty], [llvm_v4i32_ty, llvm_v4i32_ty, llvm_v4i32_ty, llvm_i8_ty], [IntrNoMem]>; - def int_x86_avx512_mask_vpermi2var_d_256 : - GCCBuiltin<"__builtin_ia32_vpermi2vard256_mask">, + def int_x86_avx512_mask_vpermi2var_d_256 : // FIXME: Remove Intrinsic<[llvm_v8i32_ty], [llvm_v8i32_ty, llvm_v8i32_ty, llvm_v8i32_ty, llvm_i8_ty], [IntrNoMem]>; - def int_x86_avx512_mask_vpermi2var_d_512 : - GCCBuiltin<"__builtin_ia32_vpermi2vard512_mask">, + def int_x86_avx512_mask_vpermi2var_d_512 : // FIXME: Remove Intrinsic<[llvm_v16i32_ty], [llvm_v16i32_ty, llvm_v16i32_ty, llvm_v16i32_ty, llvm_i16_ty], [IntrNoMem]>; - def int_x86_avx512_mask_vpermi2var_hi_128 : - GCCBuiltin<"__builtin_ia32_vpermi2varhi128_mask">, + def int_x86_avx512_mask_vpermi2var_hi_128 : // FIXME: Remove Intrinsic<[llvm_v8i16_ty], [llvm_v8i16_ty, llvm_v8i16_ty, llvm_v8i16_ty, llvm_i8_ty], [IntrNoMem]>; - def int_x86_avx512_mask_vpermi2var_hi_256 : - GCCBuiltin<"__builtin_ia32_vpermi2varhi256_mask">, + def int_x86_avx512_mask_vpermi2var_hi_256 : // FIXME: Remove Intrinsic<[llvm_v16i16_ty], [llvm_v16i16_ty, llvm_v16i16_ty, llvm_v16i16_ty, llvm_i16_ty], [IntrNoMem]>; - def int_x86_avx512_mask_vpermi2var_hi_512 : - GCCBuiltin<"__builtin_ia32_vpermi2varhi512_mask">, + def int_x86_avx512_mask_vpermi2var_hi_512 : // FIXME: Remove Intrinsic<[llvm_v32i16_ty], [llvm_v32i16_ty, llvm_v32i16_ty, llvm_v32i16_ty, llvm_i32_ty], [IntrNoMem]>; - def int_x86_avx512_mask_vpermi2var_pd_128 : - GCCBuiltin<"__builtin_ia32_vpermi2varpd128_mask">, + def int_x86_avx512_mask_vpermi2var_pd_128 : // FIXME: Remove Intrinsic<[llvm_v2f64_ty], [llvm_v2f64_ty, llvm_v2i64_ty, llvm_v2f64_ty, llvm_i8_ty], [IntrNoMem]>; - def int_x86_avx512_mask_vpermi2var_pd_256 : - GCCBuiltin<"__builtin_ia32_vpermi2varpd256_mask">, + def int_x86_avx512_mask_vpermi2var_pd_256 : // FIXME: Remove Intrinsic<[llvm_v4f64_ty], [llvm_v4f64_ty, llvm_v4i64_ty, llvm_v4f64_ty, llvm_i8_ty], [IntrNoMem]>; - def int_x86_avx512_mask_vpermi2var_pd_512 : - GCCBuiltin<"__builtin_ia32_vpermi2varpd512_mask">, + def int_x86_avx512_mask_vpermi2var_pd_512 : // FIXME: Remove Intrinsic<[llvm_v8f64_ty], [llvm_v8f64_ty, llvm_v8i64_ty, llvm_v8f64_ty, llvm_i8_ty], [IntrNoMem]>; - def int_x86_avx512_mask_vpermi2var_ps_128 : - GCCBuiltin<"__builtin_ia32_vpermi2varps128_mask">, + def int_x86_avx512_mask_vpermi2var_ps_128 : // FIXME: Remove Intrinsic<[llvm_v4f32_ty], [llvm_v4f32_ty, llvm_v4i32_ty, llvm_v4f32_ty, llvm_i8_ty], [IntrNoMem]>; - def int_x86_avx512_mask_vpermi2var_ps_256 : - GCCBuiltin<"__builtin_ia32_vpermi2varps256_mask">, + def int_x86_avx512_mask_vpermi2var_ps_256 : // FIXME: Remove Intrinsic<[llvm_v8f32_ty], [llvm_v8f32_ty, llvm_v8i32_ty, llvm_v8f32_ty, llvm_i8_ty], [IntrNoMem]>; - def int_x86_avx512_mask_vpermi2var_ps_512 : - GCCBuiltin<"__builtin_ia32_vpermi2varps512_mask">, + def int_x86_avx512_mask_vpermi2var_ps_512 : // FIXME: Remove Intrinsic<[llvm_v16f32_ty], [llvm_v16f32_ty, llvm_v16i32_ty, llvm_v16f32_ty, llvm_i16_ty], [IntrNoMem]>; - def int_x86_avx512_mask_vpermi2var_q_128 : - GCCBuiltin<"__builtin_ia32_vpermi2varq128_mask">, + def int_x86_avx512_mask_vpermi2var_q_128 : // FIXME: Remove Intrinsic<[llvm_v2i64_ty], [llvm_v2i64_ty, llvm_v2i64_ty, llvm_v2i64_ty, llvm_i8_ty], [IntrNoMem]>; - def int_x86_avx512_mask_vpermi2var_q_256 : - GCCBuiltin<"__builtin_ia32_vpermi2varq256_mask">, + def int_x86_avx512_mask_vpermi2var_q_256 : // FIXME: Remove Intrinsic<[llvm_v4i64_ty], [llvm_v4i64_ty, llvm_v4i64_ty, llvm_v4i64_ty, llvm_i8_ty], [IntrNoMem]>; - def int_x86_avx512_mask_vpermi2var_q_512 : - GCCBuiltin<"__builtin_ia32_vpermi2varq512_mask">, + def int_x86_avx512_mask_vpermi2var_q_512 : // FIXME: Remove Intrinsic<[llvm_v8i64_ty], [llvm_v8i64_ty, llvm_v8i64_ty, llvm_v8i64_ty, llvm_i8_ty], [IntrNoMem]>; - def int_x86_avx512_mask_vpermt2var_d_512: - GCCBuiltin<"__builtin_ia32_vpermt2vard512_mask">, + def int_x86_avx512_mask_vpermt2var_d_512: // FIXME: Remove Intrinsic<[llvm_v16i32_ty], [llvm_v16i32_ty, llvm_v16i32_ty, llvm_v16i32_ty, llvm_i16_ty], [IntrNoMem]>; - def int_x86_avx512_mask_vpermt2var_q_512: - GCCBuiltin<"__builtin_ia32_vpermt2varq512_mask">, + def int_x86_avx512_mask_vpermt2var_q_512: // FIXME: Remove Intrinsic<[llvm_v8i64_ty], [llvm_v8i64_ty, llvm_v8i64_ty, llvm_v8i64_ty, llvm_i8_ty], [IntrNoMem]>; - def int_x86_avx512_mask_vpermt2var_ps_512: - GCCBuiltin<"__builtin_ia32_vpermt2varps512_mask">, + def int_x86_avx512_mask_vpermt2var_ps_512: // FIXME: Remove Intrinsic<[llvm_v16f32_ty], [llvm_v16i32_ty, llvm_v16f32_ty, llvm_v16f32_ty, llvm_i16_ty], [IntrNoMem]>; - def int_x86_avx512_mask_vpermt2var_pd_512: - GCCBuiltin<"__builtin_ia32_vpermt2varpd512_mask">, + def int_x86_avx512_mask_vpermt2var_pd_512: // FIXME: Remove Intrinsic<[llvm_v8f64_ty], [llvm_v8i64_ty, llvm_v8f64_ty, llvm_v8f64_ty, llvm_i8_ty], [IntrNoMem]>; - def int_x86_avx512_mask_vpermt2var_d_128 : - GCCBuiltin<"__builtin_ia32_vpermt2vard128_mask">, + def int_x86_avx512_mask_vpermt2var_d_128 : // FIXME: Remove Intrinsic<[llvm_v4i32_ty], [llvm_v4i32_ty, llvm_v4i32_ty, llvm_v4i32_ty, llvm_i8_ty], [IntrNoMem]>; - def int_x86_avx512_maskz_vpermt2var_d_128 : - GCCBuiltin<"__builtin_ia32_vpermt2vard128_maskz">, + def int_x86_avx512_maskz_vpermt2var_d_128 : // FIXME: Remove Intrinsic<[llvm_v4i32_ty], [llvm_v4i32_ty, llvm_v4i32_ty, llvm_v4i32_ty, llvm_i8_ty], [IntrNoMem]>; - def int_x86_avx512_mask_vpermt2var_d_256 : - GCCBuiltin<"__builtin_ia32_vpermt2vard256_mask">, + def int_x86_avx512_mask_vpermt2var_d_256 : // FIXME: Remove Intrinsic<[llvm_v8i32_ty], [llvm_v8i32_ty, llvm_v8i32_ty, llvm_v8i32_ty, llvm_i8_ty], [IntrNoMem]>; - def int_x86_avx512_maskz_vpermt2var_d_256 : - GCCBuiltin<"__builtin_ia32_vpermt2vard256_maskz">, + def int_x86_avx512_maskz_vpermt2var_d_256 : // FIXME: Remove Intrinsic<[llvm_v8i32_ty], [llvm_v8i32_ty, llvm_v8i32_ty, llvm_v8i32_ty, llvm_i8_ty], [IntrNoMem]>; - def int_x86_avx512_maskz_vpermt2var_d_512 : - GCCBuiltin<"__builtin_ia32_vpermt2vard512_maskz">, + def int_x86_avx512_maskz_vpermt2var_d_512 : // FIXME: Remove Intrinsic<[llvm_v16i32_ty], [llvm_v16i32_ty, llvm_v16i32_ty, llvm_v16i32_ty, llvm_i16_ty], [IntrNoMem]>; - def int_x86_avx512_mask_vpermt2var_hi_128 : - GCCBuiltin<"__builtin_ia32_vpermt2varhi128_mask">, + def int_x86_avx512_mask_vpermt2var_hi_128 : // FIXME: Remove Intrinsic<[llvm_v8i16_ty], [llvm_v8i16_ty, llvm_v8i16_ty, llvm_v8i16_ty, llvm_i8_ty], [IntrNoMem]>; - def int_x86_avx512_maskz_vpermt2var_hi_128 : - GCCBuiltin<"__builtin_ia32_vpermt2varhi128_maskz">, + def int_x86_avx512_maskz_vpermt2var_hi_128 : // FIXME: Remove Intrinsic<[llvm_v8i16_ty], [llvm_v8i16_ty, llvm_v8i16_ty, llvm_v8i16_ty, llvm_i8_ty], [IntrNoMem]>; - def int_x86_avx512_mask_vpermt2var_hi_256 : - GCCBuiltin<"__builtin_ia32_vpermt2varhi256_mask">, + def int_x86_avx512_mask_vpermt2var_hi_256 : // FIXME: Remove Intrinsic<[llvm_v16i16_ty], [llvm_v16i16_ty, llvm_v16i16_ty, llvm_v16i16_ty, llvm_i16_ty], [IntrNoMem]>; - def int_x86_avx512_maskz_vpermt2var_hi_256 : - GCCBuiltin<"__builtin_ia32_vpermt2varhi256_maskz">, + def int_x86_avx512_maskz_vpermt2var_hi_256 : // FIXME: Remove Intrinsic<[llvm_v16i16_ty], [llvm_v16i16_ty, llvm_v16i16_ty, llvm_v16i16_ty, llvm_i16_ty], [IntrNoMem]>; - def int_x86_avx512_mask_vpermt2var_hi_512 : - GCCBuiltin<"__builtin_ia32_vpermt2varhi512_mask">, + def int_x86_avx512_mask_vpermt2var_hi_512 : // FIXME: Remove Intrinsic<[llvm_v32i16_ty], [llvm_v32i16_ty, llvm_v32i16_ty, llvm_v32i16_ty, llvm_i32_ty], [IntrNoMem]>; - def int_x86_avx512_maskz_vpermt2var_hi_512 : - GCCBuiltin<"__builtin_ia32_vpermt2varhi512_maskz">, + def int_x86_avx512_maskz_vpermt2var_hi_512 : // FIXME: Remove Intrinsic<[llvm_v32i16_ty], [llvm_v32i16_ty, llvm_v32i16_ty, llvm_v32i16_ty, llvm_i32_ty], [IntrNoMem]>; - def int_x86_avx512_mask_vpermt2var_pd_128 : - GCCBuiltin<"__builtin_ia32_vpermt2varpd128_mask">, + def int_x86_avx512_mask_vpermt2var_pd_128 : // FIXME: Remove Intrinsic<[llvm_v2f64_ty], [llvm_v2i64_ty, llvm_v2f64_ty, llvm_v2f64_ty, llvm_i8_ty], [IntrNoMem]>; - def int_x86_avx512_maskz_vpermt2var_pd_128 : - GCCBuiltin<"__builtin_ia32_vpermt2varpd128_maskz">, + def int_x86_avx512_maskz_vpermt2var_pd_128 : // FIXME: Remove Intrinsic<[llvm_v2f64_ty], [llvm_v2i64_ty, llvm_v2f64_ty, llvm_v2f64_ty, llvm_i8_ty], [IntrNoMem]>; - def int_x86_avx512_mask_vpermt2var_pd_256 : - GCCBuiltin<"__builtin_ia32_vpermt2varpd256_mask">, + def int_x86_avx512_mask_vpermt2var_pd_256 : // FIXME: Remove Intrinsic<[llvm_v4f64_ty], [llvm_v4i64_ty, llvm_v4f64_ty, llvm_v4f64_ty, llvm_i8_ty], [IntrNoMem]>; - def int_x86_avx512_maskz_vpermt2var_pd_256 : - GCCBuiltin<"__builtin_ia32_vpermt2varpd256_maskz">, + def int_x86_avx512_maskz_vpermt2var_pd_256 : // FIXME: Remove Intrinsic<[llvm_v4f64_ty], [llvm_v4i64_ty, llvm_v4f64_ty, llvm_v4f64_ty, llvm_i8_ty], [IntrNoMem]>; - def int_x86_avx512_maskz_vpermt2var_pd_512 : - GCCBuiltin<"__builtin_ia32_vpermt2varpd512_maskz">, + def int_x86_avx512_maskz_vpermt2var_pd_512 : // FIXME: Remove Intrinsic<[llvm_v8f64_ty], [llvm_v8i64_ty, llvm_v8f64_ty, llvm_v8f64_ty, llvm_i8_ty], [IntrNoMem]>; - def int_x86_avx512_mask_vpermt2var_ps_128 : - GCCBuiltin<"__builtin_ia32_vpermt2varps128_mask">, + def int_x86_avx512_mask_vpermt2var_ps_128 : // FIXME: Remove Intrinsic<[llvm_v4f32_ty], [llvm_v4i32_ty, llvm_v4f32_ty, llvm_v4f32_ty, llvm_i8_ty], [IntrNoMem]>; - def int_x86_avx512_maskz_vpermt2var_ps_128 : - GCCBuiltin<"__builtin_ia32_vpermt2varps128_maskz">, + def int_x86_avx512_maskz_vpermt2var_ps_128 : // FIXME: Remove Intrinsic<[llvm_v4f32_ty], [llvm_v4i32_ty, llvm_v4f32_ty, llvm_v4f32_ty, llvm_i8_ty], [IntrNoMem]>; - def int_x86_avx512_mask_vpermt2var_ps_256 : - GCCBuiltin<"__builtin_ia32_vpermt2varps256_mask">, + def int_x86_avx512_mask_vpermt2var_ps_256 : // FIXME: Remove Intrinsic<[llvm_v8f32_ty], [llvm_v8i32_ty, llvm_v8f32_ty, llvm_v8f32_ty, llvm_i8_ty], [IntrNoMem]>; - def int_x86_avx512_maskz_vpermt2var_ps_256 : - GCCBuiltin<"__builtin_ia32_vpermt2varps256_maskz">, + def int_x86_avx512_maskz_vpermt2var_ps_256 : // FIXME: Remove Intrinsic<[llvm_v8f32_ty], [llvm_v8i32_ty, llvm_v8f32_ty, llvm_v8f32_ty, llvm_i8_ty], [IntrNoMem]>; - def int_x86_avx512_maskz_vpermt2var_ps_512 : - GCCBuiltin<"__builtin_ia32_vpermt2varps512_maskz">, + def int_x86_avx512_maskz_vpermt2var_ps_512 : // FIXME: Remove Intrinsic<[llvm_v16f32_ty], [llvm_v16i32_ty, llvm_v16f32_ty, llvm_v16f32_ty, llvm_i16_ty], [IntrNoMem]>; - def int_x86_avx512_mask_vpermt2var_q_128 : - GCCBuiltin<"__builtin_ia32_vpermt2varq128_mask">, + def int_x86_avx512_mask_vpermt2var_q_128 : // FIXME: Remove Intrinsic<[llvm_v2i64_ty], [llvm_v2i64_ty, llvm_v2i64_ty, llvm_v2i64_ty, llvm_i8_ty], [IntrNoMem]>; - def int_x86_avx512_maskz_vpermt2var_q_128 : - GCCBuiltin<"__builtin_ia32_vpermt2varq128_maskz">, + def int_x86_avx512_maskz_vpermt2var_q_128 : // FIXME: Remove Intrinsic<[llvm_v2i64_ty], [llvm_v2i64_ty, llvm_v2i64_ty, llvm_v2i64_ty, llvm_i8_ty], [IntrNoMem]>; - def int_x86_avx512_mask_vpermt2var_q_256 : - GCCBuiltin<"__builtin_ia32_vpermt2varq256_mask">, + def int_x86_avx512_mask_vpermt2var_q_256 : // FIXME: Remove Intrinsic<[llvm_v4i64_ty], [llvm_v4i64_ty, llvm_v4i64_ty, llvm_v4i64_ty, llvm_i8_ty], [IntrNoMem]>; - def int_x86_avx512_maskz_vpermt2var_q_256 : - GCCBuiltin<"__builtin_ia32_vpermt2varq256_maskz">, + def int_x86_avx512_maskz_vpermt2var_q_256 : // FIXME: Remove Intrinsic<[llvm_v4i64_ty], [llvm_v4i64_ty, llvm_v4i64_ty, llvm_v4i64_ty, llvm_i8_ty], [IntrNoMem]>; - def int_x86_avx512_maskz_vpermt2var_q_512 : - GCCBuiltin<"__builtin_ia32_vpermt2varq512_maskz">, + def int_x86_avx512_maskz_vpermt2var_q_512 : // FIXME: Remove Intrinsic<[llvm_v8i64_ty], [llvm_v8i64_ty, llvm_v8i64_ty, llvm_v8i64_ty, llvm_i8_ty], [IntrNoMem]>; - def int_x86_avx512_mask_vpermi2var_qi_128 : - GCCBuiltin<"__builtin_ia32_vpermi2varqi128_mask">, + def int_x86_avx512_mask_vpermi2var_qi_128 : // FIXME: Remove Intrinsic<[llvm_v16i8_ty], [llvm_v16i8_ty, llvm_v16i8_ty, llvm_v16i8_ty, llvm_i16_ty], [IntrNoMem]>; - def int_x86_avx512_mask_vpermt2var_qi_128 : - GCCBuiltin<"__builtin_ia32_vpermt2varqi128_mask">, + def int_x86_avx512_mask_vpermt2var_qi_128 : // FIXME: Remove Intrinsic<[llvm_v16i8_ty], [llvm_v16i8_ty, llvm_v16i8_ty, llvm_v16i8_ty, llvm_i16_ty], [IntrNoMem]>; - def int_x86_avx512_maskz_vpermt2var_qi_128 : - GCCBuiltin<"__builtin_ia32_vpermt2varqi128_maskz">, + def int_x86_avx512_maskz_vpermt2var_qi_128 : // FIXME: Remove Intrinsic<[llvm_v16i8_ty], [llvm_v16i8_ty, llvm_v16i8_ty, llvm_v16i8_ty, llvm_i16_ty], [IntrNoMem]>; - def int_x86_avx512_mask_vpermi2var_qi_256 : - GCCBuiltin<"__builtin_ia32_vpermi2varqi256_mask">, + def int_x86_avx512_mask_vpermi2var_qi_256 : // FIXME: Remove Intrinsic<[llvm_v32i8_ty], [llvm_v32i8_ty, llvm_v32i8_ty, llvm_v32i8_ty, llvm_i32_ty], [IntrNoMem]>; - def int_x86_avx512_mask_vpermt2var_qi_256 : - GCCBuiltin<"__builtin_ia32_vpermt2varqi256_mask">, + def int_x86_avx512_mask_vpermt2var_qi_256 : // FIXME: Remove Intrinsic<[llvm_v32i8_ty], [llvm_v32i8_ty, llvm_v32i8_ty, llvm_v32i8_ty, llvm_i32_ty], [IntrNoMem]>; - def int_x86_avx512_maskz_vpermt2var_qi_256 : - GCCBuiltin<"__builtin_ia32_vpermt2varqi256_maskz">, + def int_x86_avx512_maskz_vpermt2var_qi_256 : // FIXME: Remove Intrinsic<[llvm_v32i8_ty], [llvm_v32i8_ty, llvm_v32i8_ty, llvm_v32i8_ty, llvm_i32_ty], [IntrNoMem]>; - def int_x86_avx512_mask_vpermi2var_qi_512 : - GCCBuiltin<"__builtin_ia32_vpermi2varqi512_mask">, + def int_x86_avx512_mask_vpermi2var_qi_512 : // FIXME: Remove Intrinsic<[llvm_v64i8_ty], [llvm_v64i8_ty, llvm_v64i8_ty, llvm_v64i8_ty, llvm_i64_ty], [IntrNoMem]>; - def int_x86_avx512_mask_vpermt2var_qi_512 : - GCCBuiltin<"__builtin_ia32_vpermt2varqi512_mask">, + def int_x86_avx512_mask_vpermt2var_qi_512 : // FIXME: Remove Intrinsic<[llvm_v64i8_ty], [llvm_v64i8_ty, llvm_v64i8_ty, llvm_v64i8_ty, llvm_i64_ty], [IntrNoMem]>; - def int_x86_avx512_maskz_vpermt2var_qi_512 : - GCCBuiltin<"__builtin_ia32_vpermt2varqi512_maskz">, + def int_x86_avx512_maskz_vpermt2var_qi_512 : // FIXME: Remove Intrinsic<[llvm_v64i8_ty], [llvm_v64i8_ty, llvm_v64i8_ty, llvm_v64i8_ty, llvm_i64_ty], [IntrNoMem]>; + + + + + def int_x86_avx512_vpermi2var_d_128 : + GCCBuiltin<"__builtin_ia32_vpermi2vard128">, + Intrinsic<[llvm_v4i32_ty], + [llvm_v4i32_ty, llvm_v4i32_ty, llvm_v4i32_ty], [IntrNoMem]>; + + def int_x86_avx512_vpermi2var_d_256 : + GCCBuiltin<"__builtin_ia32_vpermi2vard256">, + Intrinsic<[llvm_v8i32_ty], + [llvm_v8i32_ty, llvm_v8i32_ty, llvm_v8i32_ty], [IntrNoMem]>; + + def int_x86_avx512_vpermi2var_d_512 : + GCCBuiltin<"__builtin_ia32_vpermi2vard512">, + Intrinsic<[llvm_v16i32_ty], + [llvm_v16i32_ty, llvm_v16i32_ty, llvm_v16i32_ty], + [IntrNoMem]>; + + def int_x86_avx512_vpermi2var_hi_128 : + GCCBuiltin<"__builtin_ia32_vpermi2varhi128">, + Intrinsic<[llvm_v8i16_ty], + [llvm_v8i16_ty, llvm_v8i16_ty, llvm_v8i16_ty], [IntrNoMem]>; + + def int_x86_avx512_vpermi2var_hi_256 : + GCCBuiltin<"__builtin_ia32_vpermi2varhi256">, + Intrinsic<[llvm_v16i16_ty], + [llvm_v16i16_ty, llvm_v16i16_ty, llvm_v16i16_ty], + [IntrNoMem]>; + + def int_x86_avx512_vpermi2var_hi_512 : + GCCBuiltin<"__builtin_ia32_vpermi2varhi512">, + Intrinsic<[llvm_v32i16_ty], + [llvm_v32i16_ty, llvm_v32i16_ty, llvm_v32i16_ty], + [IntrNoMem]>; + + def int_x86_avx512_vpermi2var_pd_128 : + GCCBuiltin<"__builtin_ia32_vpermi2varpd128">, + Intrinsic<[llvm_v2f64_ty], + [llvm_v2f64_ty, llvm_v2i64_ty, llvm_v2f64_ty], [IntrNoMem]>; + + def int_x86_avx512_vpermi2var_pd_256 : + GCCBuiltin<"__builtin_ia32_vpermi2varpd256">, + Intrinsic<[llvm_v4f64_ty], + [llvm_v4f64_ty, llvm_v4i64_ty, llvm_v4f64_ty], [IntrNoMem]>; + + def int_x86_avx512_vpermi2var_pd_512 : + GCCBuiltin<"__builtin_ia32_vpermi2varpd512">, + Intrinsic<[llvm_v8f64_ty], + [llvm_v8f64_ty, llvm_v8i64_ty, llvm_v8f64_ty], [IntrNoMem]>; + + def int_x86_avx512_vpermi2var_ps_128 : + GCCBuiltin<"__builtin_ia32_vpermi2varps128">, + Intrinsic<[llvm_v4f32_ty], + [llvm_v4f32_ty, llvm_v4i32_ty, llvm_v4f32_ty], [IntrNoMem]>; + + def int_x86_avx512_vpermi2var_ps_256 : + GCCBuiltin<"__builtin_ia32_vpermi2varps256">, + Intrinsic<[llvm_v8f32_ty], + [llvm_v8f32_ty, llvm_v8i32_ty, llvm_v8f32_ty], [IntrNoMem]>; + + def int_x86_avx512_vpermi2var_ps_512 : + GCCBuiltin<"__builtin_ia32_vpermi2varps512">, + Intrinsic<[llvm_v16f32_ty], + [llvm_v16f32_ty, llvm_v16i32_ty, llvm_v16f32_ty], + [IntrNoMem]>; + + def int_x86_avx512_vpermi2var_q_128 : + GCCBuiltin<"__builtin_ia32_vpermi2varq128">, + Intrinsic<[llvm_v2i64_ty], + [llvm_v2i64_ty, llvm_v2i64_ty, llvm_v2i64_ty], [IntrNoMem]>; + + def int_x86_avx512_vpermi2var_q_256 : + GCCBuiltin<"__builtin_ia32_vpermi2varq256">, + Intrinsic<[llvm_v4i64_ty], + [llvm_v4i64_ty, llvm_v4i64_ty, llvm_v4i64_ty], [IntrNoMem]>; + + def int_x86_avx512_vpermi2var_q_512 : + GCCBuiltin<"__builtin_ia32_vpermi2varq512">, + Intrinsic<[llvm_v8i64_ty], + [llvm_v8i64_ty, llvm_v8i64_ty, llvm_v8i64_ty], [IntrNoMem]>; + + def int_x86_avx512_vpermi2var_qi_128 : + GCCBuiltin<"__builtin_ia32_vpermi2varqi128">, + Intrinsic<[llvm_v16i8_ty], + [llvm_v16i8_ty, llvm_v16i8_ty, llvm_v16i8_ty], [IntrNoMem]>; + + def int_x86_avx512_vpermi2var_qi_256 : + GCCBuiltin<"__builtin_ia32_vpermi2varqi256">, + Intrinsic<[llvm_v32i8_ty], + [llvm_v32i8_ty, llvm_v32i8_ty, llvm_v32i8_ty], [IntrNoMem]>; + + def int_x86_avx512_vpermi2var_qi_512 : + GCCBuiltin<"__builtin_ia32_vpermi2varqi512">, + Intrinsic<[llvm_v64i8_ty], + [llvm_v64i8_ty, llvm_v64i8_ty, llvm_v64i8_ty], [IntrNoMem]>; + def int_x86_avx512_vpermilvar_pd_512 : GCCBuiltin<"__builtin_ia32_vpermilvarpd512">, Intrinsic<[llvm_v8f64_ty], [llvm_v8f64_ty, llvm_v8i64_ty], diff --git a/llvm/lib/Target/X86/X86IntrinsicsInfo.h b/llvm/lib/Target/X86/X86IntrinsicsInfo.h index 3e3a62ec82e..3a79af9a1f5 100644 --- a/llvm/lib/Target/X86/X86IntrinsicsInfo.h +++ b/llvm/lib/Target/X86/X86IntrinsicsInfo.h @@ -1435,6 +1435,24 @@ static const IntrinsicData IntrinsicsWithoutChain[] = { X86_INTRINSIC_DATA(avx512_vcvtss2si64, INTR_TYPE_2OP, X86ISD::CVTS2SI_RND, 0), X86_INTRINSIC_DATA(avx512_vcvtss2usi32, INTR_TYPE_2OP, X86ISD::CVTS2UI_RND, 0), X86_INTRINSIC_DATA(avx512_vcvtss2usi64, INTR_TYPE_2OP, X86ISD::CVTS2UI_RND, 0), + X86_INTRINSIC_DATA(avx512_vpermi2var_d_128, INTR_TYPE_3OP, X86ISD::VPERMV3, 0), + X86_INTRINSIC_DATA(avx512_vpermi2var_d_256, INTR_TYPE_3OP, X86ISD::VPERMV3, 0), + X86_INTRINSIC_DATA(avx512_vpermi2var_d_512, INTR_TYPE_3OP, X86ISD::VPERMV3, 0), + X86_INTRINSIC_DATA(avx512_vpermi2var_hi_128, INTR_TYPE_3OP, X86ISD::VPERMV3, 0), + X86_INTRINSIC_DATA(avx512_vpermi2var_hi_256, INTR_TYPE_3OP, X86ISD::VPERMV3, 0), + X86_INTRINSIC_DATA(avx512_vpermi2var_hi_512, INTR_TYPE_3OP, X86ISD::VPERMV3, 0), + X86_INTRINSIC_DATA(avx512_vpermi2var_pd_128, INTR_TYPE_3OP, X86ISD::VPERMV3, 0), + X86_INTRINSIC_DATA(avx512_vpermi2var_pd_256, INTR_TYPE_3OP, X86ISD::VPERMV3, 0), + X86_INTRINSIC_DATA(avx512_vpermi2var_pd_512, INTR_TYPE_3OP, X86ISD::VPERMV3, 0), + X86_INTRINSIC_DATA(avx512_vpermi2var_ps_128, INTR_TYPE_3OP, X86ISD::VPERMV3, 0), + X86_INTRINSIC_DATA(avx512_vpermi2var_ps_256, INTR_TYPE_3OP, X86ISD::VPERMV3, 0), + X86_INTRINSIC_DATA(avx512_vpermi2var_ps_512, INTR_TYPE_3OP, X86ISD::VPERMV3, 0), + X86_INTRINSIC_DATA(avx512_vpermi2var_q_128, INTR_TYPE_3OP, X86ISD::VPERMV3, 0), + X86_INTRINSIC_DATA(avx512_vpermi2var_q_256, INTR_TYPE_3OP, X86ISD::VPERMV3, 0), + X86_INTRINSIC_DATA(avx512_vpermi2var_q_512, INTR_TYPE_3OP, X86ISD::VPERMV3, 0), + X86_INTRINSIC_DATA(avx512_vpermi2var_qi_128, INTR_TYPE_3OP, X86ISD::VPERMV3, 0), + X86_INTRINSIC_DATA(avx512_vpermi2var_qi_256, INTR_TYPE_3OP, X86ISD::VPERMV3, 0), + X86_INTRINSIC_DATA(avx512_vpermi2var_qi_512, INTR_TYPE_3OP, X86ISD::VPERMV3, 0), X86_INTRINSIC_DATA(avx512_vpermilvar_pd_512, INTR_TYPE_2OP, X86ISD::VPERMILPV, 0), X86_INTRINSIC_DATA(avx512_vpermilvar_ps_512, INTR_TYPE_2OP, X86ISD::VPERMILPV, 0), X86_INTRINSIC_DATA(avx512_vpmadd52h_uq_128 , IFMA_OP, X86ISD::VPMADD52H, 0), diff --git a/llvm/test/CodeGen/X86/avx512-intrinsics-fast-isel.ll b/llvm/test/CodeGen/X86/avx512-intrinsics-fast-isel.ll index 6b93271c119..00e54d03dc1 100644 --- a/llvm/test/CodeGen/X86/avx512-intrinsics-fast-isel.ll +++ b/llvm/test/CodeGen/X86/avx512-intrinsics-fast-isel.ll @@ -2478,5 +2478,338 @@ entry: ret <8 x i64> %2 } +declare <16 x i32> @llvm.x86.avx512.vpermi2var.d.512(<16 x i32>, <16 x i32>, <16 x i32>) + +define <8 x i64> @test_mm512_mask2_permutex2var_epi32(<8 x i64> %__A, <8 x i64> %__I, i16 zeroext %__U, <8 x i64> %__B) { +; X32-LABEL: test_mm512_mask2_permutex2var_epi32: +; X32: # %bb.0: # %entry +; X32-NEXT: kmovw {{[0-9]+}}(%esp), %k1 +; X32-NEXT: vpermi2d %zmm2, %zmm0, %zmm1 {%k1} +; X32-NEXT: vmovdqa64 %zmm1, %zmm0 +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_mask2_permutex2var_epi32: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vpermi2d %zmm2, %zmm0, %zmm1 {%k1} +; X64-NEXT: vmovdqa64 %zmm1, %zmm0 +; X64-NEXT: retq +entry: + %0 = bitcast <8 x i64> %__A to <16 x i32> + %1 = bitcast <8 x i64> %__I to <16 x i32> + %2 = bitcast <8 x i64> %__B to <16 x i32> + %3 = tail call <16 x i32> @llvm.x86.avx512.vpermi2var.d.512(<16 x i32> %0, <16 x i32> %1, <16 x i32> %2) + %4 = bitcast i16 %__U to <16 x i1> + %5 = select <16 x i1> %4, <16 x i32> %3, <16 x i32> %1 + %6 = bitcast <16 x i32> %5 to <8 x i64> + ret <8 x i64> %6 +} + +declare <8 x double> @llvm.x86.avx512.vpermi2var.pd.512(<8 x double>, <8 x i64>, <8 x double>) + +define <8 x double> @test_mm512_mask2_permutex2var_pd(<8 x double> %__A, <8 x i64> %__I, i8 zeroext %__U, <8 x double> %__B) { +; X32-LABEL: test_mm512_mask2_permutex2var_pd: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vpermi2pd %zmm2, %zmm0, %zmm1 {%k1} +; X32-NEXT: vmovapd %zmm1, %zmm0 +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_mask2_permutex2var_pd: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vpermi2pd %zmm2, %zmm0, %zmm1 {%k1} +; X64-NEXT: vmovapd %zmm1, %zmm0 +; X64-NEXT: retq +entry: + %0 = tail call <8 x double> @llvm.x86.avx512.vpermi2var.pd.512(<8 x double> %__A, <8 x i64> %__I, <8 x double> %__B) + %1 = bitcast <8 x i64> %__I to <8 x double> + %2 = bitcast i8 %__U to <8 x i1> + %3 = select <8 x i1> %2, <8 x double> %0, <8 x double> %1 + ret <8 x double> %3 +} + +declare <16 x float> @llvm.x86.avx512.vpermi2var.ps.512(<16 x float>, <16 x i32>, <16 x float>) + +define <16 x float> @test_mm512_mask2_permutex2var_ps(<16 x float> %__A, <8 x i64> %__I, i16 zeroext %__U, <16 x float> %__B) { +; X32-LABEL: test_mm512_mask2_permutex2var_ps: +; X32: # %bb.0: # %entry +; X32-NEXT: kmovw {{[0-9]+}}(%esp), %k1 +; X32-NEXT: vpermi2ps %zmm2, %zmm0, %zmm1 {%k1} +; X32-NEXT: vmovaps %zmm1, %zmm0 +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_mask2_permutex2var_ps: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vpermi2ps %zmm2, %zmm0, %zmm1 {%k1} +; X64-NEXT: vmovaps %zmm1, %zmm0 +; X64-NEXT: retq +entry: + %0 = bitcast <8 x i64> %__I to <16 x i32> + %1 = tail call <16 x float> @llvm.x86.avx512.vpermi2var.ps.512(<16 x float> %__A, <16 x i32> %0, <16 x float> %__B) + %2 = bitcast <8 x i64> %__I to <16 x float> + %3 = bitcast i16 %__U to <16 x i1> + %4 = select <16 x i1> %3, <16 x float> %1, <16 x float> %2 + ret <16 x float> %4 +} + +declare <8 x i64> @llvm.x86.avx512.vpermi2var.q.512(<8 x i64>, <8 x i64>, <8 x i64>) + +define <8 x i64> @test_mm512_mask2_permutex2var_epi64(<8 x i64> %__A, <8 x i64> %__I, i8 zeroext %__U, <8 x i64> %__B) { +; X32-LABEL: test_mm512_mask2_permutex2var_epi64: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vpermi2q %zmm2, %zmm0, %zmm1 {%k1} +; X32-NEXT: vmovdqa64 %zmm1, %zmm0 +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_mask2_permutex2var_epi64: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vpermi2q %zmm2, %zmm0, %zmm1 {%k1} +; X64-NEXT: vmovdqa64 %zmm1, %zmm0 +; X64-NEXT: retq +entry: + %0 = tail call <8 x i64> @llvm.x86.avx512.vpermi2var.q.512(<8 x i64> %__A, <8 x i64> %__I, <8 x i64> %__B) + %1 = bitcast i8 %__U to <8 x i1> + %2 = select <8 x i1> %1, <8 x i64> %0, <8 x i64> %__I + ret <8 x i64> %2 +} + +define <8 x i64> @test_mm512_permutex2var_epi32(<8 x i64> %__A, <8 x i64> %__I, <8 x i64> %__B) { +; X32-LABEL: test_mm512_permutex2var_epi32: +; X32: # %bb.0: # %entry +; X32-NEXT: vpermt2d %zmm2, %zmm1, %zmm0 +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_permutex2var_epi32: +; X64: # %bb.0: # %entry +; X64-NEXT: vpermt2d %zmm2, %zmm1, %zmm0 +; X64-NEXT: retq +entry: + %0 = bitcast <8 x i64> %__A to <16 x i32> + %1 = bitcast <8 x i64> %__I to <16 x i32> + %2 = bitcast <8 x i64> %__B to <16 x i32> + %3 = tail call <16 x i32> @llvm.x86.avx512.vpermi2var.d.512(<16 x i32> %0, <16 x i32> %1, <16 x i32> %2) + %4 = bitcast <16 x i32> %3 to <8 x i64> + ret <8 x i64> %4 +} + +define <8 x i64> @test_mm512_maskz_permutex2var_epi32(i16 zeroext %__U, <8 x i64> %__A, <8 x i64> %__I, <8 x i64> %__B) { +; X32-LABEL: test_mm512_maskz_permutex2var_epi32: +; X32: # %bb.0: # %entry +; X32-NEXT: kmovw {{[0-9]+}}(%esp), %k1 +; X32-NEXT: vpermt2d %zmm2, %zmm1, %zmm0 {%k1} {z} +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_maskz_permutex2var_epi32: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vpermt2d %zmm2, %zmm1, %zmm0 {%k1} {z} +; X64-NEXT: retq +entry: + %0 = bitcast <8 x i64> %__A to <16 x i32> + %1 = bitcast <8 x i64> %__I to <16 x i32> + %2 = bitcast <8 x i64> %__B to <16 x i32> + %3 = tail call <16 x i32> @llvm.x86.avx512.vpermi2var.d.512(<16 x i32> %0, <16 x i32> %1, <16 x i32> %2) + %4 = bitcast i16 %__U to <16 x i1> + %5 = select <16 x i1> %4, <16 x i32> %3, <16 x i32> zeroinitializer + %6 = bitcast <16 x i32> %5 to <8 x i64> + ret <8 x i64> %6 +} + +define <8 x i64> @test_mm512_mask_permutex2var_epi32(<8 x i64> %__A, i16 zeroext %__U, <8 x i64> %__I, <8 x i64> %__B) { +; X32-LABEL: test_mm512_mask_permutex2var_epi32: +; X32: # %bb.0: # %entry +; X32-NEXT: kmovw {{[0-9]+}}(%esp), %k1 +; X32-NEXT: vpermt2d %zmm2, %zmm1, %zmm0 {%k1} +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_mask_permutex2var_epi32: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vpermt2d %zmm2, %zmm1, %zmm0 {%k1} +; X64-NEXT: retq +entry: + %0 = bitcast <8 x i64> %__A to <16 x i32> + %1 = bitcast <8 x i64> %__I to <16 x i32> + %2 = bitcast <8 x i64> %__B to <16 x i32> + %3 = tail call <16 x i32> @llvm.x86.avx512.vpermi2var.d.512(<16 x i32> %0, <16 x i32> %1, <16 x i32> %2) + %4 = bitcast i16 %__U to <16 x i1> + %5 = select <16 x i1> %4, <16 x i32> %3, <16 x i32> %0 + %6 = bitcast <16 x i32> %5 to <8 x i64> + ret <8 x i64> %6 +} + +define <8 x double> @test_mm512_permutex2var_pd(<8 x double> %__A, <8 x i64> %__I, <8 x double> %__B) { +; X32-LABEL: test_mm512_permutex2var_pd: +; X32: # %bb.0: # %entry +; X32-NEXT: vpermt2pd %zmm2, %zmm1, %zmm0 +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_permutex2var_pd: +; X64: # %bb.0: # %entry +; X64-NEXT: vpermt2pd %zmm2, %zmm1, %zmm0 +; X64-NEXT: retq +entry: + %0 = tail call <8 x double> @llvm.x86.avx512.vpermi2var.pd.512(<8 x double> %__A, <8 x i64> %__I, <8 x double> %__B) + ret <8 x double> %0 +} + +define <8 x double> @test_mm512_mask_permutex2var_pd(<8 x double> %__A, i8 zeroext %__U, <8 x i64> %__I, <8 x double> %__B) { +; X32-LABEL: test_mm512_mask_permutex2var_pd: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vpermt2pd %zmm2, %zmm1, %zmm0 {%k1} +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_mask_permutex2var_pd: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vpermt2pd %zmm2, %zmm1, %zmm0 {%k1} +; X64-NEXT: retq +entry: + %0 = tail call <8 x double> @llvm.x86.avx512.vpermi2var.pd.512(<8 x double> %__A, <8 x i64> %__I, <8 x double> %__B) + %1 = bitcast i8 %__U to <8 x i1> + %2 = select <8 x i1> %1, <8 x double> %0, <8 x double> %__A + ret <8 x double> %2 +} + +define <8 x double> @test_mm512_maskz_permutex2var_pd(i8 zeroext %__U, <8 x double> %__A, <8 x i64> %__I, <8 x double> %__B) { +; X32-LABEL: test_mm512_maskz_permutex2var_pd: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vpermt2pd %zmm2, %zmm1, %zmm0 {%k1} {z} +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_maskz_permutex2var_pd: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vpermt2pd %zmm2, %zmm1, %zmm0 {%k1} {z} +; X64-NEXT: retq +entry: + %0 = tail call <8 x double> @llvm.x86.avx512.vpermi2var.pd.512(<8 x double> %__A, <8 x i64> %__I, <8 x double> %__B) + %1 = bitcast i8 %__U to <8 x i1> + %2 = select <8 x i1> %1, <8 x double> %0, <8 x double> zeroinitializer + ret <8 x double> %2 +} + +define <16 x float> @test_mm512_permutex2var_ps(<16 x float> %__A, <8 x i64> %__I, <16 x float> %__B) { +; X32-LABEL: test_mm512_permutex2var_ps: +; X32: # %bb.0: # %entry +; X32-NEXT: vpermt2ps %zmm2, %zmm1, %zmm0 +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_permutex2var_ps: +; X64: # %bb.0: # %entry +; X64-NEXT: vpermt2ps %zmm2, %zmm1, %zmm0 +; X64-NEXT: retq +entry: + %0 = bitcast <8 x i64> %__I to <16 x i32> + %1 = tail call <16 x float> @llvm.x86.avx512.vpermi2var.ps.512(<16 x float> %__A, <16 x i32> %0, <16 x float> %__B) + ret <16 x float> %1 +} + +define <16 x float> @test_mm512_mask_permutex2var_ps(<16 x float> %__A, i16 zeroext %__U, <8 x i64> %__I, <16 x float> %__B) { +; X32-LABEL: test_mm512_mask_permutex2var_ps: +; X32: # %bb.0: # %entry +; X32-NEXT: kmovw {{[0-9]+}}(%esp), %k1 +; X32-NEXT: vpermt2ps %zmm2, %zmm1, %zmm0 {%k1} +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_mask_permutex2var_ps: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vpermt2ps %zmm2, %zmm1, %zmm0 {%k1} +; X64-NEXT: retq +entry: + %0 = bitcast <8 x i64> %__I to <16 x i32> + %1 = tail call <16 x float> @llvm.x86.avx512.vpermi2var.ps.512(<16 x float> %__A, <16 x i32> %0, <16 x float> %__B) + %2 = bitcast i16 %__U to <16 x i1> + %3 = select <16 x i1> %2, <16 x float> %1, <16 x float> %__A + ret <16 x float> %3 +} + +define <16 x float> @test_mm512_maskz_permutex2var_ps(i16 zeroext %__U, <16 x float> %__A, <8 x i64> %__I, <16 x float> %__B) { +; X32-LABEL: test_mm512_maskz_permutex2var_ps: +; X32: # %bb.0: # %entry +; X32-NEXT: kmovw {{[0-9]+}}(%esp), %k1 +; X32-NEXT: vpermt2ps %zmm2, %zmm1, %zmm0 {%k1} {z} +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_maskz_permutex2var_ps: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vpermt2ps %zmm2, %zmm1, %zmm0 {%k1} {z} +; X64-NEXT: retq +entry: + %0 = bitcast <8 x i64> %__I to <16 x i32> + %1 = tail call <16 x float> @llvm.x86.avx512.vpermi2var.ps.512(<16 x float> %__A, <16 x i32> %0, <16 x float> %__B) + %2 = bitcast i16 %__U to <16 x i1> + %3 = select <16 x i1> %2, <16 x float> %1, <16 x float> zeroinitializer + ret <16 x float> %3 +} + +define <8 x i64> @test_mm512_permutex2var_epi64(<8 x i64> %__A, <8 x i64> %__I, <8 x i64> %__B) { +; X32-LABEL: test_mm512_permutex2var_epi64: +; X32: # %bb.0: # %entry +; X32-NEXT: vpermt2q %zmm2, %zmm1, %zmm0 +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_permutex2var_epi64: +; X64: # %bb.0: # %entry +; X64-NEXT: vpermt2q %zmm2, %zmm1, %zmm0 +; X64-NEXT: retq +entry: + %0 = tail call <8 x i64> @llvm.x86.avx512.vpermi2var.q.512(<8 x i64> %__A, <8 x i64> %__I, <8 x i64> %__B) + ret <8 x i64> %0 +} + +define <8 x i64> @test_mm512_mask_permutex2var_epi64(<8 x i64> %__A, i8 zeroext %__U, <8 x i64> %__I, <8 x i64> %__B) { +; X32-LABEL: test_mm512_mask_permutex2var_epi64: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vpermt2q %zmm2, %zmm1, %zmm0 {%k1} +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_mask_permutex2var_epi64: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vpermt2q %zmm2, %zmm1, %zmm0 {%k1} +; X64-NEXT: retq +entry: + %0 = tail call <8 x i64> @llvm.x86.avx512.vpermi2var.q.512(<8 x i64> %__A, <8 x i64> %__I, <8 x i64> %__B) + %1 = bitcast i8 %__U to <8 x i1> + %2 = select <8 x i1> %1, <8 x i64> %0, <8 x i64> %__A + ret <8 x i64> %2 +} + +define <8 x i64> @test_mm512_maskz_permutex2var_epi64(i8 zeroext %__U, <8 x i64> %__A, <8 x i64> %__I, <8 x i64> %__B) { +; X32-LABEL: test_mm512_maskz_permutex2var_epi64: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vpermt2q %zmm2, %zmm1, %zmm0 {%k1} {z} +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_maskz_permutex2var_epi64: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vpermt2q %zmm2, %zmm1, %zmm0 {%k1} {z} +; X64-NEXT: retq +entry: + %0 = tail call <8 x i64> @llvm.x86.avx512.vpermi2var.q.512(<8 x i64> %__A, <8 x i64> %__I, <8 x i64> %__B) + %1 = bitcast i8 %__U to <8 x i1> + %2 = select <8 x i1> %1, <8 x i64> %0, <8 x i64> zeroinitializer + ret <8 x i64> %2 +} + + !0 = !{i32 1} diff --git a/llvm/test/CodeGen/X86/avx512bw-intrinsics-fast-isel.ll b/llvm/test/CodeGen/X86/avx512bw-intrinsics-fast-isel.ll index 018487afe51..76f183542f4 100644 --- a/llvm/test/CodeGen/X86/avx512bw-intrinsics-fast-isel.ll +++ b/llvm/test/CodeGen/X86/avx512bw-intrinsics-fast-isel.ll @@ -861,5 +861,97 @@ entry: ret <4 x i64> %3 } +define <8 x i64> @test_mm512_mask2_permutex2var_epi16(<8 x i64> %__A, <8 x i64> %__I, i32 %__U, <8 x i64> %__B) { +; X32-LABEL: test_mm512_mask2_permutex2var_epi16: +; X32: # %bb.0: # %entry +; X32-NEXT: kmovd {{[0-9]+}}(%esp), %k1 +; X32-NEXT: vpermi2w %zmm2, %zmm0, %zmm1 {%k1} +; X32-NEXT: vmovdqa64 %zmm1, %zmm0 +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_mask2_permutex2var_epi16: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovd %edi, %k1 +; X64-NEXT: vpermi2w %zmm2, %zmm0, %zmm1 {%k1} +; X64-NEXT: vmovdqa64 %zmm1, %zmm0 +; X64-NEXT: retq +entry: + %0 = bitcast <8 x i64> %__A to <32 x i16> + %1 = bitcast <8 x i64> %__I to <32 x i16> + %2 = bitcast <8 x i64> %__B to <32 x i16> + %3 = tail call <32 x i16> @llvm.x86.avx512.vpermi2var.hi.512(<32 x i16> %0, <32 x i16> %1, <32 x i16> %2) + %4 = bitcast i32 %__U to <32 x i1> + %5 = select <32 x i1> %4, <32 x i16> %3, <32 x i16> %1 + %6 = bitcast <32 x i16> %5 to <8 x i64> + ret <8 x i64> %6 +} + +define <8 x i64> @test_mm512_permutex2var_epi16(<8 x i64> %__A, <8 x i64> %__I, <8 x i64> %__B) { +; X32-LABEL: test_mm512_permutex2var_epi16: +; X32: # %bb.0: # %entry +; X32-NEXT: vpermt2w %zmm2, %zmm1, %zmm0 +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_permutex2var_epi16: +; X64: # %bb.0: # %entry +; X64-NEXT: vpermt2w %zmm2, %zmm1, %zmm0 +; X64-NEXT: retq +entry: + %0 = bitcast <8 x i64> %__A to <32 x i16> + %1 = bitcast <8 x i64> %__I to <32 x i16> + %2 = bitcast <8 x i64> %__B to <32 x i16> + %3 = tail call <32 x i16> @llvm.x86.avx512.vpermi2var.hi.512(<32 x i16> %0, <32 x i16> %1, <32 x i16> %2) + %4 = bitcast <32 x i16> %3 to <8 x i64> + ret <8 x i64> %4 +} + +define <8 x i64> @test_mm512_mask_permutex2var_epi16(<8 x i64> %__A, i32 %__U, <8 x i64> %__I, <8 x i64> %__B) { +; X32-LABEL: test_mm512_mask_permutex2var_epi16: +; X32: # %bb.0: # %entry +; X32-NEXT: kmovd {{[0-9]+}}(%esp), %k1 +; X32-NEXT: vpermt2w %zmm2, %zmm1, %zmm0 {%k1} +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_mask_permutex2var_epi16: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovd %edi, %k1 +; X64-NEXT: vpermt2w %zmm2, %zmm1, %zmm0 {%k1} +; X64-NEXT: retq +entry: + %0 = bitcast <8 x i64> %__A to <32 x i16> + %1 = bitcast <8 x i64> %__I to <32 x i16> + %2 = bitcast <8 x i64> %__B to <32 x i16> + %3 = tail call <32 x i16> @llvm.x86.avx512.vpermi2var.hi.512(<32 x i16> %0, <32 x i16> %1, <32 x i16> %2) + %4 = bitcast i32 %__U to <32 x i1> + %5 = select <32 x i1> %4, <32 x i16> %3, <32 x i16> %0 + %6 = bitcast <32 x i16> %5 to <8 x i64> + ret <8 x i64> %6 +} + +define <8 x i64> @test_mm512_maskz_permutex2var_epi16(i32 %__U, <8 x i64> %__A, <8 x i64> %__I, <8 x i64> %__B) { +; X32-LABEL: test_mm512_maskz_permutex2var_epi16: +; X32: # %bb.0: # %entry +; X32-NEXT: kmovd {{[0-9]+}}(%esp), %k1 +; X32-NEXT: vpermt2w %zmm2, %zmm1, %zmm0 {%k1} {z} +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_maskz_permutex2var_epi16: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovd %edi, %k1 +; X64-NEXT: vpermt2w %zmm2, %zmm1, %zmm0 {%k1} {z} +; X64-NEXT: retq +entry: + %0 = bitcast <8 x i64> %__A to <32 x i16> + %1 = bitcast <8 x i64> %__I to <32 x i16> + %2 = bitcast <8 x i64> %__B to <32 x i16> + %3 = tail call <32 x i16> @llvm.x86.avx512.vpermi2var.hi.512(<32 x i16> %0, <32 x i16> %1, <32 x i16> %2) + %4 = bitcast i32 %__U to <32 x i1> + %5 = select <32 x i1> %4, <32 x i16> %3, <32 x i16> zeroinitializer + %6 = bitcast <32 x i16> %5 to <8 x i64> + ret <8 x i64> %6 +} + +declare <32 x i16> @llvm.x86.avx512.vpermi2var.hi.512(<32 x i16>, <32 x i16>, <32 x i16>) + !0 = !{i32 1} diff --git a/llvm/test/CodeGen/X86/avx512bwvl-intrinsics-fast-isel.ll b/llvm/test/CodeGen/X86/avx512bwvl-intrinsics-fast-isel.ll index 9ba23f46688..3d8ba8d381a 100644 --- a/llvm/test/CodeGen/X86/avx512bwvl-intrinsics-fast-isel.ll +++ b/llvm/test/CodeGen/X86/avx512bwvl-intrinsics-fast-isel.ll @@ -877,5 +877,191 @@ entry: ret <2 x i64> %3 } +define <2 x i64> @test_mm_mask2_permutex2var_epi16(<2 x i64> %__A, <2 x i64> %__I, i8 zeroext %__U, <2 x i64> %__B) { +; X32-LABEL: test_mm_mask2_permutex2var_epi16: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovd %eax, %k1 +; X32-NEXT: vpermi2w %xmm2, %xmm0, %xmm1 {%k1} +; X32-NEXT: vmovdqa %xmm1, %xmm0 +; X32-NEXT: retl +; +; X64-LABEL: test_mm_mask2_permutex2var_epi16: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovd %edi, %k1 +; X64-NEXT: vpermi2w %xmm2, %xmm0, %xmm1 {%k1} +; X64-NEXT: vmovdqa %xmm1, %xmm0 +; X64-NEXT: retq +entry: + %0 = bitcast <2 x i64> %__A to <8 x i16> + %1 = bitcast <2 x i64> %__I to <8 x i16> + %2 = bitcast <2 x i64> %__B to <8 x i16> + %3 = tail call <8 x i16> @llvm.x86.avx512.vpermi2var.hi.128(<8 x i16> %0, <8 x i16> %1, <8 x i16> %2) + %4 = bitcast i8 %__U to <8 x i1> + %5 = select <8 x i1> %4, <8 x i16> %3, <8 x i16> %1 + %6 = bitcast <8 x i16> %5 to <2 x i64> + ret <2 x i64> %6 +} + +define <4 x i64> @test_mm256_mask2_permutex2var_epi16(<4 x i64> %__A, <4 x i64> %__I, i16 zeroext %__U, <4 x i64> %__B) { +; X32-LABEL: test_mm256_mask2_permutex2var_epi16: +; X32: # %bb.0: # %entry +; X32-NEXT: kmovw {{[0-9]+}}(%esp), %k1 +; X32-NEXT: vpermi2w %ymm2, %ymm0, %ymm1 {%k1} +; X32-NEXT: vmovdqa %ymm1, %ymm0 +; X32-NEXT: retl +; +; X64-LABEL: test_mm256_mask2_permutex2var_epi16: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovd %edi, %k1 +; X64-NEXT: vpermi2w %ymm2, %ymm0, %ymm1 {%k1} +; X64-NEXT: vmovdqa %ymm1, %ymm0 +; X64-NEXT: retq +entry: + %0 = bitcast <4 x i64> %__A to <16 x i16> + %1 = bitcast <4 x i64> %__I to <16 x i16> + %2 = bitcast <4 x i64> %__B to <16 x i16> + %3 = tail call <16 x i16> @llvm.x86.avx512.vpermi2var.hi.256(<16 x i16> %0, <16 x i16> %1, <16 x i16> %2) + %4 = bitcast i16 %__U to <16 x i1> + %5 = select <16 x i1> %4, <16 x i16> %3, <16 x i16> %1 + %6 = bitcast <16 x i16> %5 to <4 x i64> + ret <4 x i64> %6 +} + +define <2 x i64> @test_mm_permutex2var_epi16(<2 x i64> %__A, <2 x i64> %__I, <2 x i64> %__B) { +; X32-LABEL: test_mm_permutex2var_epi16: +; X32: # %bb.0: # %entry +; X32-NEXT: vpermt2w %xmm2, %xmm1, %xmm0 +; X32-NEXT: retl +; +; X64-LABEL: test_mm_permutex2var_epi16: +; X64: # %bb.0: # %entry +; X64-NEXT: vpermt2w %xmm2, %xmm1, %xmm0 +; X64-NEXT: retq +entry: + %0 = bitcast <2 x i64> %__A to <8 x i16> + %1 = bitcast <2 x i64> %__I to <8 x i16> + %2 = bitcast <2 x i64> %__B to <8 x i16> + %3 = tail call <8 x i16> @llvm.x86.avx512.vpermi2var.hi.128(<8 x i16> %0, <8 x i16> %1, <8 x i16> %2) + %4 = bitcast <8 x i16> %3 to <2 x i64> + ret <2 x i64> %4 +} + +define <2 x i64> @test_mm_mask_permutex2var_epi16(<2 x i64> %__A, i8 zeroext %__U, <2 x i64> %__I, <2 x i64> %__B) { +; X32-LABEL: test_mm_mask_permutex2var_epi16: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovd %eax, %k1 +; X32-NEXT: vpermt2w %xmm2, %xmm1, %xmm0 {%k1} +; X32-NEXT: retl +; +; X64-LABEL: test_mm_mask_permutex2var_epi16: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovd %edi, %k1 +; X64-NEXT: vpermt2w %xmm2, %xmm1, %xmm0 {%k1} +; X64-NEXT: retq +entry: + %0 = bitcast <2 x i64> %__A to <8 x i16> + %1 = bitcast <2 x i64> %__I to <8 x i16> + %2 = bitcast <2 x i64> %__B to <8 x i16> + %3 = tail call <8 x i16> @llvm.x86.avx512.vpermi2var.hi.128(<8 x i16> %0, <8 x i16> %1, <8 x i16> %2) + %4 = bitcast i8 %__U to <8 x i1> + %5 = select <8 x i1> %4, <8 x i16> %3, <8 x i16> %0 + %6 = bitcast <8 x i16> %5 to <2 x i64> + ret <2 x i64> %6 +} + +define <2 x i64> @test_mm_maskz_permutex2var_epi16(i8 zeroext %__U, <2 x i64> %__A, <2 x i64> %__I, <2 x i64> %__B) { +; X32-LABEL: test_mm_maskz_permutex2var_epi16: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovd %eax, %k1 +; X32-NEXT: vpermt2w %xmm2, %xmm1, %xmm0 {%k1} {z} +; X32-NEXT: retl +; +; X64-LABEL: test_mm_maskz_permutex2var_epi16: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovd %edi, %k1 +; X64-NEXT: vpermt2w %xmm2, %xmm1, %xmm0 {%k1} {z} +; X64-NEXT: retq +entry: + %0 = bitcast <2 x i64> %__A to <8 x i16> + %1 = bitcast <2 x i64> %__I to <8 x i16> + %2 = bitcast <2 x i64> %__B to <8 x i16> + %3 = tail call <8 x i16> @llvm.x86.avx512.vpermi2var.hi.128(<8 x i16> %0, <8 x i16> %1, <8 x i16> %2) + %4 = bitcast i8 %__U to <8 x i1> + %5 = select <8 x i1> %4, <8 x i16> %3, <8 x i16> zeroinitializer + %6 = bitcast <8 x i16> %5 to <2 x i64> + ret <2 x i64> %6 +} + +define <4 x i64> @test_mm256_permutex2var_epi16(<4 x i64> %__A, <4 x i64> %__I, <4 x i64> %__B) { +; X32-LABEL: test_mm256_permutex2var_epi16: +; X32: # %bb.0: # %entry +; X32-NEXT: vpermt2w %ymm2, %ymm1, %ymm0 +; X32-NEXT: retl +; +; X64-LABEL: test_mm256_permutex2var_epi16: +; X64: # %bb.0: # %entry +; X64-NEXT: vpermt2w %ymm2, %ymm1, %ymm0 +; X64-NEXT: retq +entry: + %0 = bitcast <4 x i64> %__A to <16 x i16> + %1 = bitcast <4 x i64> %__I to <16 x i16> + %2 = bitcast <4 x i64> %__B to <16 x i16> + %3 = tail call <16 x i16> @llvm.x86.avx512.vpermi2var.hi.256(<16 x i16> %0, <16 x i16> %1, <16 x i16> %2) + %4 = bitcast <16 x i16> %3 to <4 x i64> + ret <4 x i64> %4 +} + +define <4 x i64> @test_mm256_mask_permutex2var_epi16(<4 x i64> %__A, i16 zeroext %__U, <4 x i64> %__I, <4 x i64> %__B) { +; X32-LABEL: test_mm256_mask_permutex2var_epi16: +; X32: # %bb.0: # %entry +; X32-NEXT: kmovw {{[0-9]+}}(%esp), %k1 +; X32-NEXT: vpermt2w %ymm2, %ymm1, %ymm0 {%k1} +; X32-NEXT: retl +; +; X64-LABEL: test_mm256_mask_permutex2var_epi16: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovd %edi, %k1 +; X64-NEXT: vpermt2w %ymm2, %ymm1, %ymm0 {%k1} +; X64-NEXT: retq +entry: + %0 = bitcast <4 x i64> %__A to <16 x i16> + %1 = bitcast <4 x i64> %__I to <16 x i16> + %2 = bitcast <4 x i64> %__B to <16 x i16> + %3 = tail call <16 x i16> @llvm.x86.avx512.vpermi2var.hi.256(<16 x i16> %0, <16 x i16> %1, <16 x i16> %2) + %4 = bitcast i16 %__U to <16 x i1> + %5 = select <16 x i1> %4, <16 x i16> %3, <16 x i16> %0 + %6 = bitcast <16 x i16> %5 to <4 x i64> + ret <4 x i64> %6 +} + +define <4 x i64> @test_mm256_maskz_permutex2var_epi16(i16 zeroext %__U, <4 x i64> %__A, <4 x i64> %__I, <4 x i64> %__B) { +; X32-LABEL: test_mm256_maskz_permutex2var_epi16: +; X32: # %bb.0: # %entry +; X32-NEXT: kmovw {{[0-9]+}}(%esp), %k1 +; X32-NEXT: vpermt2w %ymm2, %ymm1, %ymm0 {%k1} {z} +; X32-NEXT: retl +; +; X64-LABEL: test_mm256_maskz_permutex2var_epi16: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovd %edi, %k1 +; X64-NEXT: vpermt2w %ymm2, %ymm1, %ymm0 {%k1} {z} +; X64-NEXT: retq +entry: + %0 = bitcast <4 x i64> %__A to <16 x i16> + %1 = bitcast <4 x i64> %__I to <16 x i16> + %2 = bitcast <4 x i64> %__B to <16 x i16> + %3 = tail call <16 x i16> @llvm.x86.avx512.vpermi2var.hi.256(<16 x i16> %0, <16 x i16> %1, <16 x i16> %2) + %4 = bitcast i16 %__U to <16 x i1> + %5 = select <16 x i1> %4, <16 x i16> %3, <16 x i16> zeroinitializer + %6 = bitcast <16 x i16> %5 to <4 x i64> + ret <4 x i64> %6 +} + +declare <8 x i16> @llvm.x86.avx512.vpermi2var.hi.128(<8 x i16>, <8 x i16>, <8 x i16>) +declare <16 x i16> @llvm.x86.avx512.vpermi2var.hi.256(<16 x i16>, <16 x i16>, <16 x i16>) + !0 = !{i32 1} diff --git a/llvm/test/CodeGen/X86/avx512vbmi-intrinsics-fast-isel.ll b/llvm/test/CodeGen/X86/avx512vbmi-intrinsics-fast-isel.ll new file mode 100644 index 00000000000..c405116250d --- /dev/null +++ b/llvm/test/CodeGen/X86/avx512vbmi-intrinsics-fast-isel.ll @@ -0,0 +1,103 @@ +; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py +; RUN: llc < %s -fast-isel -mtriple=i386-unknown-unknown -mattr=+avx512f,+avx512vbmi | FileCheck %s --check-prefix=ALL --check-prefix=X32 +; RUN: llc < %s -fast-isel -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+avx512vbmi | FileCheck %s --check-prefix=ALL --check-prefix=X64 + +; NOTE: This should use IR equivalent to what is generated by clang/test/CodeGen/avx512vbmi-builtins.c + +define <8 x i64> @test_mm512_mask2_permutex2var_epi8(<8 x i64> %__A, <8 x i64> %__I, i64 %__U, <8 x i64> %__B) { +; X32-LABEL: test_mm512_mask2_permutex2var_epi8: +; X32: # %bb.0: # %entry +; X32-NEXT: kmovd {{[0-9]+}}(%esp), %k0 +; X32-NEXT: kmovd {{[0-9]+}}(%esp), %k1 +; X32-NEXT: kunpckdq %k1, %k0, %k1 +; X32-NEXT: vpermi2b %zmm2, %zmm0, %zmm1 {%k1} +; X32-NEXT: vmovdqa64 %zmm1, %zmm0 +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_mask2_permutex2var_epi8: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovq %rdi, %k1 +; X64-NEXT: vpermi2b %zmm2, %zmm0, %zmm1 {%k1} +; X64-NEXT: vmovdqa64 %zmm1, %zmm0 +; X64-NEXT: retq +entry: + %0 = bitcast <8 x i64> %__A to <64 x i8> + %1 = bitcast <8 x i64> %__I to <64 x i8> + %2 = bitcast <8 x i64> %__B to <64 x i8> + %3 = tail call <64 x i8> @llvm.x86.avx512.vpermi2var.qi.512(<64 x i8> %0, <64 x i8> %1, <64 x i8> %2) + %4 = bitcast i64 %__U to <64 x i1> + %5 = select <64 x i1> %4, <64 x i8> %3, <64 x i8> %1 + %6 = bitcast <64 x i8> %5 to <8 x i64> + ret <8 x i64> %6 +} + +define <8 x i64> @test_mm512_permutex2var_epi8(<8 x i64> %__A, <8 x i64> %__I, <8 x i64> %__B) { +; X32-LABEL: test_mm512_permutex2var_epi8: +; X32: # %bb.0: # %entry +; X32-NEXT: vpermt2b %zmm2, %zmm1, %zmm0 +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_permutex2var_epi8: +; X64: # %bb.0: # %entry +; X64-NEXT: vpermt2b %zmm2, %zmm1, %zmm0 +; X64-NEXT: retq +entry: + %0 = bitcast <8 x i64> %__A to <64 x i8> + %1 = bitcast <8 x i64> %__I to <64 x i8> + %2 = bitcast <8 x i64> %__B to <64 x i8> + %3 = tail call <64 x i8> @llvm.x86.avx512.vpermi2var.qi.512(<64 x i8> %0, <64 x i8> %1, <64 x i8> %2) + %4 = bitcast <64 x i8> %3 to <8 x i64> + ret <8 x i64> %4 +} + +define <8 x i64> @test_mm512_mask_permutex2var_epi8(<8 x i64> %__A, i64 %__U, <8 x i64> %__I, <8 x i64> %__B) { +; X32-LABEL: test_mm512_mask_permutex2var_epi8: +; X32: # %bb.0: # %entry +; X32-NEXT: kmovd {{[0-9]+}}(%esp), %k0 +; X32-NEXT: kmovd {{[0-9]+}}(%esp), %k1 +; X32-NEXT: kunpckdq %k1, %k0, %k1 +; X32-NEXT: vpermt2b %zmm2, %zmm1, %zmm0 {%k1} +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_mask_permutex2var_epi8: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovq %rdi, %k1 +; X64-NEXT: vpermt2b %zmm2, %zmm1, %zmm0 {%k1} +; X64-NEXT: retq +entry: + %0 = bitcast <8 x i64> %__A to <64 x i8> + %1 = bitcast <8 x i64> %__I to <64 x i8> + %2 = bitcast <8 x i64> %__B to <64 x i8> + %3 = tail call <64 x i8> @llvm.x86.avx512.vpermi2var.qi.512(<64 x i8> %0, <64 x i8> %1, <64 x i8> %2) + %4 = bitcast i64 %__U to <64 x i1> + %5 = select <64 x i1> %4, <64 x i8> %3, <64 x i8> %0 + %6 = bitcast <64 x i8> %5 to <8 x i64> + ret <8 x i64> %6 +} + +define <8 x i64> @test_mm512_maskz_permutex2var_epi8(i64 %__U, <8 x i64> %__A, <8 x i64> %__I, <8 x i64> %__B) { +; X32-LABEL: test_mm512_maskz_permutex2var_epi8: +; X32: # %bb.0: # %entry +; X32-NEXT: kmovd {{[0-9]+}}(%esp), %k0 +; X32-NEXT: kmovd {{[0-9]+}}(%esp), %k1 +; X32-NEXT: kunpckdq %k1, %k0, %k1 +; X32-NEXT: vpermt2b %zmm2, %zmm1, %zmm0 {%k1} {z} +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_maskz_permutex2var_epi8: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovq %rdi, %k1 +; X64-NEXT: vpermt2b %zmm2, %zmm1, %zmm0 {%k1} {z} +; X64-NEXT: retq +entry: + %0 = bitcast <8 x i64> %__A to <64 x i8> + %1 = bitcast <8 x i64> %__I to <64 x i8> + %2 = bitcast <8 x i64> %__B to <64 x i8> + %3 = tail call <64 x i8> @llvm.x86.avx512.vpermi2var.qi.512(<64 x i8> %0, <64 x i8> %1, <64 x i8> %2) + %4 = bitcast i64 %__U to <64 x i1> + %5 = select <64 x i1> %4, <64 x i8> %3, <64 x i8> zeroinitializer + %6 = bitcast <64 x i8> %5 to <8 x i64> + ret <8 x i64> %6 +} + +declare <64 x i8> @llvm.x86.avx512.vpermi2var.qi.512(<64 x i8>, <64 x i8>, <64 x i8>) diff --git a/llvm/test/CodeGen/X86/avx512vbmivl-intrinsics-fast-isel.ll b/llvm/test/CodeGen/X86/avx512vbmivl-intrinsics-fast-isel.ll new file mode 100644 index 00000000000..cc565293064 --- /dev/null +++ b/llvm/test/CodeGen/X86/avx512vbmivl-intrinsics-fast-isel.ll @@ -0,0 +1,188 @@ +; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py +; RUN: llc < %s -fast-isel -mtriple=i386-unknown-unknown -mattr=+avx512f,+avx512vbmi,+avx512vl | FileCheck %s --check-prefix=ALL --check-prefix=X32 +; RUN: llc < %s -fast-isel -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+avx512vbmi,+avx512vl | FileCheck %s --check-prefix=ALL --check-prefix=X64 + +; NOTE: This should use IR equivalent to what is generated by clang/test/CodeGen/avx512vlbw-builtins.c + +define <2 x i64> @test_mm_mask2_permutex2var_epi8(<2 x i64> %__A, <2 x i64> %__I, i16 zeroext %__U, <2 x i64> %__B) { +; X32-LABEL: test_mm_mask2_permutex2var_epi8: +; X32: # %bb.0: # %entry +; X32-NEXT: kmovw {{[0-9]+}}(%esp), %k1 +; X32-NEXT: vpermi2b %xmm2, %xmm0, %xmm1 {%k1} +; X32-NEXT: vmovdqa %xmm1, %xmm0 +; X32-NEXT: retl +; +; X64-LABEL: test_mm_mask2_permutex2var_epi8: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovd %edi, %k1 +; X64-NEXT: vpermi2b %xmm2, %xmm0, %xmm1 {%k1} +; X64-NEXT: vmovdqa %xmm1, %xmm0 +; X64-NEXT: retq +entry: + %0 = bitcast <2 x i64> %__A to <16 x i8> + %1 = bitcast <2 x i64> %__I to <16 x i8> + %2 = bitcast <2 x i64> %__B to <16 x i8> + %3 = tail call <16 x i8> @llvm.x86.avx512.vpermi2var.qi.128(<16 x i8> %0, <16 x i8> %1, <16 x i8> %2) + %4 = bitcast i16 %__U to <16 x i1> + %5 = select <16 x i1> %4, <16 x i8> %3, <16 x i8> %1 + %6 = bitcast <16 x i8> %5 to <2 x i64> + ret <2 x i64> %6 +} + +define <4 x i64> @test_mm256_mask2_permutex2var_epi8(<4 x i64> %__A, <4 x i64> %__I, i32 %__U, <4 x i64> %__B) { +; X32-LABEL: test_mm256_mask2_permutex2var_epi8: +; X32: # %bb.0: # %entry +; X32-NEXT: kmovd {{[0-9]+}}(%esp), %k1 +; X32-NEXT: vpermi2b %ymm2, %ymm0, %ymm1 {%k1} +; X32-NEXT: vmovdqa %ymm1, %ymm0 +; X32-NEXT: retl +; +; X64-LABEL: test_mm256_mask2_permutex2var_epi8: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovd %edi, %k1 +; X64-NEXT: vpermi2b %ymm2, %ymm0, %ymm1 {%k1} +; X64-NEXT: vmovdqa %ymm1, %ymm0 +; X64-NEXT: retq +entry: + %0 = bitcast <4 x i64> %__A to <32 x i8> + %1 = bitcast <4 x i64> %__I to <32 x i8> + %2 = bitcast <4 x i64> %__B to <32 x i8> + %3 = tail call <32 x i8> @llvm.x86.avx512.vpermi2var.qi.256(<32 x i8> %0, <32 x i8> %1, <32 x i8> %2) + %4 = bitcast i32 %__U to <32 x i1> + %5 = select <32 x i1> %4, <32 x i8> %3, <32 x i8> %1 + %6 = bitcast <32 x i8> %5 to <4 x i64> + ret <4 x i64> %6 +} + +define <2 x i64> @test_mm_permutex2var_epi8(<2 x i64> %__A, <2 x i64> %__I, <2 x i64> %__B) { +; X32-LABEL: test_mm_permutex2var_epi8: +; X32: # %bb.0: # %entry +; X32-NEXT: vpermt2b %xmm2, %xmm1, %xmm0 +; X32-NEXT: retl +; +; X64-LABEL: test_mm_permutex2var_epi8: +; X64: # %bb.0: # %entry +; X64-NEXT: vpermt2b %xmm2, %xmm1, %xmm0 +; X64-NEXT: retq +entry: + %0 = bitcast <2 x i64> %__A to <16 x i8> + %1 = bitcast <2 x i64> %__I to <16 x i8> + %2 = bitcast <2 x i64> %__B to <16 x i8> + %3 = tail call <16 x i8> @llvm.x86.avx512.vpermi2var.qi.128(<16 x i8> %0, <16 x i8> %1, <16 x i8> %2) + %4 = bitcast <16 x i8> %3 to <2 x i64> + ret <2 x i64> %4 +} + +define <2 x i64> @test_mm_mask_permutex2var_epi8(<2 x i64> %__A, i16 zeroext %__U, <2 x i64> %__I, <2 x i64> %__B) { +; X32-LABEL: test_mm_mask_permutex2var_epi8: +; X32: # %bb.0: # %entry +; X32-NEXT: kmovw {{[0-9]+}}(%esp), %k1 +; X32-NEXT: vpermt2b %xmm2, %xmm1, %xmm0 {%k1} +; X32-NEXT: retl +; +; X64-LABEL: test_mm_mask_permutex2var_epi8: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovd %edi, %k1 +; X64-NEXT: vpermt2b %xmm2, %xmm1, %xmm0 {%k1} +; X64-NEXT: retq +entry: + %0 = bitcast <2 x i64> %__A to <16 x i8> + %1 = bitcast <2 x i64> %__I to <16 x i8> + %2 = bitcast <2 x i64> %__B to <16 x i8> + %3 = tail call <16 x i8> @llvm.x86.avx512.vpermi2var.qi.128(<16 x i8> %0, <16 x i8> %1, <16 x i8> %2) + %4 = bitcast i16 %__U to <16 x i1> + %5 = select <16 x i1> %4, <16 x i8> %3, <16 x i8> %0 + %6 = bitcast <16 x i8> %5 to <2 x i64> + ret <2 x i64> %6 +} + +define <2 x i64> @test_mm_maskz_permutex2var_epi8(i16 zeroext %__U, <2 x i64> %__A, <2 x i64> %__I, <2 x i64> %__B) { +; X32-LABEL: test_mm_maskz_permutex2var_epi8: +; X32: # %bb.0: # %entry +; X32-NEXT: kmovw {{[0-9]+}}(%esp), %k1 +; X32-NEXT: vpermt2b %xmm2, %xmm1, %xmm0 {%k1} {z} +; X32-NEXT: retl +; +; X64-LABEL: test_mm_maskz_permutex2var_epi8: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovd %edi, %k1 +; X64-NEXT: vpermt2b %xmm2, %xmm1, %xmm0 {%k1} {z} +; X64-NEXT: retq +entry: + %0 = bitcast <2 x i64> %__A to <16 x i8> + %1 = bitcast <2 x i64> %__I to <16 x i8> + %2 = bitcast <2 x i64> %__B to <16 x i8> + %3 = tail call <16 x i8> @llvm.x86.avx512.vpermi2var.qi.128(<16 x i8> %0, <16 x i8> %1, <16 x i8> %2) + %4 = bitcast i16 %__U to <16 x i1> + %5 = select <16 x i1> %4, <16 x i8> %3, <16 x i8> zeroinitializer + %6 = bitcast <16 x i8> %5 to <2 x i64> + ret <2 x i64> %6 +} + +define <4 x i64> @test_mm256_permutex2var_epi8(<4 x i64> %__A, <4 x i64> %__I, <4 x i64> %__B) { +; X32-LABEL: test_mm256_permutex2var_epi8: +; X32: # %bb.0: # %entry +; X32-NEXT: vpermt2b %ymm2, %ymm1, %ymm0 +; X32-NEXT: retl +; +; X64-LABEL: test_mm256_permutex2var_epi8: +; X64: # %bb.0: # %entry +; X64-NEXT: vpermt2b %ymm2, %ymm1, %ymm0 +; X64-NEXT: retq +entry: + %0 = bitcast <4 x i64> %__A to <32 x i8> + %1 = bitcast <4 x i64> %__I to <32 x i8> + %2 = bitcast <4 x i64> %__B to <32 x i8> + %3 = tail call <32 x i8> @llvm.x86.avx512.vpermi2var.qi.256(<32 x i8> %0, <32 x i8> %1, <32 x i8> %2) + %4 = bitcast <32 x i8> %3 to <4 x i64> + ret <4 x i64> %4 +} + +define <4 x i64> @test_mm256_mask_permutex2var_epi8(<4 x i64> %__A, i32 %__U, <4 x i64> %__I, <4 x i64> %__B) { +; X32-LABEL: test_mm256_mask_permutex2var_epi8: +; X32: # %bb.0: # %entry +; X32-NEXT: kmovd {{[0-9]+}}(%esp), %k1 +; X32-NEXT: vpermt2b %ymm2, %ymm1, %ymm0 {%k1} +; X32-NEXT: retl +; +; X64-LABEL: test_mm256_mask_permutex2var_epi8: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovd %edi, %k1 +; X64-NEXT: vpermt2b %ymm2, %ymm1, %ymm0 {%k1} +; X64-NEXT: retq +entry: + %0 = bitcast <4 x i64> %__A to <32 x i8> + %1 = bitcast <4 x i64> %__I to <32 x i8> + %2 = bitcast <4 x i64> %__B to <32 x i8> + %3 = tail call <32 x i8> @llvm.x86.avx512.vpermi2var.qi.256(<32 x i8> %0, <32 x i8> %1, <32 x i8> %2) + %4 = bitcast i32 %__U to <32 x i1> + %5 = select <32 x i1> %4, <32 x i8> %3, <32 x i8> %0 + %6 = bitcast <32 x i8> %5 to <4 x i64> + ret <4 x i64> %6 +} + +define <4 x i64> @test_mm256_maskz_permutex2var_epi8(i32 %__U, <4 x i64> %__A, <4 x i64> %__I, <4 x i64> %__B) { +; X32-LABEL: test_mm256_maskz_permutex2var_epi8: +; X32: # %bb.0: # %entry +; X32-NEXT: kmovd {{[0-9]+}}(%esp), %k1 +; X32-NEXT: vpermt2b %ymm2, %ymm1, %ymm0 {%k1} {z} +; X32-NEXT: retl +; +; X64-LABEL: test_mm256_maskz_permutex2var_epi8: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovd %edi, %k1 +; X64-NEXT: vpermt2b %ymm2, %ymm1, %ymm0 {%k1} {z} +; X64-NEXT: retq +entry: + %0 = bitcast <4 x i64> %__A to <32 x i8> + %1 = bitcast <4 x i64> %__I to <32 x i8> + %2 = bitcast <4 x i64> %__B to <32 x i8> + %3 = tail call <32 x i8> @llvm.x86.avx512.vpermi2var.qi.256(<32 x i8> %0, <32 x i8> %1, <32 x i8> %2) + %4 = bitcast i32 %__U to <32 x i1> + %5 = select <32 x i1> %4, <32 x i8> %3, <32 x i8> zeroinitializer + %6 = bitcast <32 x i8> %5 to <4 x i64> + ret <4 x i64> %6 +} + +declare <16 x i8> @llvm.x86.avx512.vpermi2var.qi.128(<16 x i8>, <16 x i8>, <16 x i8>) +declare <32 x i8> @llvm.x86.avx512.vpermi2var.qi.256(<32 x i8>, <32 x i8>, <32 x i8>) diff --git a/llvm/test/CodeGen/X86/avx512vl-intrinsics-fast-isel.ll b/llvm/test/CodeGen/X86/avx512vl-intrinsics-fast-isel.ll index 586dac8172c..0f5c20e46b3 100644 --- a/llvm/test/CodeGen/X86/avx512vl-intrinsics-fast-isel.ll +++ b/llvm/test/CodeGen/X86/avx512vl-intrinsics-fast-isel.ll @@ -3804,6 +3804,685 @@ entry: ret <4 x i64> %2 } +define <2 x i64> @test_mm_mask2_permutex2var_epi32(<2 x i64> %__A, <2 x i64> %__I, i8 zeroext %__U, <2 x i64> %__B) { +; X32-LABEL: test_mm_mask2_permutex2var_epi32: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vpermi2d %xmm2, %xmm0, %xmm1 {%k1} +; X32-NEXT: vmovdqa %xmm1, %xmm0 +; X32-NEXT: retl +; +; X64-LABEL: test_mm_mask2_permutex2var_epi32: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vpermi2d %xmm2, %xmm0, %xmm1 {%k1} +; X64-NEXT: vmovdqa %xmm1, %xmm0 +; X64-NEXT: retq +entry: + %0 = bitcast <2 x i64> %__A to <4 x i32> + %1 = bitcast <2 x i64> %__I to <4 x i32> + %2 = bitcast <2 x i64> %__B to <4 x i32> + %3 = tail call <4 x i32> @llvm.x86.avx512.vpermi2var.d.128(<4 x i32> %0, <4 x i32> %1, <4 x i32> %2) + %4 = bitcast i8 %__U to <8 x i1> + %extract.i = shufflevector <8 x i1> %4, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3> + %5 = select <4 x i1> %extract.i, <4 x i32> %3, <4 x i32> %1 + %6 = bitcast <4 x i32> %5 to <2 x i64> + ret <2 x i64> %6 +} + +define <4 x i64> @test_mm256_mask2_permutex2var_epi32(<4 x i64> %__A, <4 x i64> %__I, i8 zeroext %__U, <4 x i64> %__B) { +; X32-LABEL: test_mm256_mask2_permutex2var_epi32: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vpermi2d %ymm2, %ymm0, %ymm1 {%k1} +; X32-NEXT: vmovdqa %ymm1, %ymm0 +; X32-NEXT: retl +; +; X64-LABEL: test_mm256_mask2_permutex2var_epi32: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vpermi2d %ymm2, %ymm0, %ymm1 {%k1} +; X64-NEXT: vmovdqa %ymm1, %ymm0 +; X64-NEXT: retq +entry: + %0 = bitcast <4 x i64> %__A to <8 x i32> + %1 = bitcast <4 x i64> %__I to <8 x i32> + %2 = bitcast <4 x i64> %__B to <8 x i32> + %3 = tail call <8 x i32> @llvm.x86.avx512.vpermi2var.d.256(<8 x i32> %0, <8 x i32> %1, <8 x i32> %2) + %4 = bitcast i8 %__U to <8 x i1> + %5 = select <8 x i1> %4, <8 x i32> %3, <8 x i32> %1 + %6 = bitcast <8 x i32> %5 to <4 x i64> + ret <4 x i64> %6 +} + +define <2 x double> @test_mm_mask2_permutex2var_pd(<2 x double> %__A, <2 x i64> %__I, i8 zeroext %__U, <2 x double> %__B) { +; X32-LABEL: test_mm_mask2_permutex2var_pd: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vpermi2pd %xmm2, %xmm0, %xmm1 {%k1} +; X32-NEXT: vmovapd %xmm1, %xmm0 +; X32-NEXT: retl +; +; X64-LABEL: test_mm_mask2_permutex2var_pd: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vpermi2pd %xmm2, %xmm0, %xmm1 {%k1} +; X64-NEXT: vmovapd %xmm1, %xmm0 +; X64-NEXT: retq +entry: + %0 = tail call <2 x double> @llvm.x86.avx512.vpermi2var.pd.128(<2 x double> %__A, <2 x i64> %__I, <2 x double> %__B) + %1 = bitcast <2 x i64> %__I to <2 x double> + %2 = bitcast i8 %__U to <8 x i1> + %extract.i = shufflevector <8 x i1> %2, <8 x i1> undef, <2 x i32> <i32 0, i32 1> + %3 = select <2 x i1> %extract.i, <2 x double> %0, <2 x double> %1 + ret <2 x double> %3 +} + +define <4 x double> @test_mm256_mask2_permutex2var_pd(<4 x double> %__A, <4 x i64> %__I, i8 zeroext %__U, <4 x double> %__B) { +; X32-LABEL: test_mm256_mask2_permutex2var_pd: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vpermi2pd %ymm2, %ymm0, %ymm1 {%k1} +; X32-NEXT: vmovapd %ymm1, %ymm0 +; X32-NEXT: retl +; +; X64-LABEL: test_mm256_mask2_permutex2var_pd: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vpermi2pd %ymm2, %ymm0, %ymm1 {%k1} +; X64-NEXT: vmovapd %ymm1, %ymm0 +; X64-NEXT: retq +entry: + %0 = tail call <4 x double> @llvm.x86.avx512.vpermi2var.pd.256(<4 x double> %__A, <4 x i64> %__I, <4 x double> %__B) + %1 = bitcast <4 x i64> %__I to <4 x double> + %2 = bitcast i8 %__U to <8 x i1> + %extract.i = shufflevector <8 x i1> %2, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3> + %3 = select <4 x i1> %extract.i, <4 x double> %0, <4 x double> %1 + ret <4 x double> %3 +} + +define <4 x float> @test_mm_mask2_permutex2var_ps(<4 x float> %__A, <2 x i64> %__I, i8 zeroext %__U, <4 x float> %__B) { +; X32-LABEL: test_mm_mask2_permutex2var_ps: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vpermi2ps %xmm2, %xmm0, %xmm1 {%k1} +; X32-NEXT: vmovaps %xmm1, %xmm0 +; X32-NEXT: retl +; +; X64-LABEL: test_mm_mask2_permutex2var_ps: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vpermi2ps %xmm2, %xmm0, %xmm1 {%k1} +; X64-NEXT: vmovaps %xmm1, %xmm0 +; X64-NEXT: retq +entry: + %0 = bitcast <2 x i64> %__I to <4 x i32> + %1 = tail call <4 x float> @llvm.x86.avx512.vpermi2var.ps.128(<4 x float> %__A, <4 x i32> %0, <4 x float> %__B) + %2 = bitcast <2 x i64> %__I to <4 x float> + %3 = bitcast i8 %__U to <8 x i1> + %extract.i = shufflevector <8 x i1> %3, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3> + %4 = select <4 x i1> %extract.i, <4 x float> %1, <4 x float> %2 + ret <4 x float> %4 +} + +define <8 x float> @test_mm256_mask2_permutex2var_ps(<8 x float> %__A, <4 x i64> %__I, i8 zeroext %__U, <8 x float> %__B) { +; X32-LABEL: test_mm256_mask2_permutex2var_ps: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vpermi2ps %ymm2, %ymm0, %ymm1 {%k1} +; X32-NEXT: vmovaps %ymm1, %ymm0 +; X32-NEXT: retl +; +; X64-LABEL: test_mm256_mask2_permutex2var_ps: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vpermi2ps %ymm2, %ymm0, %ymm1 {%k1} +; X64-NEXT: vmovaps %ymm1, %ymm0 +; X64-NEXT: retq +entry: + %0 = bitcast <4 x i64> %__I to <8 x i32> + %1 = tail call <8 x float> @llvm.x86.avx512.vpermi2var.ps.256(<8 x float> %__A, <8 x i32> %0, <8 x float> %__B) + %2 = bitcast <4 x i64> %__I to <8 x float> + %3 = bitcast i8 %__U to <8 x i1> + %4 = select <8 x i1> %3, <8 x float> %1, <8 x float> %2 + ret <8 x float> %4 +} + +define <2 x i64> @test_mm_mask2_permutex2var_epi64(<2 x i64> %__A, <2 x i64> %__I, i8 zeroext %__U, <2 x i64> %__B) { +; X32-LABEL: test_mm_mask2_permutex2var_epi64: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vpermi2q %xmm2, %xmm0, %xmm1 {%k1} +; X32-NEXT: vmovdqa %xmm1, %xmm0 +; X32-NEXT: retl +; +; X64-LABEL: test_mm_mask2_permutex2var_epi64: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vpermi2q %xmm2, %xmm0, %xmm1 {%k1} +; X64-NEXT: vmovdqa %xmm1, %xmm0 +; X64-NEXT: retq +entry: + %0 = tail call <2 x i64> @llvm.x86.avx512.vpermi2var.q.128(<2 x i64> %__A, <2 x i64> %__I, <2 x i64> %__B) + %1 = bitcast i8 %__U to <8 x i1> + %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <2 x i32> <i32 0, i32 1> + %2 = select <2 x i1> %extract.i, <2 x i64> %0, <2 x i64> %__I + ret <2 x i64> %2 +} + +define <4 x i64> @test_mm256_mask2_permutex2var_epi64(<4 x i64> %__A, <4 x i64> %__I, i8 zeroext %__U, <4 x i64> %__B) { +; X32-LABEL: test_mm256_mask2_permutex2var_epi64: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vpermi2q %ymm2, %ymm0, %ymm1 {%k1} +; X32-NEXT: vmovdqa %ymm1, %ymm0 +; X32-NEXT: retl +; +; X64-LABEL: test_mm256_mask2_permutex2var_epi64: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vpermi2q %ymm2, %ymm0, %ymm1 {%k1} +; X64-NEXT: vmovdqa %ymm1, %ymm0 +; X64-NEXT: retq +entry: + %0 = tail call <4 x i64> @llvm.x86.avx512.vpermi2var.q.256(<4 x i64> %__A, <4 x i64> %__I, <4 x i64> %__B) + %1 = bitcast i8 %__U to <8 x i1> + %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3> + %2 = select <4 x i1> %extract.i, <4 x i64> %0, <4 x i64> %__I + ret <4 x i64> %2 +} + +define <2 x i64> @test_mm_permutex2var_epi32(<2 x i64> %__A, <2 x i64> %__I, <2 x i64> %__B) { +; X32-LABEL: test_mm_permutex2var_epi32: +; X32: # %bb.0: # %entry +; X32-NEXT: vpermt2d %xmm2, %xmm1, %xmm0 +; X32-NEXT: retl +; +; X64-LABEL: test_mm_permutex2var_epi32: +; X64: # %bb.0: # %entry +; X64-NEXT: vpermt2d %xmm2, %xmm1, %xmm0 +; X64-NEXT: retq +entry: + %0 = bitcast <2 x i64> %__A to <4 x i32> + %1 = bitcast <2 x i64> %__I to <4 x i32> + %2 = bitcast <2 x i64> %__B to <4 x i32> + %3 = tail call <4 x i32> @llvm.x86.avx512.vpermi2var.d.128(<4 x i32> %0, <4 x i32> %1, <4 x i32> %2) + %4 = bitcast <4 x i32> %3 to <2 x i64> + ret <2 x i64> %4 +} + +define <2 x i64> @test_mm_mask_permutex2var_epi32(<2 x i64> %__A, i8 zeroext %__U, <2 x i64> %__I, <2 x i64> %__B) { +; X32-LABEL: test_mm_mask_permutex2var_epi32: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vpermt2d %xmm2, %xmm1, %xmm0 {%k1} +; X32-NEXT: retl +; +; X64-LABEL: test_mm_mask_permutex2var_epi32: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vpermt2d %xmm2, %xmm1, %xmm0 {%k1} +; X64-NEXT: retq +entry: + %0 = bitcast <2 x i64> %__A to <4 x i32> + %1 = bitcast <2 x i64> %__I to <4 x i32> + %2 = bitcast <2 x i64> %__B to <4 x i32> + %3 = tail call <4 x i32> @llvm.x86.avx512.vpermi2var.d.128(<4 x i32> %0, <4 x i32> %1, <4 x i32> %2) + %4 = bitcast i8 %__U to <8 x i1> + %extract.i = shufflevector <8 x i1> %4, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3> + %5 = select <4 x i1> %extract.i, <4 x i32> %3, <4 x i32> %0 + %6 = bitcast <4 x i32> %5 to <2 x i64> + ret <2 x i64> %6 +} + +define <2 x i64> @test_mm_maskz_permutex2var_epi32(i8 zeroext %__U, <2 x i64> %__A, <2 x i64> %__I, <2 x i64> %__B) { +; X32-LABEL: test_mm_maskz_permutex2var_epi32: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vpermt2d %xmm2, %xmm1, %xmm0 {%k1} {z} +; X32-NEXT: retl +; +; X64-LABEL: test_mm_maskz_permutex2var_epi32: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vpermt2d %xmm2, %xmm1, %xmm0 {%k1} {z} +; X64-NEXT: retq +entry: + %0 = bitcast <2 x i64> %__A to <4 x i32> + %1 = bitcast <2 x i64> %__I to <4 x i32> + %2 = bitcast <2 x i64> %__B to <4 x i32> + %3 = tail call <4 x i32> @llvm.x86.avx512.vpermi2var.d.128(<4 x i32> %0, <4 x i32> %1, <4 x i32> %2) + %4 = bitcast i8 %__U to <8 x i1> + %extract.i = shufflevector <8 x i1> %4, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3> + %5 = select <4 x i1> %extract.i, <4 x i32> %3, <4 x i32> zeroinitializer + %6 = bitcast <4 x i32> %5 to <2 x i64> + ret <2 x i64> %6 +} + +define <4 x i64> @test_mm256_permutex2var_epi32(<4 x i64> %__A, <4 x i64> %__I, <4 x i64> %__B) { +; X32-LABEL: test_mm256_permutex2var_epi32: +; X32: # %bb.0: # %entry +; X32-NEXT: vpermt2d %ymm2, %ymm1, %ymm0 +; X32-NEXT: retl +; +; X64-LABEL: test_mm256_permutex2var_epi32: +; X64: # %bb.0: # %entry +; X64-NEXT: vpermt2d %ymm2, %ymm1, %ymm0 +; X64-NEXT: retq +entry: + %0 = bitcast <4 x i64> %__A to <8 x i32> + %1 = bitcast <4 x i64> %__I to <8 x i32> + %2 = bitcast <4 x i64> %__B to <8 x i32> + %3 = tail call <8 x i32> @llvm.x86.avx512.vpermi2var.d.256(<8 x i32> %0, <8 x i32> %1, <8 x i32> %2) + %4 = bitcast <8 x i32> %3 to <4 x i64> + ret <4 x i64> %4 +} + +define <4 x i64> @test_mm256_mask_permutex2var_epi32(<4 x i64> %__A, i8 zeroext %__U, <4 x i64> %__I, <4 x i64> %__B) { +; X32-LABEL: test_mm256_mask_permutex2var_epi32: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vpermt2d %ymm2, %ymm1, %ymm0 {%k1} +; X32-NEXT: retl +; +; X64-LABEL: test_mm256_mask_permutex2var_epi32: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vpermt2d %ymm2, %ymm1, %ymm0 {%k1} +; X64-NEXT: retq +entry: + %0 = bitcast <4 x i64> %__A to <8 x i32> + %1 = bitcast <4 x i64> %__I to <8 x i32> + %2 = bitcast <4 x i64> %__B to <8 x i32> + %3 = tail call <8 x i32> @llvm.x86.avx512.vpermi2var.d.256(<8 x i32> %0, <8 x i32> %1, <8 x i32> %2) + %4 = bitcast i8 %__U to <8 x i1> + %5 = select <8 x i1> %4, <8 x i32> %3, <8 x i32> %0 + %6 = bitcast <8 x i32> %5 to <4 x i64> + ret <4 x i64> %6 +} + +define <4 x i64> @test_mm256_maskz_permutex2var_epi32(i8 zeroext %__U, <4 x i64> %__A, <4 x i64> %__I, <4 x i64> %__B) { +; X32-LABEL: test_mm256_maskz_permutex2var_epi32: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vpermt2d %ymm2, %ymm1, %ymm0 {%k1} {z} +; X32-NEXT: retl +; +; X64-LABEL: test_mm256_maskz_permutex2var_epi32: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vpermt2d %ymm2, %ymm1, %ymm0 {%k1} {z} +; X64-NEXT: retq +entry: + %0 = bitcast <4 x i64> %__A to <8 x i32> + %1 = bitcast <4 x i64> %__I to <8 x i32> + %2 = bitcast <4 x i64> %__B to <8 x i32> + %3 = tail call <8 x i32> @llvm.x86.avx512.vpermi2var.d.256(<8 x i32> %0, <8 x i32> %1, <8 x i32> %2) + %4 = bitcast i8 %__U to <8 x i1> + %5 = select <8 x i1> %4, <8 x i32> %3, <8 x i32> zeroinitializer + %6 = bitcast <8 x i32> %5 to <4 x i64> + ret <4 x i64> %6 +} + +define <2 x double> @test_mm_permutex2var_pd(<2 x double> %__A, <2 x i64> %__I, <2 x double> %__B) { +; X32-LABEL: test_mm_permutex2var_pd: +; X32: # %bb.0: # %entry +; X32-NEXT: vpermt2pd %xmm2, %xmm1, %xmm0 +; X32-NEXT: retl +; +; X64-LABEL: test_mm_permutex2var_pd: +; X64: # %bb.0: # %entry +; X64-NEXT: vpermt2pd %xmm2, %xmm1, %xmm0 +; X64-NEXT: retq +entry: + %0 = tail call <2 x double> @llvm.x86.avx512.vpermi2var.pd.128(<2 x double> %__A, <2 x i64> %__I, <2 x double> %__B) + ret <2 x double> %0 +} + +define <2 x double> @test_mm_mask_permutex2var_pd(<2 x double> %__A, i8 zeroext %__U, <2 x i64> %__I, <2 x double> %__B) { +; X32-LABEL: test_mm_mask_permutex2var_pd: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vpermt2pd %xmm2, %xmm1, %xmm0 {%k1} +; X32-NEXT: retl +; +; X64-LABEL: test_mm_mask_permutex2var_pd: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vpermt2pd %xmm2, %xmm1, %xmm0 {%k1} +; X64-NEXT: retq +entry: + %0 = tail call <2 x double> @llvm.x86.avx512.vpermi2var.pd.128(<2 x double> %__A, <2 x i64> %__I, <2 x double> %__B) + %1 = bitcast i8 %__U to <8 x i1> + %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <2 x i32> <i32 0, i32 1> + %2 = select <2 x i1> %extract.i, <2 x double> %0, <2 x double> %__A + ret <2 x double> %2 +} + +define <2 x double> @test_mm_maskz_permutex2var_pd(i8 zeroext %__U, <2 x double> %__A, <2 x i64> %__I, <2 x double> %__B) { +; X32-LABEL: test_mm_maskz_permutex2var_pd: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vpermt2pd %xmm2, %xmm1, %xmm0 {%k1} {z} +; X32-NEXT: retl +; +; X64-LABEL: test_mm_maskz_permutex2var_pd: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vpermt2pd %xmm2, %xmm1, %xmm0 {%k1} {z} +; X64-NEXT: retq +entry: + %0 = tail call <2 x double> @llvm.x86.avx512.vpermi2var.pd.128(<2 x double> %__A, <2 x i64> %__I, <2 x double> %__B) + %1 = bitcast i8 %__U to <8 x i1> + %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <2 x i32> <i32 0, i32 1> + %2 = select <2 x i1> %extract.i, <2 x double> %0, <2 x double> zeroinitializer + ret <2 x double> %2 +} + +define <4 x double> @test_mm256_permutex2var_pd(<4 x double> %__A, <4 x i64> %__I, <4 x double> %__B) { +; X32-LABEL: test_mm256_permutex2var_pd: +; X32: # %bb.0: # %entry +; X32-NEXT: vpermt2pd %ymm2, %ymm1, %ymm0 +; X32-NEXT: retl +; +; X64-LABEL: test_mm256_permutex2var_pd: +; X64: # %bb.0: # %entry +; X64-NEXT: vpermt2pd %ymm2, %ymm1, %ymm0 +; X64-NEXT: retq +entry: + %0 = tail call <4 x double> @llvm.x86.avx512.vpermi2var.pd.256(<4 x double> %__A, <4 x i64> %__I, <4 x double> %__B) + ret <4 x double> %0 +} + +define <4 x double> @test_mm256_mask_permutex2var_pd(<4 x double> %__A, i8 zeroext %__U, <4 x i64> %__I, <4 x double> %__B) { +; X32-LABEL: test_mm256_mask_permutex2var_pd: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vpermt2pd %ymm2, %ymm1, %ymm0 {%k1} +; X32-NEXT: retl +; +; X64-LABEL: test_mm256_mask_permutex2var_pd: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vpermt2pd %ymm2, %ymm1, %ymm0 {%k1} +; X64-NEXT: retq +entry: + %0 = tail call <4 x double> @llvm.x86.avx512.vpermi2var.pd.256(<4 x double> %__A, <4 x i64> %__I, <4 x double> %__B) + %1 = bitcast i8 %__U to <8 x i1> + %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3> + %2 = select <4 x i1> %extract.i, <4 x double> %0, <4 x double> %__A + ret <4 x double> %2 +} + +define <4 x double> @test_mm256_maskz_permutex2var_pd(i8 zeroext %__U, <4 x double> %__A, <4 x i64> %__I, <4 x double> %__B) { +; X32-LABEL: test_mm256_maskz_permutex2var_pd: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vpermt2pd %ymm2, %ymm1, %ymm0 {%k1} {z} +; X32-NEXT: retl +; +; X64-LABEL: test_mm256_maskz_permutex2var_pd: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vpermt2pd %ymm2, %ymm1, %ymm0 {%k1} {z} +; X64-NEXT: retq +entry: + %0 = tail call <4 x double> @llvm.x86.avx512.vpermi2var.pd.256(<4 x double> %__A, <4 x i64> %__I, <4 x double> %__B) + %1 = bitcast i8 %__U to <8 x i1> + %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3> + %2 = select <4 x i1> %extract.i, <4 x double> %0, <4 x double> zeroinitializer + ret <4 x double> %2 +} + +define <4 x float> @test_mm_permutex2var_ps(<4 x float> %__A, <2 x i64> %__I, <4 x float> %__B) { +; X32-LABEL: test_mm_permutex2var_ps: +; X32: # %bb.0: # %entry +; X32-NEXT: vpermt2ps %xmm2, %xmm1, %xmm0 +; X32-NEXT: retl +; +; X64-LABEL: test_mm_permutex2var_ps: +; X64: # %bb.0: # %entry +; X64-NEXT: vpermt2ps %xmm2, %xmm1, %xmm0 +; X64-NEXT: retq +entry: + %0 = bitcast <2 x i64> %__I to <4 x i32> + %1 = tail call <4 x float> @llvm.x86.avx512.vpermi2var.ps.128(<4 x float> %__A, <4 x i32> %0, <4 x float> %__B) + ret <4 x float> %1 +} + +define <4 x float> @test_mm_mask_permutex2var_ps(<4 x float> %__A, i8 zeroext %__U, <2 x i64> %__I, <4 x float> %__B) { +; X32-LABEL: test_mm_mask_permutex2var_ps: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vpermt2ps %xmm2, %xmm1, %xmm0 {%k1} +; X32-NEXT: retl +; +; X64-LABEL: test_mm_mask_permutex2var_ps: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vpermt2ps %xmm2, %xmm1, %xmm0 {%k1} +; X64-NEXT: retq +entry: + %0 = bitcast <2 x i64> %__I to <4 x i32> + %1 = tail call <4 x float> @llvm.x86.avx512.vpermi2var.ps.128(<4 x float> %__A, <4 x i32> %0, <4 x float> %__B) + %2 = bitcast i8 %__U to <8 x i1> + %extract.i = shufflevector <8 x i1> %2, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3> + %3 = select <4 x i1> %extract.i, <4 x float> %1, <4 x float> %__A + ret <4 x float> %3 +} + +define <4 x float> @test_mm_maskz_permutex2var_ps(i8 zeroext %__U, <4 x float> %__A, <2 x i64> %__I, <4 x float> %__B) { +; X32-LABEL: test_mm_maskz_permutex2var_ps: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vpermt2ps %xmm2, %xmm1, %xmm0 {%k1} {z} +; X32-NEXT: retl +; +; X64-LABEL: test_mm_maskz_permutex2var_ps: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vpermt2ps %xmm2, %xmm1, %xmm0 {%k1} {z} +; X64-NEXT: retq +entry: + %0 = bitcast <2 x i64> %__I to <4 x i32> + %1 = tail call <4 x float> @llvm.x86.avx512.vpermi2var.ps.128(<4 x float> %__A, <4 x i32> %0, <4 x float> %__B) + %2 = bitcast i8 %__U to <8 x i1> + %extract.i = shufflevector <8 x i1> %2, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3> + %3 = select <4 x i1> %extract.i, <4 x float> %1, <4 x float> zeroinitializer + ret <4 x float> %3 +} + +define <8 x float> @test_mm256_permutex2var_ps(<8 x float> %__A, <4 x i64> %__I, <8 x float> %__B) { +; X32-LABEL: test_mm256_permutex2var_ps: +; X32: # %bb.0: # %entry +; X32-NEXT: vpermt2ps %ymm2, %ymm1, %ymm0 +; X32-NEXT: retl +; +; X64-LABEL: test_mm256_permutex2var_ps: +; X64: # %bb.0: # %entry +; X64-NEXT: vpermt2ps %ymm2, %ymm1, %ymm0 +; X64-NEXT: retq +entry: + %0 = bitcast <4 x i64> %__I to <8 x i32> + %1 = tail call <8 x float> @llvm.x86.avx512.vpermi2var.ps.256(<8 x float> %__A, <8 x i32> %0, <8 x float> %__B) + ret <8 x float> %1 +} + +define <8 x float> @test_mm256_mask_permutex2var_ps(<8 x float> %__A, i8 zeroext %__U, <4 x i64> %__I, <8 x float> %__B) { +; X32-LABEL: test_mm256_mask_permutex2var_ps: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vpermt2ps %ymm2, %ymm1, %ymm0 {%k1} +; X32-NEXT: retl +; +; X64-LABEL: test_mm256_mask_permutex2var_ps: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vpermt2ps %ymm2, %ymm1, %ymm0 {%k1} +; X64-NEXT: retq +entry: + %0 = bitcast <4 x i64> %__I to <8 x i32> + %1 = tail call <8 x float> @llvm.x86.avx512.vpermi2var.ps.256(<8 x float> %__A, <8 x i32> %0, <8 x float> %__B) + %2 = bitcast i8 %__U to <8 x i1> + %3 = select <8 x i1> %2, <8 x float> %1, <8 x float> %__A + ret <8 x float> %3 +} + +define <8 x float> @test_mm256_maskz_permutex2var_ps(i8 zeroext %__U, <8 x float> %__A, <4 x i64> %__I, <8 x float> %__B) { +; X32-LABEL: test_mm256_maskz_permutex2var_ps: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vpermt2ps %ymm2, %ymm1, %ymm0 {%k1} {z} +; X32-NEXT: retl +; +; X64-LABEL: test_mm256_maskz_permutex2var_ps: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vpermt2ps %ymm2, %ymm1, %ymm0 {%k1} {z} +; X64-NEXT: retq +entry: + %0 = bitcast <4 x i64> %__I to <8 x i32> + %1 = tail call <8 x float> @llvm.x86.avx512.vpermi2var.ps.256(<8 x float> %__A, <8 x i32> %0, <8 x float> %__B) + %2 = bitcast i8 %__U to <8 x i1> + %3 = select <8 x i1> %2, <8 x float> %1, <8 x float> zeroinitializer + ret <8 x float> %3 +} + +define <2 x i64> @test_mm_permutex2var_epi64(<2 x i64> %__A, <2 x i64> %__I, <2 x i64> %__B) { +; X32-LABEL: test_mm_permutex2var_epi64: +; X32: # %bb.0: # %entry +; X32-NEXT: vpermt2q %xmm2, %xmm1, %xmm0 +; X32-NEXT: retl +; +; X64-LABEL: test_mm_permutex2var_epi64: +; X64: # %bb.0: # %entry +; X64-NEXT: vpermt2q %xmm2, %xmm1, %xmm0 +; X64-NEXT: retq +entry: + %0 = tail call <2 x i64> @llvm.x86.avx512.vpermi2var.q.128(<2 x i64> %__A, <2 x i64> %__I, <2 x i64> %__B) + ret <2 x i64> %0 +} + +define <2 x i64> @test_mm_mask_permutex2var_epi64(<2 x i64> %__A, i8 zeroext %__U, <2 x i64> %__I, <2 x i64> %__B) { +; X32-LABEL: test_mm_mask_permutex2var_epi64: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vpermt2q %xmm2, %xmm1, %xmm0 {%k1} +; X32-NEXT: retl +; +; X64-LABEL: test_mm_mask_permutex2var_epi64: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vpermt2q %xmm2, %xmm1, %xmm0 {%k1} +; X64-NEXT: retq +entry: + %0 = tail call <2 x i64> @llvm.x86.avx512.vpermi2var.q.128(<2 x i64> %__A, <2 x i64> %__I, <2 x i64> %__B) + %1 = bitcast i8 %__U to <8 x i1> + %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <2 x i32> <i32 0, i32 1> + %2 = select <2 x i1> %extract.i, <2 x i64> %0, <2 x i64> %__A + ret <2 x i64> %2 +} + +define <2 x i64> @test_mm_maskz_permutex2var_epi64(i8 zeroext %__U, <2 x i64> %__A, <2 x i64> %__I, <2 x i64> %__B) { +; X32-LABEL: test_mm_maskz_permutex2var_epi64: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vpermt2q %xmm2, %xmm1, %xmm0 {%k1} {z} +; X32-NEXT: retl +; +; X64-LABEL: test_mm_maskz_permutex2var_epi64: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vpermt2q %xmm2, %xmm1, %xmm0 {%k1} {z} +; X64-NEXT: retq +entry: + %0 = tail call <2 x i64> @llvm.x86.avx512.vpermi2var.q.128(<2 x i64> %__A, <2 x i64> %__I, <2 x i64> %__B) + %1 = bitcast i8 %__U to <8 x i1> + %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <2 x i32> <i32 0, i32 1> + %2 = select <2 x i1> %extract.i, <2 x i64> %0, <2 x i64> zeroinitializer + ret <2 x i64> %2 +} + +define <4 x i64> @test_mm256_permutex2var_epi64(<4 x i64> %__A, <4 x i64> %__I, <4 x i64> %__B) { +; X32-LABEL: test_mm256_permutex2var_epi64: +; X32: # %bb.0: # %entry +; X32-NEXT: vpermt2q %ymm2, %ymm1, %ymm0 +; X32-NEXT: retl +; +; X64-LABEL: test_mm256_permutex2var_epi64: +; X64: # %bb.0: # %entry +; X64-NEXT: vpermt2q %ymm2, %ymm1, %ymm0 +; X64-NEXT: retq +entry: + %0 = tail call <4 x i64> @llvm.x86.avx512.vpermi2var.q.256(<4 x i64> %__A, <4 x i64> %__I, <4 x i64> %__B) + ret <4 x i64> %0 +} + +define <4 x i64> @test_mm256_mask_permutex2var_epi64(<4 x i64> %__A, i8 zeroext %__U, <4 x i64> %__I, <4 x i64> %__B) { +; X32-LABEL: test_mm256_mask_permutex2var_epi64: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vpermt2q %ymm2, %ymm1, %ymm0 {%k1} +; X32-NEXT: retl +; +; X64-LABEL: test_mm256_mask_permutex2var_epi64: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vpermt2q %ymm2, %ymm1, %ymm0 {%k1} +; X64-NEXT: retq +entry: + %0 = tail call <4 x i64> @llvm.x86.avx512.vpermi2var.q.256(<4 x i64> %__A, <4 x i64> %__I, <4 x i64> %__B) + %1 = bitcast i8 %__U to <8 x i1> + %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3> + %2 = select <4 x i1> %extract.i, <4 x i64> %0, <4 x i64> %__A + ret <4 x i64> %2 +} + +define <4 x i64> @test_mm256_maskz_permutex2var_epi64(i8 zeroext %__U, <4 x i64> %__A, <4 x i64> %__I, <4 x i64> %__B) { +; X32-LABEL: test_mm256_maskz_permutex2var_epi64: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vpermt2q %ymm2, %ymm1, %ymm0 {%k1} {z} +; X32-NEXT: retl +; +; X64-LABEL: test_mm256_maskz_permutex2var_epi64: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vpermt2q %ymm2, %ymm1, %ymm0 {%k1} {z} +; X64-NEXT: retq +entry: + %0 = tail call <4 x i64> @llvm.x86.avx512.vpermi2var.q.256(<4 x i64> %__A, <4 x i64> %__I, <4 x i64> %__B) + %1 = bitcast i8 %__U to <8 x i1> + %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3> + %2 = select <4 x i1> %extract.i, <4 x i64> %0, <4 x i64> zeroinitializer + ret <4 x i64> %2 +} + + declare <4 x i32> @llvm.x86.avx512.mask.cvtpd2dq.128(<2 x double>, <4 x i32>, i8) declare <4 x i32> @llvm.x86.avx.cvt.pd2dq.256(<4 x double>) declare <4 x float> @llvm.x86.avx512.mask.cvtpd2ps(<2 x double>, <4 x float>, i8) @@ -3823,5 +4502,13 @@ declare <8 x i32> @llvm.x86.avx.cvtt.ps2dq.256(<8 x float>) declare <4 x i32> @llvm.x86.avx512.mask.cvttps2udq.128(<4 x float>, <4 x i32>, i8) declare <8 x i32> @llvm.x86.avx512.mask.cvttps2udq.256(<8 x float>, <8 x i32>, i8) declare <8 x i16> @llvm.x86.avx512.mask.pmov.dw.256(<8 x i32>, <8 x i16>, i8) +declare <4 x i32> @llvm.x86.avx512.vpermi2var.d.128(<4 x i32>, <4 x i32>, <4 x i32>) +declare <8 x i32> @llvm.x86.avx512.vpermi2var.d.256(<8 x i32>, <8 x i32>, <8 x i32>) +declare <2 x double> @llvm.x86.avx512.vpermi2var.pd.128(<2 x double>, <2 x i64>, <2 x double>) +declare <4 x double> @llvm.x86.avx512.vpermi2var.pd.256(<4 x double>, <4 x i64>, <4 x double>) +declare <4 x float> @llvm.x86.avx512.vpermi2var.ps.128(<4 x float>, <4 x i32>, <4 x float>) +declare <8 x float> @llvm.x86.avx512.vpermi2var.ps.256(<8 x float>, <8 x i32>, <8 x float>) +declare <2 x i64> @llvm.x86.avx512.vpermi2var.q.128(<2 x i64>, <2 x i64>, <2 x i64>) +declare <4 x i64> @llvm.x86.avx512.vpermi2var.q.256(<4 x i64>, <4 x i64>, <4 x i64>) !0 = !{i32 1} |
