diff options
Diffstat (limited to 'include/llvm/IR/IntrinsicsX86.td')
| -rw-r--r-- | include/llvm/IR/IntrinsicsX86.td | 581 |
1 files changed, 503 insertions, 78 deletions
diff --git a/include/llvm/IR/IntrinsicsX86.td b/include/llvm/IR/IntrinsicsX86.td index 59ff946..60deb32 100644 --- a/include/llvm/IR/IntrinsicsX86.td +++ b/include/llvm/IR/IntrinsicsX86.td @@ -453,19 +453,6 @@ let TargetPrefix = "x86" in { // All intrinsics start with "llvm.x86.". def int_x86_sse2_psrai_d : GCCBuiltin<"__builtin_ia32_psradi128">, Intrinsic<[llvm_v4i32_ty], [llvm_v4i32_ty, llvm_i32_ty], [IntrNoMem]>; - - def int_x86_sse2_psll_dq : GCCBuiltin<"__builtin_ia32_pslldqi128">, - Intrinsic<[llvm_v2i64_ty], [llvm_v2i64_ty, - llvm_i32_ty], [IntrNoMem]>; - def int_x86_sse2_psrl_dq : GCCBuiltin<"__builtin_ia32_psrldqi128">, - Intrinsic<[llvm_v2i64_ty], [llvm_v2i64_ty, - llvm_i32_ty], [IntrNoMem]>; - def int_x86_sse2_psll_dq_bs : GCCBuiltin<"__builtin_ia32_pslldqi128_byteshift">, - Intrinsic<[llvm_v2i64_ty], [llvm_v2i64_ty, - llvm_i32_ty], [IntrNoMem]>; - def int_x86_sse2_psrl_dq_bs : GCCBuiltin<"__builtin_ia32_psrldqi128_byteshift">, - Intrinsic<[llvm_v2i64_ty], [llvm_v2i64_ty, - llvm_i32_ty], [IntrNoMem]>; } // Conversion ops @@ -1363,6 +1350,12 @@ let TargetPrefix = "x86" in { // All intrinsics start with "llvm.x86.". def int_x86_avx512_mask_loadu_pd_512 : GCCBuiltin<"__builtin_ia32_loadupd512_mask">, Intrinsic<[llvm_v8f64_ty], [llvm_ptr_ty, llvm_v8f64_ty, llvm_i8_ty], [IntrReadArgMem]>; + def int_x86_avx512_mask_load_ps_512 : GCCBuiltin<"__builtin_ia32_loadaps512_mask">, + Intrinsic<[llvm_v16f32_ty], [llvm_ptr_ty, llvm_v16f32_ty, llvm_i16_ty], + [IntrReadArgMem]>; + def int_x86_avx512_mask_load_pd_512 : GCCBuiltin<"__builtin_ia32_loadapd512_mask">, + Intrinsic<[llvm_v8f64_ty], [llvm_ptr_ty, llvm_v8f64_ty, llvm_i8_ty], + [IntrReadArgMem]>; } // Conditional store ops @@ -1389,6 +1382,14 @@ let TargetPrefix = "x86" in { // All intrinsics start with "llvm.x86.". GCCBuiltin<"__builtin_ia32_storeupd512_mask">, Intrinsic<[], [llvm_ptr_ty, llvm_v8f64_ty, llvm_i8_ty], [IntrReadWriteArgMem]>; + def int_x86_avx512_mask_store_ps_512 : + GCCBuiltin<"__builtin_ia32_storeaps512_mask">, + Intrinsic<[], [llvm_ptr_ty, llvm_v16f32_ty, llvm_i16_ty], + [IntrReadWriteArgMem]>; + def int_x86_avx512_mask_store_pd_512 : + GCCBuiltin<"__builtin_ia32_storeapd512_mask">, + Intrinsic<[], [llvm_ptr_ty, llvm_v8f64_ty, llvm_i8_ty], + [IntrReadWriteArgMem]>; def int_x86_avx512_mask_store_ss : GCCBuiltin<"__builtin_ia32_storess_mask">, Intrinsic<[], [llvm_ptr_ty, llvm_v4f32_ty, llvm_i8_ty], @@ -1572,19 +1573,6 @@ let TargetPrefix = "x86" in { // All intrinsics start with "llvm.x86.". Intrinsic<[llvm_v8i32_ty], [llvm_v8i32_ty, llvm_i32_ty], [IntrNoMem]>; - def int_x86_avx2_psll_dq : GCCBuiltin<"__builtin_ia32_pslldqi256">, - Intrinsic<[llvm_v4i64_ty], [llvm_v4i64_ty, - llvm_i32_ty], [IntrNoMem]>; - def int_x86_avx2_psrl_dq : GCCBuiltin<"__builtin_ia32_psrldqi256">, - Intrinsic<[llvm_v4i64_ty], [llvm_v4i64_ty, - llvm_i32_ty], [IntrNoMem]>; - def int_x86_avx2_psll_dq_bs : GCCBuiltin<"__builtin_ia32_pslldqi256_byteshift">, - Intrinsic<[llvm_v4i64_ty], [llvm_v4i64_ty, - llvm_i32_ty], [IntrNoMem]>; - def int_x86_avx2_psrl_dq_bs : GCCBuiltin<"__builtin_ia32_psrldqi256_byteshift">, - Intrinsic<[llvm_v4i64_ty], [llvm_v4i64_ty, - llvm_i32_ty], [IntrNoMem]>; - def int_x86_avx512_mask_pslli_d : GCCBuiltin<"__builtin_ia32_pslldi512">, Intrinsic<[llvm_v16i32_ty], [llvm_v16i32_ty, llvm_i32_ty, llvm_v16i32_ty, llvm_i16_ty], [IntrNoMem]>; @@ -1603,6 +1591,25 @@ let TargetPrefix = "x86" in { // All intrinsics start with "llvm.x86.". def int_x86_avx512_mask_psrai_q : GCCBuiltin<"__builtin_ia32_psraqi512">, Intrinsic<[llvm_v8i64_ty], [llvm_v8i64_ty, llvm_i32_ty, llvm_v8i64_ty, llvm_i8_ty], [IntrNoMem]>; + + def int_x86_avx512_mask_psll_d : GCCBuiltin<"__builtin_ia32_pslld512_mask">, + Intrinsic<[llvm_v16i32_ty], [llvm_v16i32_ty, + llvm_v4i32_ty, llvm_v16i32_ty, llvm_i16_ty], [IntrNoMem]>; + def int_x86_avx512_mask_psll_q : GCCBuiltin<"__builtin_ia32_psllq512_mask">, + Intrinsic<[llvm_v8i64_ty], [llvm_v8i64_ty, + llvm_v2i64_ty, llvm_v8i64_ty, llvm_i8_ty], [IntrNoMem]>; + def int_x86_avx512_mask_psrl_d : GCCBuiltin<"__builtin_ia32_psrld512_mask">, + Intrinsic<[llvm_v16i32_ty], [llvm_v16i32_ty, + llvm_v4i32_ty, llvm_v16i32_ty, llvm_i16_ty], [IntrNoMem]>; + def int_x86_avx512_mask_psrl_q : GCCBuiltin<"__builtin_ia32_psrlq512_mask">, + Intrinsic<[llvm_v8i64_ty], [llvm_v8i64_ty, + llvm_v2i64_ty, llvm_v8i64_ty, llvm_i8_ty], [IntrNoMem]>; + def int_x86_avx512_mask_psra_d : GCCBuiltin<"__builtin_ia32_psrad512_mask">, + Intrinsic<[llvm_v16i32_ty], [llvm_v16i32_ty, + llvm_v4i32_ty, llvm_v16i32_ty, llvm_i16_ty], [IntrNoMem]>; + def int_x86_avx512_mask_psra_q : GCCBuiltin<"__builtin_ia32_psraq512_mask">, + Intrinsic<[llvm_v8i64_ty], [llvm_v8i64_ty, + llvm_v2i64_ty, llvm_v8i64_ty, llvm_i8_ty], [IntrNoMem]>; } // Pack ops. @@ -1911,6 +1918,31 @@ let TargetPrefix = "x86" in { // All intrinsics start with "llvm.x86.". def int_x86_avx2_psrav_d_256 : GCCBuiltin<"__builtin_ia32_psrav8si">, Intrinsic<[llvm_v8i32_ty], [llvm_v8i32_ty, llvm_v8i32_ty], [IntrNoMem]>; + + def int_x86_avx512_mask_psllv_d : GCCBuiltin<"__builtin_ia32_psllv16si_mask">, + Intrinsic<[llvm_v16i32_ty], [llvm_v16i32_ty, + llvm_v16i32_ty, llvm_v16i32_ty, llvm_i16_ty], + [IntrNoMem]>; + def int_x86_avx512_mask_psllv_q : GCCBuiltin<"__builtin_ia32_psllv8di_mask">, + Intrinsic<[llvm_v8i64_ty], [llvm_v8i64_ty, + llvm_v8i64_ty, llvm_v8i64_ty, llvm_i8_ty], + [IntrNoMem]>; + def int_x86_avx512_mask_psrav_d : GCCBuiltin<"__builtin_ia32_psrav16si_mask">, + Intrinsic<[llvm_v16i32_ty], [llvm_v16i32_ty, + llvm_v16i32_ty, llvm_v16i32_ty, llvm_i16_ty], + [IntrNoMem]>; + def int_x86_avx512_mask_psrav_q : GCCBuiltin<"__builtin_ia32_psrav8di_mask">, + Intrinsic<[llvm_v8i64_ty], [llvm_v8i64_ty, + llvm_v8i64_ty, llvm_v8i64_ty, llvm_i8_ty], + [IntrNoMem]>; + def int_x86_avx512_mask_psrlv_d : GCCBuiltin<"__builtin_ia32_psrlv16si_mask">, + Intrinsic<[llvm_v16i32_ty], [llvm_v16i32_ty, + llvm_v16i32_ty, llvm_v16i32_ty, llvm_i16_ty], + [IntrNoMem]>; + def int_x86_avx512_mask_psrlv_q : GCCBuiltin<"__builtin_ia32_psrlv8di_mask">, + Intrinsic<[llvm_v8i64_ty], [llvm_v8i64_ty, + llvm_v8i64_ty, llvm_v8i64_ty, llvm_i8_ty], + [IntrNoMem]>; } // Gather ops @@ -2029,11 +2061,31 @@ let TargetPrefix = "x86" in { // All intrinsics start with "llvm.x86.". [llvm_v16f32_ty, llvm_v16f32_ty, llvm_v16f32_ty, llvm_i16_ty, llvm_i32_ty], [IntrNoMem]>; + def int_x86_fma_mask_vfmadd_ps_256 : GCCBuiltin<"__builtin_ia32_vfmaddps256_mask">, + Intrinsic<[llvm_v8f32_ty], + [llvm_v8f32_ty, llvm_v8f32_ty, llvm_v8f32_ty, + llvm_i8_ty], + [IntrNoMem]>; + def int_x86_fma_mask_vfmadd_ps_128 : GCCBuiltin<"__builtin_ia32_vfmaddps128_mask">, + Intrinsic<[llvm_v4f32_ty], + [llvm_v4f32_ty, llvm_v4f32_ty, llvm_v4f32_ty, + llvm_i8_ty], + [IntrNoMem]>; def int_x86_fma_mask_vfmadd_pd_512 : GCCBuiltin<"__builtin_ia32_vfmaddpd512_mask">, Intrinsic<[llvm_v8f64_ty], [llvm_v8f64_ty, llvm_v8f64_ty, llvm_v8f64_ty, llvm_i8_ty, llvm_i32_ty], [IntrNoMem]>; + def int_x86_fma_mask_vfmadd_pd_256 : GCCBuiltin<"__builtin_ia32_vfmaddpd256_mask">, + Intrinsic<[llvm_v4f64_ty], + [llvm_v4f64_ty, llvm_v4f64_ty, llvm_v4f64_ty, + llvm_i8_ty], + [IntrNoMem]>; + def int_x86_fma_mask_vfmadd_pd_128 : GCCBuiltin<"__builtin_ia32_vfmaddpd128_mask">, + Intrinsic<[llvm_v2f64_ty], + [llvm_v2f64_ty, llvm_v2f64_ty, llvm_v2f64_ty, + llvm_i8_ty], + [IntrNoMem]>; def int_x86_fma_vfmsub_ss : GCCBuiltin<"__builtin_ia32_vfmsubss">, Intrinsic<[llvm_v4f32_ty], [llvm_v4f32_ty, llvm_v4f32_ty, llvm_v4f32_ty], @@ -2063,11 +2115,31 @@ let TargetPrefix = "x86" in { // All intrinsics start with "llvm.x86.". [llvm_v16f32_ty, llvm_v16f32_ty, llvm_v16f32_ty, llvm_i16_ty, llvm_i32_ty], [IntrNoMem]>; + def int_x86_fma_mask_vfmsub_ps_256 : GCCBuiltin<"__builtin_ia32_vfmsubps256_mask">, + Intrinsic<[llvm_v8f32_ty], + [llvm_v8f32_ty, llvm_v8f32_ty, llvm_v8f32_ty, + llvm_i8_ty], + [IntrNoMem]>; + def int_x86_fma_mask_vfmsub_ps_128 : GCCBuiltin<"__builtin_ia32_vfmsubps128_mask">, + Intrinsic<[llvm_v4f32_ty], + [llvm_v4f32_ty, llvm_v4f32_ty, llvm_v4f32_ty, + llvm_i8_ty], + [IntrNoMem]>; def int_x86_fma_mask_vfmsub_pd_512 : GCCBuiltin<"__builtin_ia32_vfmsubpd512_mask">, Intrinsic<[llvm_v8f64_ty], [llvm_v8f64_ty, llvm_v8f64_ty, llvm_v8f64_ty, llvm_i8_ty, llvm_i32_ty], [IntrNoMem]>; + def int_x86_fma_mask_vfmsub_pd_256 : GCCBuiltin<"__builtin_ia32_vfmsubpd256_mask">, + Intrinsic<[llvm_v4f64_ty], + [llvm_v4f64_ty, llvm_v4f64_ty, llvm_v4f64_ty, + llvm_i8_ty], + [IntrNoMem]>; + def int_x86_fma_mask_vfmsub_pd_128 : GCCBuiltin<"__builtin_ia32_vfmsubpd128_mask">, + Intrinsic<[llvm_v2f64_ty], + [llvm_v2f64_ty, llvm_v2f64_ty, llvm_v2f64_ty, + llvm_i8_ty], + [IntrNoMem]>; def int_x86_fma_vfnmadd_ss : GCCBuiltin<"__builtin_ia32_vfnmaddss">, Intrinsic<[llvm_v4f32_ty], [llvm_v4f32_ty, llvm_v4f32_ty, llvm_v4f32_ty], @@ -2097,11 +2169,31 @@ let TargetPrefix = "x86" in { // All intrinsics start with "llvm.x86.". [llvm_v16f32_ty, llvm_v16f32_ty, llvm_v16f32_ty, llvm_i16_ty, llvm_i32_ty], [IntrNoMem]>; + def int_x86_fma_mask_vfnmadd_ps_256 : GCCBuiltin<"__builtin_ia32_vfnmaddps256_mask">, + Intrinsic<[llvm_v8f32_ty], + [llvm_v8f32_ty, llvm_v8f32_ty, llvm_v8f32_ty, + llvm_i8_ty], + [IntrNoMem]>; + def int_x86_fma_mask_vfnmadd_ps_128 : GCCBuiltin<"__builtin_ia32_vfnmaddps128_mask">, + Intrinsic<[llvm_v4f32_ty], + [llvm_v4f32_ty, llvm_v4f32_ty, llvm_v4f32_ty, + llvm_i8_ty], + [IntrNoMem]>; def int_x86_fma_mask_vfnmadd_pd_512 : GCCBuiltin<"__builtin_ia32_vfnmaddpd512_mask">, Intrinsic<[llvm_v8f64_ty], [llvm_v8f64_ty, llvm_v8f64_ty, llvm_v8f64_ty, llvm_i8_ty, llvm_i32_ty], [IntrNoMem]>; + def int_x86_fma_mask_vfnmadd_pd_256 : GCCBuiltin<"__builtin_ia32_vfnmaddpd256_mask">, + Intrinsic<[llvm_v4f64_ty], + [llvm_v4f64_ty, llvm_v4f64_ty, llvm_v4f64_ty, + llvm_i8_ty], + [IntrNoMem]>; + def int_x86_fma_mask_vfnmadd_pd_128 : GCCBuiltin<"__builtin_ia32_vfnmaddpd128_mask">, + Intrinsic<[llvm_v2f64_ty], + [llvm_v2f64_ty, llvm_v2f64_ty, llvm_v2f64_ty, + llvm_i8_ty], + [IntrNoMem]>; def int_x86_fma_vfnmsub_ss : GCCBuiltin<"__builtin_ia32_vfnmsubss">, Intrinsic<[llvm_v4f32_ty], [llvm_v4f32_ty, llvm_v4f32_ty, llvm_v4f32_ty], @@ -2131,11 +2223,31 @@ let TargetPrefix = "x86" in { // All intrinsics start with "llvm.x86.". [llvm_v16f32_ty, llvm_v16f32_ty, llvm_v16f32_ty, llvm_i16_ty, llvm_i32_ty], [IntrNoMem]>; + def int_x86_fma_mask_vfnmsub_ps_256 : GCCBuiltin<"__builtin_ia32_vfnmsubps256_mask">, + Intrinsic<[llvm_v8f32_ty], + [llvm_v8f32_ty, llvm_v8f32_ty, llvm_v8f32_ty, + llvm_i8_ty], + [IntrNoMem]>; + def int_x86_fma_mask_vfnmsub_ps_128 : GCCBuiltin<"__builtin_ia32_vfnmsubps128_mask">, + Intrinsic<[llvm_v4f32_ty], + [llvm_v4f32_ty, llvm_v4f32_ty, llvm_v4f32_ty, + llvm_i8_ty], + [IntrNoMem]>; def int_x86_fma_mask_vfnmsub_pd_512 : GCCBuiltin<"__builtin_ia32_vfnmsubpd512_mask">, Intrinsic<[llvm_v8f64_ty], [llvm_v8f64_ty, llvm_v8f64_ty, llvm_v8f64_ty, llvm_i8_ty, llvm_i32_ty], [IntrNoMem]>; + def int_x86_fma_mask_vfnmsub_pd_256 : GCCBuiltin<"__builtin_ia32_vfnmsubpd256_mask">, + Intrinsic<[llvm_v4f64_ty], + [llvm_v4f64_ty, llvm_v4f64_ty, llvm_v4f64_ty, + llvm_i8_ty], + [IntrNoMem]>; + def int_x86_fma_mask_vfnmsub_pd_128 : GCCBuiltin<"__builtin_ia32_vfnmsubpd128_mask">, + Intrinsic<[llvm_v2f64_ty], + [llvm_v2f64_ty, llvm_v2f64_ty, llvm_v2f64_ty, + llvm_i8_ty], + [IntrNoMem]>; def int_x86_fma_vfmaddsub_ps : GCCBuiltin<"__builtin_ia32_vfmaddsubps">, Intrinsic<[llvm_v4f32_ty], [llvm_v4f32_ty, llvm_v4f32_ty, llvm_v4f32_ty], @@ -2159,11 +2271,31 @@ let TargetPrefix = "x86" in { // All intrinsics start with "llvm.x86.". [llvm_v16f32_ty, llvm_v16f32_ty, llvm_v16f32_ty, llvm_i16_ty, llvm_i32_ty], [IntrNoMem]>; + def int_x86_fma_mask_vfmaddsub_ps_256 : GCCBuiltin<"__builtin_ia32_vfmaddsubps256_mask">, + Intrinsic<[llvm_v8f32_ty], + [llvm_v8f32_ty, llvm_v8f32_ty, llvm_v8f32_ty, + llvm_i8_ty], + [IntrNoMem]>; + def int_x86_fma_mask_vfmaddsub_ps_128 : GCCBuiltin<"__builtin_ia32_vfmaddsubps128_mask">, + Intrinsic<[llvm_v4f32_ty], + [llvm_v4f32_ty, llvm_v4f32_ty, llvm_v4f32_ty, + llvm_i8_ty], + [IntrNoMem]>; def int_x86_fma_mask_vfmaddsub_pd_512 : GCCBuiltin<"__builtin_ia32_vfmaddsubpd512_mask">, Intrinsic<[llvm_v8f64_ty], [llvm_v8f64_ty, llvm_v8f64_ty, llvm_v8f64_ty, llvm_i8_ty, llvm_i32_ty], [IntrNoMem]>; + def int_x86_fma_mask_vfmaddsub_pd_256 : GCCBuiltin<"__builtin_ia32_vfmaddsubpd256_mask">, + Intrinsic<[llvm_v4f64_ty], + [llvm_v4f64_ty, llvm_v4f64_ty, llvm_v4f64_ty, + llvm_i8_ty], + [IntrNoMem]>; + def int_x86_fma_mask_vfmaddsub_pd_128 : GCCBuiltin<"__builtin_ia32_vfmaddsubpd128_mask">, + Intrinsic<[llvm_v2f64_ty], + [llvm_v2f64_ty, llvm_v2f64_ty, llvm_v2f64_ty, + llvm_i8_ty], + [IntrNoMem]>; def int_x86_fma_vfmsubadd_ps : GCCBuiltin<"__builtin_ia32_vfmsubaddps">, Intrinsic<[llvm_v4f32_ty], [llvm_v4f32_ty, llvm_v4f32_ty, llvm_v4f32_ty], @@ -2187,11 +2319,31 @@ let TargetPrefix = "x86" in { // All intrinsics start with "llvm.x86.". [llvm_v16f32_ty, llvm_v16f32_ty, llvm_v16f32_ty, llvm_i16_ty, llvm_i32_ty], [IntrNoMem]>; + def int_x86_fma_mask_vfmsubadd_ps_256 : GCCBuiltin<"__builtin_ia32_vfmsubaddps256_mask">, + Intrinsic<[llvm_v8f32_ty], + [llvm_v8f32_ty, llvm_v8f32_ty, llvm_v8f32_ty, + llvm_i8_ty], + [IntrNoMem]>; + def int_x86_fma_mask_vfmsubadd_ps_128 : GCCBuiltin<"__builtin_ia32_vfmsubaddps128_mask">, + Intrinsic<[llvm_v4f32_ty], + [llvm_v4f32_ty, llvm_v4f32_ty, llvm_v4f32_ty, + llvm_i8_ty], + [IntrNoMem]>; def int_x86_fma_mask_vfmsubadd_pd_512 : GCCBuiltin<"__builtin_ia32_vfmsubaddpd512_mask">, Intrinsic<[llvm_v8f64_ty], [llvm_v8f64_ty, llvm_v8f64_ty, llvm_v8f64_ty, llvm_i8_ty, llvm_i32_ty], [IntrNoMem]>; + def int_x86_fma_mask_vfmsubadd_pd_256 : GCCBuiltin<"__builtin_ia32_vfmsubaddpd256_mask">, + Intrinsic<[llvm_v4f64_ty], + [llvm_v4f64_ty, llvm_v4f64_ty, llvm_v4f64_ty, + llvm_i8_ty], + [IntrNoMem]>; + def int_x86_fma_mask_vfmsubadd_pd_128 : GCCBuiltin<"__builtin_ia32_vfmsubaddpd128_mask">, + Intrinsic<[llvm_v2f64_ty], + [llvm_v2f64_ty, llvm_v2f64_ty, llvm_v2f64_ty, + llvm_i8_ty], + [IntrNoMem]>; } //===----------------------------------------------------------------------===// @@ -3003,6 +3155,31 @@ let TargetPrefix = "x86" in { // All intrinsics start with "llvm.x86.". // Arithmetic ops let TargetPrefix = "x86" in { // All intrinsics start with "llvm.x86.". + + def int_x86_avx512_mask_add_ps_512 : GCCBuiltin<"__builtin_ia32_addps512_mask">, + Intrinsic<[llvm_v16f32_ty], [llvm_v16f32_ty, llvm_v16f32_ty, + llvm_v16f32_ty, llvm_i16_ty, llvm_i32_ty], [IntrNoMem]>; + def int_x86_avx512_mask_add_pd_512 : GCCBuiltin<"__builtin_ia32_addpd512_mask">, + Intrinsic<[llvm_v8f64_ty], [llvm_v8f64_ty, llvm_v8f64_ty, + llvm_v8f64_ty, llvm_i8_ty, llvm_i32_ty], [IntrNoMem]>; + def int_x86_avx512_mask_sub_ps_512 : GCCBuiltin<"__builtin_ia32_subps512_mask">, + Intrinsic<[llvm_v16f32_ty], [llvm_v16f32_ty, llvm_v16f32_ty, + llvm_v16f32_ty, llvm_i16_ty, llvm_i32_ty], [IntrNoMem]>; + def int_x86_avx512_mask_sub_pd_512 : GCCBuiltin<"__builtin_ia32_subpd512_mask">, + Intrinsic<[llvm_v8f64_ty], [llvm_v8f64_ty, llvm_v8f64_ty, + llvm_v8f64_ty, llvm_i8_ty, llvm_i32_ty], [IntrNoMem]>; + def int_x86_avx512_mask_mul_ps_512 : GCCBuiltin<"__builtin_ia32_mulps512_mask">, + Intrinsic<[llvm_v16f32_ty], [llvm_v16f32_ty, llvm_v16f32_ty, + llvm_v16f32_ty, llvm_i16_ty, llvm_i32_ty], [IntrNoMem]>; + def int_x86_avx512_mask_mul_pd_512 : GCCBuiltin<"__builtin_ia32_mulpd512_mask">, + Intrinsic<[llvm_v8f64_ty], [llvm_v8f64_ty, llvm_v8f64_ty, + llvm_v8f64_ty, llvm_i8_ty, llvm_i32_ty], [IntrNoMem]>; + def int_x86_avx512_mask_div_ps_512 : GCCBuiltin<"__builtin_ia32_divps512_mask">, + Intrinsic<[llvm_v16f32_ty], [llvm_v16f32_ty, llvm_v16f32_ty, + llvm_v16f32_ty, llvm_i16_ty, llvm_i32_ty], [IntrNoMem]>; + def int_x86_avx512_mask_div_pd_512 : GCCBuiltin<"__builtin_ia32_divpd512_mask">, + Intrinsic<[llvm_v8f64_ty], [llvm_v8f64_ty, llvm_v8f64_ty, + llvm_v8f64_ty, llvm_i8_ty, llvm_i32_ty], [IntrNoMem]>; def int_x86_avx512_mask_max_ps_512 : GCCBuiltin<"__builtin_ia32_maxps512_mask">, Intrinsic<[llvm_v16f32_ty], [llvm_v16f32_ty, llvm_v16f32_ty, llvm_v16f32_ty, llvm_i16_ty, llvm_i32_ty], [IntrNoMem]>; @@ -3016,12 +3193,14 @@ let TargetPrefix = "x86" in { // All intrinsics start with "llvm.x86.". Intrinsic<[llvm_v8f64_ty], [llvm_v8f64_ty, llvm_v8f64_ty, llvm_v8f64_ty, llvm_i8_ty, llvm_i32_ty], [IntrNoMem]>; - def int_x86_avx512_rndscale_ss : GCCBuiltin<"__builtin_ia32_rndscaless">, - Intrinsic<[llvm_v4f32_ty], [llvm_v4f32_ty, llvm_v4f32_ty, - llvm_i32_ty], [IntrNoMem]>; - def int_x86_avx512_rndscale_sd : GCCBuiltin<"__builtin_ia32_rndscalesd">, - Intrinsic<[llvm_v2f64_ty], [llvm_v2f64_ty, llvm_v2f64_ty, - llvm_i32_ty], [IntrNoMem]>; + def int_x86_avx512_mask_rndscale_ss : GCCBuiltin<"__builtin_ia32_rndscaless_mask">, + Intrinsic<[llvm_v4f32_ty], [llvm_v4f32_ty, llvm_v4f32_ty, llvm_v4f32_ty, + llvm_i8_ty, llvm_i32_ty, llvm_i32_ty], + [IntrNoMem]>; + def int_x86_avx512_mask_rndscale_sd : GCCBuiltin<"__builtin_ia32_rndscalesd_mask">, + Intrinsic<[llvm_v2f64_ty], [llvm_v2f64_ty, llvm_v2f64_ty, llvm_v2f64_ty, + llvm_i8_ty, llvm_i32_ty, llvm_i32_ty], + [IntrNoMem]>; def int_x86_avx512_sqrt_ss : GCCBuiltin<"__builtin_ia32_sqrtrndss">, Intrinsic<[llvm_v4f32_ty], [llvm_v4f32_ty, llvm_v4f32_ty], [IntrNoMem]>; @@ -3102,22 +3281,6 @@ let TargetPrefix = "x86" in { // All intrinsics start with "llvm.x86.". [IntrNoMem]>; } -// Integer shift ops. -let TargetPrefix = "x86" in { // All intrinsics start with "llvm.x86.". - def int_x86_avx512_psll_dq : GCCBuiltin<"__builtin_ia32_pslldqi512">, - Intrinsic<[llvm_v8i64_ty], [llvm_v8i64_ty, - llvm_i32_ty], [IntrNoMem]>; - def int_x86_avx512_psrl_dq : GCCBuiltin<"__builtin_ia32_psrldqi512">, - Intrinsic<[llvm_v8i64_ty], [llvm_v8i64_ty, - llvm_i32_ty], [IntrNoMem]>; - def int_x86_avx512_psll_dq_bs : GCCBuiltin<"__builtin_ia32_pslldqi512_byteshift">, - Intrinsic<[llvm_v8i64_ty], [llvm_v8i64_ty, - llvm_i32_ty], [IntrNoMem]>; - def int_x86_avx512_psrl_dq_bs : GCCBuiltin<"__builtin_ia32_psrldqi512_byteshift">, - Intrinsic<[llvm_v8i64_ty], [llvm_v8i64_ty, - llvm_i32_ty], [IntrNoMem]>; -} - // Gather and Scatter ops let TargetPrefix = "x86" in { def int_x86_avx512_gather_dpd_512 : GCCBuiltin<"__builtin_ia32_gathersiv8df">, @@ -3226,22 +3389,22 @@ let TargetPrefix = "x86" in { GCCBuiltin<"__builtin_ia32_vpconflictsi_512_mask">, Intrinsic<[llvm_v16i32_ty], [llvm_v16i32_ty, llvm_v16i32_ty, llvm_i16_ty], - []>; + [IntrNoMem]>; def int_x86_avx512_mask_conflict_q_512 : GCCBuiltin<"__builtin_ia32_vpconflictdi_512_mask">, Intrinsic<[llvm_v8i64_ty], [llvm_v8i64_ty, llvm_v8i64_ty, llvm_i8_ty], - []>; + [IntrNoMem]>; def int_x86_avx512_mask_lzcnt_d_512 : GCCBuiltin<"__builtin_ia32_vplzcntd_512_mask">, Intrinsic<[llvm_v16i32_ty], [llvm_v16i32_ty, llvm_v16i32_ty, llvm_i16_ty], - []>; + [IntrNoMem]>; def int_x86_avx512_mask_lzcnt_q_512 : GCCBuiltin<"__builtin_ia32_vplzcntq_512_mask">, Intrinsic<[llvm_v8i64_ty], [llvm_v8i64_ty, llvm_v8i64_ty, llvm_i8_ty], - []>; + [IntrNoMem]>; } // Vector blend @@ -3250,10 +3413,26 @@ let TargetPrefix = "x86" in { // All intrinsics start with "llvm.x86.". Intrinsic<[llvm_v16f32_ty], [llvm_v16f32_ty, llvm_v16f32_ty, llvm_i16_ty], [IntrNoMem]>; + def int_x86_avx512_mask_blend_ps_256 : GCCBuiltin<"__builtin_ia32_blendmps_256_mask">, + Intrinsic<[llvm_v8f32_ty], + [llvm_v8f32_ty, llvm_v8f32_ty, llvm_i8_ty], + [IntrNoMem]>; + def int_x86_avx512_mask_blend_ps_128 : GCCBuiltin<"__builtin_ia32_blendmps_128_mask">, + Intrinsic<[llvm_v4f32_ty], + [llvm_v4f32_ty, llvm_v4f32_ty, llvm_i8_ty], + [IntrNoMem]>; def int_x86_avx512_mask_blend_pd_512 : GCCBuiltin<"__builtin_ia32_blendmpd_512_mask">, Intrinsic<[llvm_v8f64_ty], [llvm_v8f64_ty, llvm_v8f64_ty, llvm_i8_ty], [IntrNoMem]>; + def int_x86_avx512_mask_blend_pd_256 : GCCBuiltin<"__builtin_ia32_blendmpd_256_mask">, + Intrinsic<[llvm_v4f64_ty], + [llvm_v4f64_ty, llvm_v4f64_ty, llvm_i8_ty], + [IntrNoMem]>; + def int_x86_avx512_mask_blend_pd_128 : GCCBuiltin<"__builtin_ia32_blendmpd_128_mask">, + Intrinsic<[llvm_v2f64_ty], + [llvm_v2f64_ty, llvm_v2f64_ty, llvm_i8_ty], + [IntrNoMem]>; def int_x86_avx512_mask_blend_d_512 : GCCBuiltin<"__builtin_ia32_blendmd_512_mask">, Intrinsic<[llvm_v16i32_ty], @@ -3263,6 +3442,48 @@ let TargetPrefix = "x86" in { // All intrinsics start with "llvm.x86.". Intrinsic<[llvm_v8i64_ty], [llvm_v8i64_ty, llvm_v8i64_ty, llvm_i8_ty], [IntrNoMem]>; + def int_x86_avx512_mask_blend_d_256 : GCCBuiltin<"__builtin_ia32_blendmd_256_mask">, + Intrinsic<[llvm_v8i32_ty], + [llvm_v8i32_ty, llvm_v8i32_ty, llvm_i8_ty], + [IntrNoMem]>; + def int_x86_avx512_mask_blend_q_256 : GCCBuiltin<"__builtin_ia32_blendmq_256_mask">, + Intrinsic<[llvm_v4i64_ty], + [llvm_v4i64_ty, llvm_v4i64_ty, llvm_i8_ty], + [IntrNoMem]>; + def int_x86_avx512_mask_blend_d_128 : GCCBuiltin<"__builtin_ia32_blendmd_128_mask">, + Intrinsic<[llvm_v4i32_ty], + [llvm_v4i32_ty, llvm_v4i32_ty, llvm_i8_ty], + [IntrNoMem]>; + def int_x86_avx512_mask_blend_q_128 : GCCBuiltin<"__builtin_ia32_blendmq_128_mask">, + Intrinsic<[llvm_v2i64_ty], + [llvm_v2i64_ty, llvm_v2i64_ty, llvm_i8_ty], + [IntrNoMem]>; + + def int_x86_avx512_mask_blend_w_512 : GCCBuiltin<"__builtin_ia32_blendmw_512_mask">, + Intrinsic<[llvm_v32i16_ty], + [llvm_v32i16_ty, llvm_v32i16_ty, llvm_i32_ty], + [IntrNoMem]>; + def int_x86_avx512_mask_blend_w_256 : GCCBuiltin<"__builtin_ia32_blendmw_256_mask">, + Intrinsic<[llvm_v16i16_ty], + [llvm_v16i16_ty, llvm_v16i16_ty, llvm_i16_ty], + [IntrNoMem]>; + def int_x86_avx512_mask_blend_w_128 : GCCBuiltin<"__builtin_ia32_blendmw_128_mask">, + Intrinsic<[llvm_v8i16_ty], + [llvm_v8i16_ty, llvm_v8i16_ty, llvm_i8_ty], + [IntrNoMem]>; + def int_x86_avx512_mask_blend_b_512 : GCCBuiltin<"__builtin_ia32_blendmb_512_mask">, + Intrinsic<[llvm_v64i8_ty], + [llvm_v64i8_ty, llvm_v64i8_ty, llvm_i64_ty], + [IntrNoMem]>; + def int_x86_avx512_mask_blend_b_256 : GCCBuiltin<"__builtin_ia32_blendmb_256_mask">, + Intrinsic<[llvm_v32i8_ty], + [llvm_v32i8_ty, llvm_v32i8_ty, llvm_i32_ty], + [IntrNoMem]>; + def int_x86_avx512_mask_blend_b_128 : GCCBuiltin<"__builtin_ia32_blendmb_128_mask">, + Intrinsic<[llvm_v16i8_ty], + [llvm_v16i8_ty, llvm_v16i8_ty, llvm_i16_ty], + [IntrNoMem]>; + } let TargetPrefix = "x86" in { @@ -3307,29 +3528,29 @@ let TargetPrefix = "x86" in { [IntrNoMem]>; def int_x86_avx512_mask_cmp_b_512: GCCBuiltin<"__builtin_ia32_cmpb512_mask">, - Intrinsic<[llvm_i64_ty], [llvm_v64i8_ty, llvm_v64i8_ty, llvm_i32_ty, + Intrinsic<[llvm_i64_ty], [llvm_v64i8_ty, llvm_v64i8_ty, llvm_i8_ty, llvm_i64_ty], [IntrNoMem]>; def int_x86_avx512_mask_cmp_w_512: GCCBuiltin<"__builtin_ia32_cmpw512_mask">, - Intrinsic<[llvm_i32_ty], [llvm_v32i16_ty, llvm_v32i16_ty, llvm_i32_ty, + Intrinsic<[llvm_i32_ty], [llvm_v32i16_ty, llvm_v32i16_ty, llvm_i8_ty, llvm_i32_ty], [IntrNoMem]>; def int_x86_avx512_mask_cmp_d_512: GCCBuiltin<"__builtin_ia32_cmpd512_mask">, - Intrinsic<[llvm_i16_ty], [llvm_v16i32_ty, llvm_v16i32_ty, llvm_i32_ty, + Intrinsic<[llvm_i16_ty], [llvm_v16i32_ty, llvm_v16i32_ty, llvm_i8_ty, llvm_i16_ty], [IntrNoMem ]>; def int_x86_avx512_mask_cmp_q_512: GCCBuiltin<"__builtin_ia32_cmpq512_mask">, - Intrinsic<[llvm_i8_ty], [llvm_v8i64_ty, llvm_v8i64_ty, llvm_i32_ty, + Intrinsic<[llvm_i8_ty], [llvm_v8i64_ty, llvm_v8i64_ty, llvm_i8_ty, llvm_i8_ty], [IntrNoMem]>; def int_x86_avx512_mask_ucmp_b_512: GCCBuiltin<"__builtin_ia32_ucmpb512_mask">, - Intrinsic<[llvm_i64_ty], [llvm_v64i8_ty, llvm_v64i8_ty, llvm_i32_ty, + Intrinsic<[llvm_i64_ty], [llvm_v64i8_ty, llvm_v64i8_ty, llvm_i8_ty, llvm_i64_ty], [IntrNoMem]>; def int_x86_avx512_mask_ucmp_w_512: GCCBuiltin<"__builtin_ia32_ucmpw512_mask">, - Intrinsic<[llvm_i32_ty], [llvm_v32i16_ty, llvm_v32i16_ty, llvm_i32_ty, + Intrinsic<[llvm_i32_ty], [llvm_v32i16_ty, llvm_v32i16_ty, llvm_i8_ty, llvm_i32_ty], [IntrNoMem]>; def int_x86_avx512_mask_ucmp_d_512: GCCBuiltin<"__builtin_ia32_ucmpd512_mask">, - Intrinsic<[llvm_i16_ty], [llvm_v16i32_ty, llvm_v16i32_ty, llvm_i32_ty, + Intrinsic<[llvm_i16_ty], [llvm_v16i32_ty, llvm_v16i32_ty, llvm_i8_ty, llvm_i16_ty], [IntrNoMem]>; def int_x86_avx512_mask_ucmp_q_512: GCCBuiltin<"__builtin_ia32_ucmpq512_mask">, - Intrinsic<[llvm_i8_ty], [llvm_v8i64_ty, llvm_v8i64_ty, llvm_i32_ty, + Intrinsic<[llvm_i8_ty], [llvm_v8i64_ty, llvm_v8i64_ty, llvm_i8_ty, llvm_i8_ty], [IntrNoMem]>; // 256-bit @@ -3360,29 +3581,29 @@ let TargetPrefix = "x86" in { [IntrNoMem]>; def int_x86_avx512_mask_cmp_b_256: GCCBuiltin<"__builtin_ia32_cmpb256_mask">, - Intrinsic<[llvm_i32_ty], [llvm_v32i8_ty, llvm_v32i8_ty, llvm_i32_ty, + Intrinsic<[llvm_i32_ty], [llvm_v32i8_ty, llvm_v32i8_ty, llvm_i8_ty, llvm_i32_ty], [IntrNoMem]>; def int_x86_avx512_mask_cmp_w_256: GCCBuiltin<"__builtin_ia32_cmpw256_mask">, - Intrinsic<[llvm_i16_ty], [llvm_v16i16_ty, llvm_v16i16_ty, llvm_i32_ty, + Intrinsic<[llvm_i16_ty], [llvm_v16i16_ty, llvm_v16i16_ty, llvm_i8_ty, llvm_i16_ty], [IntrNoMem]>; def int_x86_avx512_mask_cmp_d_256: GCCBuiltin<"__builtin_ia32_cmpd256_mask">, - Intrinsic<[llvm_i8_ty], [llvm_v8i32_ty, llvm_v8i32_ty, llvm_i32_ty, + Intrinsic<[llvm_i8_ty], [llvm_v8i32_ty, llvm_v8i32_ty, llvm_i8_ty, llvm_i8_ty], [IntrNoMem]>; def int_x86_avx512_mask_cmp_q_256: GCCBuiltin<"__builtin_ia32_cmpq256_mask">, - Intrinsic<[llvm_i8_ty], [llvm_v4i64_ty, llvm_v4i64_ty, llvm_i32_ty, + Intrinsic<[llvm_i8_ty], [llvm_v4i64_ty, llvm_v4i64_ty, llvm_i8_ty, llvm_i8_ty], [IntrNoMem]>; def int_x86_avx512_mask_ucmp_b_256: GCCBuiltin<"__builtin_ia32_ucmpb256_mask">, - Intrinsic<[llvm_i32_ty], [llvm_v32i8_ty, llvm_v32i8_ty, llvm_i32_ty, + Intrinsic<[llvm_i32_ty], [llvm_v32i8_ty, llvm_v32i8_ty, llvm_i8_ty, llvm_i32_ty], [IntrNoMem]>; def int_x86_avx512_mask_ucmp_w_256: GCCBuiltin<"__builtin_ia32_ucmpw256_mask">, - Intrinsic<[llvm_i16_ty], [llvm_v16i16_ty, llvm_v16i16_ty, llvm_i32_ty, + Intrinsic<[llvm_i16_ty], [llvm_v16i16_ty, llvm_v16i16_ty, llvm_i8_ty, llvm_i16_ty], [IntrNoMem]>; def int_x86_avx512_mask_ucmp_d_256: GCCBuiltin<"__builtin_ia32_ucmpd256_mask">, - Intrinsic<[llvm_i8_ty], [llvm_v8i32_ty, llvm_v8i32_ty, llvm_i32_ty, + Intrinsic<[llvm_i8_ty], [llvm_v8i32_ty, llvm_v8i32_ty, llvm_i8_ty, llvm_i8_ty], [IntrNoMem]>; def int_x86_avx512_mask_ucmp_q_256: GCCBuiltin<"__builtin_ia32_ucmpq256_mask">, - Intrinsic<[llvm_i8_ty], [llvm_v4i64_ty, llvm_v4i64_ty, llvm_i32_ty, + Intrinsic<[llvm_i8_ty], [llvm_v4i64_ty, llvm_v4i64_ty, llvm_i8_ty, llvm_i8_ty], [IntrNoMem]>; // 128-bit @@ -3413,39 +3634,243 @@ let TargetPrefix = "x86" in { [IntrNoMem]>; def int_x86_avx512_mask_cmp_b_128: GCCBuiltin<"__builtin_ia32_cmpb128_mask">, - Intrinsic<[llvm_i16_ty], [llvm_v16i8_ty, llvm_v16i8_ty, llvm_i32_ty, + Intrinsic<[llvm_i16_ty], [llvm_v16i8_ty, llvm_v16i8_ty, llvm_i8_ty, llvm_i16_ty], [IntrNoMem]>; def int_x86_avx512_mask_cmp_w_128: GCCBuiltin<"__builtin_ia32_cmpw128_mask">, - Intrinsic<[llvm_i8_ty], [llvm_v8i16_ty, llvm_v8i16_ty, llvm_i32_ty, + Intrinsic<[llvm_i8_ty], [llvm_v8i16_ty, llvm_v8i16_ty, llvm_i8_ty, llvm_i8_ty], [IntrNoMem]>; def int_x86_avx512_mask_cmp_d_128: GCCBuiltin<"__builtin_ia32_cmpd128_mask">, - Intrinsic<[llvm_i8_ty], [llvm_v4i32_ty, llvm_v4i32_ty, llvm_i32_ty, + Intrinsic<[llvm_i8_ty], [llvm_v4i32_ty, llvm_v4i32_ty, llvm_i8_ty, llvm_i8_ty], [IntrNoMem]>; def int_x86_avx512_mask_cmp_q_128: GCCBuiltin<"__builtin_ia32_cmpq128_mask">, - Intrinsic<[llvm_i8_ty], [llvm_v2i64_ty, llvm_v2i64_ty, llvm_i32_ty, + Intrinsic<[llvm_i8_ty], [llvm_v2i64_ty, llvm_v2i64_ty, llvm_i8_ty, llvm_i8_ty], [IntrNoMem]>; def int_x86_avx512_mask_ucmp_b_128: GCCBuiltin<"__builtin_ia32_ucmpb128_mask">, - Intrinsic<[llvm_i16_ty], [llvm_v16i8_ty, llvm_v16i8_ty, llvm_i32_ty, + Intrinsic<[llvm_i16_ty], [llvm_v16i8_ty, llvm_v16i8_ty, llvm_i8_ty, llvm_i16_ty], [IntrNoMem]>; def int_x86_avx512_mask_ucmp_w_128: GCCBuiltin<"__builtin_ia32_ucmpw128_mask">, - Intrinsic<[llvm_i8_ty], [llvm_v8i16_ty, llvm_v8i16_ty, llvm_i32_ty, + Intrinsic<[llvm_i8_ty], [llvm_v8i16_ty, llvm_v8i16_ty, llvm_i8_ty, llvm_i8_ty], [IntrNoMem]>; def int_x86_avx512_mask_ucmp_d_128: GCCBuiltin<"__builtin_ia32_ucmpd128_mask">, - Intrinsic<[llvm_i8_ty], [llvm_v4i32_ty, llvm_v4i32_ty, llvm_i32_ty, + Intrinsic<[llvm_i8_ty], [llvm_v4i32_ty, llvm_v4i32_ty, llvm_i8_ty, llvm_i8_ty], [IntrNoMem]>; def int_x86_avx512_mask_ucmp_q_128: GCCBuiltin<"__builtin_ia32_ucmpq128_mask">, - Intrinsic<[llvm_i8_ty], [llvm_v2i64_ty, llvm_v2i64_ty, llvm_i32_ty, + Intrinsic<[llvm_i8_ty], [llvm_v2i64_ty, llvm_v2i64_ty, llvm_i8_ty, llvm_i8_ty], [IntrNoMem]>; } +// Compress, Expand +let TargetPrefix = "x86" in { + def int_x86_avx512_mask_compress_ps_512 : + GCCBuiltin<"__builtin_ia32_compresssf512_mask">, + Intrinsic<[llvm_v16f32_ty], [llvm_v16f32_ty, llvm_v16f32_ty, + llvm_i16_ty], [IntrNoMem]>; + def int_x86_avx512_mask_compress_pd_512 : + GCCBuiltin<"__builtin_ia32_compressdf512_mask">, + Intrinsic<[llvm_v8f64_ty], [llvm_v8f64_ty, llvm_v8f64_ty, + llvm_i8_ty], [IntrNoMem]>; + def int_x86_avx512_mask_compress_ps_256 : + GCCBuiltin<"__builtin_ia32_compresssf256_mask">, + Intrinsic<[llvm_v8f32_ty], [llvm_v8f32_ty, llvm_v8f32_ty, + llvm_i8_ty], [IntrNoMem]>; + def int_x86_avx512_mask_compress_pd_256 : + GCCBuiltin<"__builtin_ia32_compressdf256_mask">, + Intrinsic<[llvm_v4f64_ty], [llvm_v4f64_ty, llvm_v4f64_ty, + llvm_i8_ty], [IntrNoMem]>; + def int_x86_avx512_mask_compress_ps_128 : + GCCBuiltin<"__builtin_ia32_compresssf128_mask">, + Intrinsic<[llvm_v4f32_ty], [llvm_v4f32_ty, llvm_v4f32_ty, + llvm_i8_ty], [IntrNoMem]>; + def int_x86_avx512_mask_compress_pd_128 : + GCCBuiltin<"__builtin_ia32_compressdf128_mask">, + Intrinsic<[llvm_v2f64_ty], [llvm_v2f64_ty, llvm_v2f64_ty, + llvm_i8_ty], [IntrNoMem]>; + + def int_x86_avx512_mask_compress_store_ps_512 : + GCCBuiltin<"__builtin_ia32_compressstoresf512_mask">, + Intrinsic<[], [llvm_ptr_ty, llvm_v16f32_ty, + llvm_i16_ty], [IntrReadWriteArgMem]>; + def int_x86_avx512_mask_compress_store_pd_512 : + GCCBuiltin<"__builtin_ia32_compressstoredf512_mask">, + Intrinsic<[], [llvm_ptr_ty, llvm_v8f64_ty, + llvm_i8_ty], [IntrReadWriteArgMem]>; + def int_x86_avx512_mask_compress_store_ps_256 : + GCCBuiltin<"__builtin_ia32_compressstoresf256_mask">, + Intrinsic<[], [llvm_ptr_ty, llvm_v8f32_ty, + llvm_i8_ty], [IntrReadWriteArgMem]>; + def int_x86_avx512_mask_compress_store_pd_256 : + GCCBuiltin<"__builtin_ia32_compressstoredf256_mask">, + Intrinsic<[], [llvm_ptr_ty, llvm_v4f64_ty, + llvm_i8_ty], [IntrReadWriteArgMem]>; + def int_x86_avx512_mask_compress_store_ps_128 : + GCCBuiltin<"__builtin_ia32_compressstoresf128_mask">, + Intrinsic<[], [llvm_ptr_ty, llvm_v4f32_ty, + llvm_i8_ty], [IntrReadWriteArgMem]>; + def int_x86_avx512_mask_compress_store_pd_128 : + GCCBuiltin<"__builtin_ia32_compressstoredf128_mask">, + Intrinsic<[], [llvm_ptr_ty, llvm_v2f64_ty, + llvm_i8_ty], [IntrReadWriteArgMem]>; + + def int_x86_avx512_mask_compress_d_512 : + GCCBuiltin<"__builtin_ia32_compresssi512_mask">, + Intrinsic<[llvm_v16i32_ty], [llvm_v16i32_ty, llvm_v16i32_ty, + llvm_i16_ty], [IntrNoMem]>; + def int_x86_avx512_mask_compress_q_512 : + GCCBuiltin<"__builtin_ia32_compressdi512_mask">, + Intrinsic<[llvm_v8i64_ty], [llvm_v8i64_ty, llvm_v8i64_ty, + llvm_i8_ty], [IntrNoMem]>; + def int_x86_avx512_mask_compress_d_256 : + GCCBuiltin<"__builtin_ia32_compresssi256_mask">, + Intrinsic<[llvm_v8i32_ty], [llvm_v8i32_ty, llvm_v8i32_ty, + llvm_i8_ty], [IntrNoMem]>; + def int_x86_avx512_mask_compress_q_256 : + GCCBuiltin<"__builtin_ia32_compressdi256_mask">, + Intrinsic<[llvm_v4i64_ty], [llvm_v4i64_ty, llvm_v4i64_ty, + llvm_i8_ty], [IntrNoMem]>; + def int_x86_avx512_mask_compress_d_128 : + GCCBuiltin<"__builtin_ia32_compresssi128_mask">, + Intrinsic<[llvm_v4i32_ty], [llvm_v4i32_ty, llvm_v4i32_ty, + llvm_i8_ty], [IntrNoMem]>; + def int_x86_avx512_mask_compress_q_128 : + GCCBuiltin<"__builtin_ia32_compressdi128_mask">, + Intrinsic<[llvm_v2i64_ty], [llvm_v2i64_ty, llvm_v2i64_ty, + llvm_i8_ty], [IntrNoMem]>; + + def int_x86_avx512_mask_compress_store_d_512 : + GCCBuiltin<"__builtin_ia32_compressstoresi512_mask">, + Intrinsic<[], [llvm_ptr_ty, llvm_v16i32_ty, + llvm_i16_ty], [IntrReadWriteArgMem]>; + def int_x86_avx512_mask_compress_store_q_512 : + GCCBuiltin<"__builtin_ia32_compressstoredi512_mask">, + Intrinsic<[], [llvm_ptr_ty, llvm_v8i64_ty, + llvm_i8_ty], [IntrReadWriteArgMem]>; + def int_x86_avx512_mask_compress_store_d_256 : + GCCBuiltin<"__builtin_ia32_compressstoresi256_mask">, + Intrinsic<[], [llvm_ptr_ty, llvm_v8i32_ty, + llvm_i8_ty], [IntrReadWriteArgMem]>; + def int_x86_avx512_mask_compress_store_q_256 : + GCCBuiltin<"__builtin_ia32_compressstoredi256_mask">, + Intrinsic<[], [llvm_ptr_ty, llvm_v4i64_ty, + llvm_i8_ty], [IntrReadWriteArgMem]>; + def int_x86_avx512_mask_compress_store_d_128 : + GCCBuiltin<"__builtin_ia32_compressstoresi128_mask">, + Intrinsic<[], [llvm_ptr_ty, llvm_v4i32_ty, + llvm_i8_ty], [IntrReadWriteArgMem]>; + def int_x86_avx512_mask_compress_store_q_128 : + GCCBuiltin<"__builtin_ia32_compressstoredi128_mask">, + Intrinsic<[], [llvm_ptr_ty, llvm_v2i64_ty, + llvm_i8_ty], [IntrReadWriteArgMem]>; + +// expand + def int_x86_avx512_mask_expand_ps_512 : + GCCBuiltin<"__builtin_ia32_expandsf512_mask">, + Intrinsic<[llvm_v16f32_ty], [llvm_v16f32_ty, llvm_v16f32_ty, + llvm_i16_ty], [IntrNoMem]>; + def int_x86_avx512_mask_expand_pd_512 : + GCCBuiltin<"__builtin_ia32_expanddf512_mask">, + Intrinsic<[llvm_v8f64_ty], [llvm_v8f64_ty, llvm_v8f64_ty, + llvm_i8_ty], [IntrNoMem]>; + def int_x86_avx512_mask_expand_ps_256 : + GCCBuiltin<"__builtin_ia32_expandsf256_mask">, + Intrinsic<[llvm_v8f32_ty], [llvm_v8f32_ty, llvm_v8f32_ty, + llvm_i8_ty], [IntrNoMem]>; + def int_x86_avx512_mask_expand_pd_256 : + GCCBuiltin<"__builtin_ia32_expanddf256_mask">, + Intrinsic<[llvm_v4f64_ty], [llvm_v4f64_ty, llvm_v4f64_ty, + llvm_i8_ty], [IntrNoMem]>; + def int_x86_avx512_mask_expand_ps_128 : + GCCBuiltin<"__builtin_ia32_expandsf128_mask">, + Intrinsic<[llvm_v4f32_ty], [llvm_v4f32_ty, llvm_v4f32_ty, + llvm_i8_ty], [IntrNoMem]>; + def int_x86_avx512_mask_expand_pd_128 : + GCCBuiltin<"__builtin_ia32_expanddf128_mask">, + Intrinsic<[llvm_v2f64_ty], [llvm_v2f64_ty, llvm_v2f64_ty, + llvm_i8_ty], [IntrNoMem]>; + + def int_x86_avx512_mask_expand_load_ps_512 : + GCCBuiltin<"__builtin_ia32_expandloadsf512_mask">, + Intrinsic<[llvm_v16f32_ty], [llvm_ptr_ty, llvm_v16f32_ty, + llvm_i16_ty], [IntrReadArgMem]>; + def int_x86_avx512_mask_expand_load_pd_512 : + GCCBuiltin<"__builtin_ia32_expandloaddf512_mask">, + Intrinsic<[llvm_v8f64_ty], [llvm_ptr_ty, llvm_v8f64_ty, + llvm_i8_ty], [IntrReadArgMem]>; + def int_x86_avx512_mask_expand_load_ps_256 : + GCCBuiltin<"__builtin_ia32_expandloadsf256_mask">, + Intrinsic<[llvm_v8f32_ty], [llvm_ptr_ty, llvm_v8f32_ty, + llvm_i8_ty], [IntrReadArgMem]>; + def int_x86_avx512_mask_expand_load_pd_256 : + GCCBuiltin<"__builtin_ia32_expandloaddf256_mask">, + Intrinsic<[llvm_v4f64_ty], [llvm_ptr_ty, llvm_v4f64_ty, + llvm_i8_ty], [IntrReadArgMem]>; + def int_x86_avx512_mask_expand_load_ps_128 : + GCCBuiltin<"__builtin_ia32_expandloadsf128_mask">, + Intrinsic<[llvm_v4f32_ty], [llvm_ptr_ty, llvm_v4f32_ty, + llvm_i8_ty], [IntrReadArgMem]>; + def int_x86_avx512_mask_expand_load_pd_128 : + GCCBuiltin<"__builtin_ia32_expandloaddf128_mask">, + Intrinsic<[llvm_v2f64_ty], [llvm_ptr_ty, llvm_v2f64_ty, + llvm_i8_ty], [IntrReadArgMem]>; + + def int_x86_avx512_mask_expand_d_512 : + GCCBuiltin<"__builtin_ia32_expandsi512_mask">, + Intrinsic<[llvm_v16i32_ty], [llvm_v16i32_ty, llvm_v16i32_ty, + llvm_i16_ty], [IntrNoMem]>; + def int_x86_avx512_mask_expand_q_512 : + GCCBuiltin<"__builtin_ia32_expanddi512_mask">, + Intrinsic<[llvm_v8i64_ty], [llvm_v8i64_ty, llvm_v8i64_ty, + llvm_i8_ty], [IntrNoMem]>; + def int_x86_avx512_mask_expand_d_256 : + GCCBuiltin<"__builtin_ia32_expandsi256_mask">, + Intrinsic<[llvm_v8i32_ty], [llvm_v8i32_ty, llvm_v8i32_ty, + llvm_i8_ty], [IntrNoMem]>; + def int_x86_avx512_mask_expand_q_256 : + GCCBuiltin<"__builtin_ia32_expanddi256_mask">, + Intrinsic<[llvm_v4i64_ty], [llvm_v4i64_ty, llvm_v4i64_ty, + llvm_i8_ty], [IntrNoMem]>; + def int_x86_avx512_mask_expand_d_128 : + GCCBuiltin<"__builtin_ia32_expandsi128_mask">, + Intrinsic<[llvm_v4i32_ty], [llvm_v4i32_ty, llvm_v4i32_ty, + llvm_i8_ty], [IntrNoMem]>; + def int_x86_avx512_mask_expand_q_128 : + GCCBuiltin<"__builtin_ia32_expanddi128_mask">, + Intrinsic<[llvm_v2i64_ty], [llvm_v2i64_ty, llvm_v2i64_ty, + llvm_i8_ty], [IntrNoMem]>; + + def int_x86_avx512_mask_expand_load_d_512 : + GCCBuiltin<"__builtin_ia32_expandloadsi512_mask">, + Intrinsic<[llvm_v16i32_ty], [llvm_ptr_ty, llvm_v16i32_ty, + llvm_i16_ty], [IntrReadArgMem]>; + def int_x86_avx512_mask_expand_load_q_512 : + GCCBuiltin<"__builtin_ia32_expandloaddi512_mask">, + Intrinsic<[llvm_v8i64_ty], [llvm_ptr_ty, llvm_v8i64_ty, + llvm_i8_ty], [IntrReadArgMem]>; + def int_x86_avx512_mask_expand_load_d_256 : + GCCBuiltin<"__builtin_ia32_expandloadsi256_mask">, + Intrinsic<[llvm_v8i32_ty], [llvm_ptr_ty, llvm_v8i32_ty, + llvm_i8_ty], [IntrReadArgMem]>; + def int_x86_avx512_mask_expand_load_q_256 : + GCCBuiltin<"__builtin_ia32_expandloaddi256_mask">, + Intrinsic<[llvm_v4i64_ty], [llvm_ptr_ty, llvm_v4i64_ty, + llvm_i8_ty], [IntrReadArgMem]>; + def int_x86_avx512_mask_expand_load_d_128 : + GCCBuiltin<"__builtin_ia32_expandloadsi128_mask">, + Intrinsic<[llvm_v4i32_ty], [llvm_ptr_ty, llvm_v4i32_ty, + llvm_i8_ty], [IntrReadArgMem]>; + def int_x86_avx512_mask_expand_load_q_128 : + GCCBuiltin<"__builtin_ia32_expandloaddi128_mask">, + Intrinsic<[llvm_v2i64_ty], [llvm_ptr_ty, llvm_v2i64_ty, + llvm_i8_ty], [IntrReadArgMem]>; + +} // Misc. let TargetPrefix = "x86" in { def int_x86_avx512_mask_cmp_ps_512 : GCCBuiltin<"__builtin_ia32_cmpps512_mask">, - Intrinsic<[llvm_i16_ty], [llvm_v16f32_ty, llvm_v16f32_ty, llvm_i32_ty, + Intrinsic<[llvm_i16_ty], [llvm_v16f32_ty, llvm_v16f32_ty, llvm_i8_ty, llvm_i16_ty, llvm_i32_ty], [IntrNoMem]>; def int_x86_avx512_mask_cmp_pd_512 : GCCBuiltin<"__builtin_ia32_cmppd512_mask">, - Intrinsic<[llvm_i8_ty], [llvm_v8f64_ty, llvm_v8f64_ty, llvm_i32_ty, + Intrinsic<[llvm_i8_ty], [llvm_v8f64_ty, llvm_v8f64_ty, llvm_i8_ty, llvm_i8_ty, llvm_i32_ty], [IntrNoMem]>; def int_x86_avx512_mask_pand_d_512 : GCCBuiltin<"__builtin_ia32_pandd512_mask">, Intrinsic<[llvm_v16i32_ty], [llvm_v16i32_ty, llvm_v16i32_ty, |