mirror of
https://github.com/RPCS3/llvm-mirror.git
synced 2024-11-24 11:42:57 +01:00
e521de4e60
a temporary workaround for the 2-wide vector_shuffle problem (i.e. its mask would have type v2i32 which is not legal). llvm-svn: 27964
538 lines
27 KiB
TableGen
538 lines
27 KiB
TableGen
//===- IntrinsicsX86.td - Defines X86 intrinsics -----------*- tablegen -*-===//
|
|
//
|
|
// The LLVM Compiler Infrastructure
|
|
//
|
|
// This file was developed by Chris Lattner and is distributed under the
|
|
// University of Illinois Open Source License. See LICENSE.TXT for details.
|
|
//
|
|
//===----------------------------------------------------------------------===//
|
|
//
|
|
// This file defines all of the X86-specific intrinsics.
|
|
//
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
// SSE1
|
|
|
|
// Arithmetic ops
|
|
let TargetPrefix = "x86" in { // All intrinsics start with "llvm.x86.".
|
|
def int_x86_sse_add_ss : GCCBuiltin<"__builtin_ia32_addss">,
|
|
Intrinsic<[llvm_v4f32_ty, llvm_v4f32_ty,
|
|
llvm_v4f32_ty], [IntrNoMem]>;
|
|
def int_x86_sse_sub_ss : GCCBuiltin<"__builtin_ia32_subss">,
|
|
Intrinsic<[llvm_v4f32_ty, llvm_v4f32_ty,
|
|
llvm_v4f32_ty], [IntrNoMem]>;
|
|
def int_x86_sse_mul_ss : GCCBuiltin<"__builtin_ia32_mulss">,
|
|
Intrinsic<[llvm_v4f32_ty, llvm_v4f32_ty,
|
|
llvm_v4f32_ty], [IntrNoMem]>;
|
|
def int_x86_sse_div_ss : GCCBuiltin<"__builtin_ia32_divss">,
|
|
Intrinsic<[llvm_v4f32_ty, llvm_v4f32_ty,
|
|
llvm_v4f32_ty], [IntrNoMem]>;
|
|
def int_x86_sse_sqrt_ss : GCCBuiltin<"__builtin_ia32_sqrtss">,
|
|
Intrinsic<[llvm_v4f32_ty, llvm_v4f32_ty],
|
|
[IntrNoMem]>;
|
|
def int_x86_sse_sqrt_ps : GCCBuiltin<"__builtin_ia32_sqrtps">,
|
|
Intrinsic<[llvm_v4f32_ty, llvm_v4f32_ty],
|
|
[IntrNoMem]>;
|
|
def int_x86_sse_rcp_ss : GCCBuiltin<"__builtin_ia32_rcpss">,
|
|
Intrinsic<[llvm_v4f32_ty, llvm_v4f32_ty],
|
|
[IntrNoMem]>;
|
|
def int_x86_sse_rcp_ps : GCCBuiltin<"__builtin_ia32_rcpps">,
|
|
Intrinsic<[llvm_v4f32_ty, llvm_v4f32_ty],
|
|
[IntrNoMem]>;
|
|
def int_x86_sse_rsqrt_ss : GCCBuiltin<"__builtin_ia32_rsqrtss">,
|
|
Intrinsic<[llvm_v4f32_ty, llvm_v4f32_ty],
|
|
[IntrNoMem]>;
|
|
def int_x86_sse_rsqrt_ps : GCCBuiltin<"__builtin_ia32_rsqrtps">,
|
|
Intrinsic<[llvm_v4f32_ty, llvm_v4f32_ty],
|
|
[IntrNoMem]>;
|
|
def int_x86_sse_min_ss : GCCBuiltin<"__builtin_ia32_minss">,
|
|
Intrinsic<[llvm_v4f32_ty, llvm_v4f32_ty,
|
|
llvm_v4f32_ty], [IntrNoMem]>;
|
|
def int_x86_sse_min_ps : GCCBuiltin<"__builtin_ia32_minps">,
|
|
Intrinsic<[llvm_v4f32_ty, llvm_v4f32_ty,
|
|
llvm_v4f32_ty], [IntrNoMem]>;
|
|
def int_x86_sse_max_ss : GCCBuiltin<"__builtin_ia32_maxss">,
|
|
Intrinsic<[llvm_v4f32_ty, llvm_v4f32_ty,
|
|
llvm_v4f32_ty], [IntrNoMem]>;
|
|
def int_x86_sse_max_ps : GCCBuiltin<"__builtin_ia32_maxps">,
|
|
Intrinsic<[llvm_v4f32_ty, llvm_v4f32_ty,
|
|
llvm_v4f32_ty], [IntrNoMem]>;
|
|
}
|
|
|
|
// Comparison ops
|
|
let TargetPrefix = "x86" in { // All intrinsics start with "llvm.x86.".
|
|
def int_x86_sse_cmp_ss :
|
|
Intrinsic<[llvm_v4f32_ty, llvm_v4f32_ty,
|
|
llvm_v4f32_ty, llvm_sbyte_ty], [IntrNoMem]>;
|
|
def int_x86_sse_cmp_ps :
|
|
Intrinsic<[llvm_v4f32_ty, llvm_v4f32_ty,
|
|
llvm_v4f32_ty, llvm_sbyte_ty], [IntrNoMem]>;
|
|
def int_x86_sse_comieq_ss : GCCBuiltin<"__builtin_ia32_comieq">,
|
|
Intrinsic<[llvm_int_ty, llvm_v4f32_ty,
|
|
llvm_v4f32_ty], [IntrNoMem]>;
|
|
def int_x86_sse_comilt_ss : GCCBuiltin<"__builtin_ia32_comilt">,
|
|
Intrinsic<[llvm_int_ty, llvm_v4f32_ty,
|
|
llvm_v4f32_ty], [IntrNoMem]>;
|
|
def int_x86_sse_comile_ss : GCCBuiltin<"__builtin_ia32_comile">,
|
|
Intrinsic<[llvm_int_ty, llvm_v4f32_ty,
|
|
llvm_v4f32_ty], [IntrNoMem]>;
|
|
def int_x86_sse_comigt_ss : GCCBuiltin<"__builtin_ia32_comigt">,
|
|
Intrinsic<[llvm_int_ty, llvm_v4f32_ty,
|
|
llvm_v4f32_ty], [IntrNoMem]>;
|
|
def int_x86_sse_comige_ss : GCCBuiltin<"__builtin_ia32_comige">,
|
|
Intrinsic<[llvm_int_ty, llvm_v4f32_ty,
|
|
llvm_v4f32_ty], [IntrNoMem]>;
|
|
def int_x86_sse_comineq_ss : GCCBuiltin<"__builtin_ia32_comineq">,
|
|
Intrinsic<[llvm_int_ty, llvm_v4f32_ty,
|
|
llvm_v4f32_ty], [IntrNoMem]>;
|
|
def int_x86_sse_ucomieq_ss : GCCBuiltin<"__builtin_ia32_ucomieq">,
|
|
Intrinsic<[llvm_int_ty, llvm_v4f32_ty,
|
|
llvm_v4f32_ty], [IntrNoMem]>;
|
|
def int_x86_sse_ucomilt_ss : GCCBuiltin<"__builtin_ia32_ucomilt">,
|
|
Intrinsic<[llvm_int_ty, llvm_v4f32_ty,
|
|
llvm_v4f32_ty], [IntrNoMem]>;
|
|
def int_x86_sse_ucomile_ss : GCCBuiltin<"__builtin_ia32_ucomile">,
|
|
Intrinsic<[llvm_int_ty, llvm_v4f32_ty,
|
|
llvm_v4f32_ty], [IntrNoMem]>;
|
|
def int_x86_sse_ucomigt_ss : GCCBuiltin<"__builtin_ia32_ucomigt">,
|
|
Intrinsic<[llvm_int_ty, llvm_v4f32_ty,
|
|
llvm_v4f32_ty], [IntrNoMem]>;
|
|
def int_x86_sse_ucomige_ss : GCCBuiltin<"__builtin_ia32_ucomige">,
|
|
Intrinsic<[llvm_int_ty, llvm_v4f32_ty,
|
|
llvm_v4f32_ty], [IntrNoMem]>;
|
|
def int_x86_sse_ucomineq_ss : GCCBuiltin<"__builtin_ia32_ucomineq">,
|
|
Intrinsic<[llvm_int_ty, llvm_v4f32_ty,
|
|
llvm_v4f32_ty], [IntrNoMem]>;
|
|
}
|
|
|
|
|
|
// Conversion ops
|
|
let TargetPrefix = "x86" in { // All intrinsics start with "llvm.x86.".
|
|
def int_x86_sse_cvtss2si : GCCBuiltin<"__builtin_ia32_cvtss2si">,
|
|
Intrinsic<[llvm_int_ty, llvm_v4f32_ty], [IntrNoMem]>;
|
|
def int_x86_sse_cvttss2si : GCCBuiltin<"__builtin_ia32_cvttss2si">,
|
|
Intrinsic<[llvm_int_ty, llvm_v4f32_ty], [IntrNoMem]>;
|
|
def int_x86_sse_cvtsi2ss : GCCBuiltin<"__builtin_ia32_cvtsi2ss">,
|
|
Intrinsic<[llvm_v4f32_ty, llvm_v4f32_ty,
|
|
llvm_int_ty], [IntrNoMem]>;
|
|
}
|
|
|
|
// SIMD load ops
|
|
let TargetPrefix = "x86" in { // All intrinsics start with "llvm.x86.".
|
|
def int_x86_sse_loadu_ps : GCCBuiltin<"__builtin_ia32_loadups">,
|
|
Intrinsic<[llvm_v4f32_ty, llvm_ptr_ty], [IntrReadMem]>;
|
|
}
|
|
|
|
// SIMD store ops
|
|
let TargetPrefix = "x86" in { // All intrinsics start with "llvm.x86.".
|
|
def int_x86_sse_storeu_ps : GCCBuiltin<"__builtin_ia32_storeups">,
|
|
Intrinsic<[llvm_void_ty, llvm_ptr_ty,
|
|
llvm_v4f32_ty], [IntrWriteMem]>;
|
|
}
|
|
|
|
// Cacheability support ops
|
|
let TargetPrefix = "x86" in { // All intrinsics start with "llvm.x86.".
|
|
def int_x86_sse_movnt_ps : GCCBuiltin<"__builtin_ia32_movntps">,
|
|
Intrinsic<[llvm_void_ty, llvm_ptr_ty,
|
|
llvm_v4f32_ty], [IntrWriteMem]>;
|
|
def int_x86_sse_sfence : GCCBuiltin<"__builtin_ia32_sfence">,
|
|
Intrinsic<[llvm_void_ty], [IntrWriteMem]>;
|
|
}
|
|
|
|
// Control register.
|
|
let TargetPrefix = "x86" in { // All intrinsics start with "llvm.x86.".
|
|
def int_x86_sse_stmxcsr :
|
|
Intrinsic<[llvm_void_ty, llvm_ptr_ty], [IntrWriteMem]>;
|
|
def int_x86_sse_ldmxcsr :
|
|
Intrinsic<[llvm_void_ty, llvm_ptr_ty], [IntrWriteMem]>;
|
|
}
|
|
|
|
// Misc.
|
|
let TargetPrefix = "x86" in { // All intrinsics start with "llvm.x86.".
|
|
def int_x86_sse_movmsk_ps : GCCBuiltin<"__builtin_ia32_movmskps">,
|
|
Intrinsic<[llvm_int_ty, llvm_v4f32_ty], [IntrNoMem]>;
|
|
}
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
// SSE2
|
|
|
|
// FP arithmetic ops
|
|
let TargetPrefix = "x86" in { // All intrinsics start with "llvm.x86.".
|
|
def int_x86_sse2_add_sd : GCCBuiltin<"__builtin_ia32_addsd">,
|
|
Intrinsic<[llvm_v2f64_ty, llvm_v2f64_ty,
|
|
llvm_v2f64_ty], [IntrNoMem]>;
|
|
def int_x86_sse2_sub_sd : GCCBuiltin<"__builtin_ia32_subsd">,
|
|
Intrinsic<[llvm_v2f64_ty, llvm_v2f64_ty,
|
|
llvm_v2f64_ty], [IntrNoMem]>;
|
|
def int_x86_sse2_mul_sd : GCCBuiltin<"__builtin_ia32_mulsd">,
|
|
Intrinsic<[llvm_v2f64_ty, llvm_v2f64_ty,
|
|
llvm_v2f64_ty], [IntrNoMem]>;
|
|
def int_x86_sse2_div_sd : GCCBuiltin<"__builtin_ia32_divsd">,
|
|
Intrinsic<[llvm_v2f64_ty, llvm_v2f64_ty,
|
|
llvm_v2f64_ty], [IntrNoMem]>;
|
|
def int_x86_sse2_sqrt_sd : GCCBuiltin<"__builtin_ia32_sqrtsd">,
|
|
Intrinsic<[llvm_v2f64_ty, llvm_v2f64_ty],
|
|
[IntrNoMem]>;
|
|
def int_x86_sse2_sqrt_pd : GCCBuiltin<"__builtin_ia32_sqrtpd">,
|
|
Intrinsic<[llvm_v2f64_ty, llvm_v2f64_ty],
|
|
[IntrNoMem]>;
|
|
def int_x86_sse2_rcp_sd : GCCBuiltin<"__builtin_ia32_rcpsd">,
|
|
Intrinsic<[llvm_v2f64_ty, llvm_v2f64_ty],
|
|
[IntrNoMem]>;
|
|
def int_x86_sse2_rcp_pd : GCCBuiltin<"__builtin_ia32_rcppd">,
|
|
Intrinsic<[llvm_v2f64_ty, llvm_v2f64_ty],
|
|
[IntrNoMem]>;
|
|
def int_x86_sse2_rsqrt_sd : GCCBuiltin<"__builtin_ia32_rsqrtsd">,
|
|
Intrinsic<[llvm_v2f64_ty, llvm_v2f64_ty],
|
|
[IntrNoMem]>;
|
|
def int_x86_sse2_rsqrt_pd : GCCBuiltin<"__builtin_ia32_rsqrtpd">,
|
|
Intrinsic<[llvm_v2f64_ty, llvm_v2f64_ty],
|
|
[IntrNoMem]>;
|
|
def int_x86_sse2_min_sd : GCCBuiltin<"__builtin_ia32_minsd">,
|
|
Intrinsic<[llvm_v2f64_ty, llvm_v2f64_ty,
|
|
llvm_v2f64_ty], [IntrNoMem]>;
|
|
def int_x86_sse2_min_pd : GCCBuiltin<"__builtin_ia32_minpd">,
|
|
Intrinsic<[llvm_v2f64_ty, llvm_v2f64_ty,
|
|
llvm_v2f64_ty], [IntrNoMem]>;
|
|
def int_x86_sse2_max_sd : GCCBuiltin<"__builtin_ia32_maxsd">,
|
|
Intrinsic<[llvm_v2f64_ty, llvm_v2f64_ty,
|
|
llvm_v2f64_ty], [IntrNoMem]>;
|
|
def int_x86_sse2_max_pd : GCCBuiltin<"__builtin_ia32_maxpd">,
|
|
Intrinsic<[llvm_v2f64_ty, llvm_v2f64_ty,
|
|
llvm_v2f64_ty], [IntrNoMem]>;
|
|
}
|
|
|
|
// FP comparison ops
|
|
let TargetPrefix = "x86" in { // All intrinsics start with "llvm.x86.".
|
|
def int_x86_sse2_cmp_sd :
|
|
Intrinsic<[llvm_v2f64_ty, llvm_v2f64_ty,
|
|
llvm_v2f64_ty, llvm_sbyte_ty], [IntrNoMem]>;
|
|
def int_x86_sse2_cmp_pd :
|
|
Intrinsic<[llvm_v2f64_ty, llvm_v2f64_ty,
|
|
llvm_v2f64_ty, llvm_sbyte_ty], [IntrNoMem]>;
|
|
def int_x86_sse2_comieq_sd : GCCBuiltin<"__builtin_ia32_comisdeq">,
|
|
Intrinsic<[llvm_int_ty, llvm_v2f64_ty,
|
|
llvm_v2f64_ty], [IntrNoMem]>;
|
|
def int_x86_sse2_comilt_sd : GCCBuiltin<"__builtin_ia32_comisdlt">,
|
|
Intrinsic<[llvm_int_ty, llvm_v2f64_ty,
|
|
llvm_v2f64_ty], [IntrNoMem]>;
|
|
def int_x86_sse2_comile_sd : GCCBuiltin<"__builtin_ia32_comisdle">,
|
|
Intrinsic<[llvm_int_ty, llvm_v2f64_ty,
|
|
llvm_v2f64_ty], [IntrNoMem]>;
|
|
def int_x86_sse2_comigt_sd : GCCBuiltin<"__builtin_ia32_comisdgt">,
|
|
Intrinsic<[llvm_int_ty, llvm_v2f64_ty,
|
|
llvm_v2f64_ty], [IntrNoMem]>;
|
|
def int_x86_sse2_comige_sd : GCCBuiltin<"__builtin_ia32_comisdge">,
|
|
Intrinsic<[llvm_int_ty, llvm_v2f64_ty,
|
|
llvm_v2f64_ty], [IntrNoMem]>;
|
|
def int_x86_sse2_comineq_sd : GCCBuiltin<"__builtin_ia32_comisdneq">,
|
|
Intrinsic<[llvm_int_ty, llvm_v2f64_ty,
|
|
llvm_v2f64_ty], [IntrNoMem]>;
|
|
def int_x86_sse2_ucomieq_sd : GCCBuiltin<"__builtin_ia32_ucomisdeq">,
|
|
Intrinsic<[llvm_int_ty, llvm_v2f64_ty,
|
|
llvm_v2f64_ty], [IntrNoMem]>;
|
|
def int_x86_sse2_ucomilt_sd : GCCBuiltin<"__builtin_ia32_ucomisdlt">,
|
|
Intrinsic<[llvm_int_ty, llvm_v2f64_ty,
|
|
llvm_v2f64_ty], [IntrNoMem]>;
|
|
def int_x86_sse2_ucomile_sd : GCCBuiltin<"__builtin_ia32_ucomisdle">,
|
|
Intrinsic<[llvm_int_ty, llvm_v2f64_ty,
|
|
llvm_v2f64_ty], [IntrNoMem]>;
|
|
def int_x86_sse2_ucomigt_sd : GCCBuiltin<"__builtin_ia32_ucomisdgt">,
|
|
Intrinsic<[llvm_int_ty, llvm_v2f64_ty,
|
|
llvm_v2f64_ty], [IntrNoMem]>;
|
|
def int_x86_sse2_ucomige_sd : GCCBuiltin<"__builtin_ia32_ucomisdge">,
|
|
Intrinsic<[llvm_int_ty, llvm_v2f64_ty,
|
|
llvm_v2f64_ty], [IntrNoMem]>;
|
|
def int_x86_sse2_ucomineq_sd : GCCBuiltin<"__builtin_ia32_ucomisdneq">,
|
|
Intrinsic<[llvm_int_ty, llvm_v2f64_ty,
|
|
llvm_v2f64_ty], [IntrNoMem]>;
|
|
}
|
|
|
|
// Integer arithmetic ops.
|
|
let TargetPrefix = "x86" in { // All intrinsics start with "llvm.x86.".
|
|
def int_x86_sse2_padds_b : GCCBuiltin<"__builtin_ia32_paddsb128">,
|
|
Intrinsic<[llvm_v16i8_ty, llvm_v16i8_ty,
|
|
llvm_v16i8_ty], [IntrNoMem]>;
|
|
def int_x86_sse2_padds_w : GCCBuiltin<"__builtin_ia32_paddsw128">,
|
|
Intrinsic<[llvm_v8i16_ty, llvm_v8i16_ty,
|
|
llvm_v8i16_ty], [IntrNoMem]>;
|
|
def int_x86_sse2_paddus_b : GCCBuiltin<"__builtin_ia32_paddusb128">,
|
|
Intrinsic<[llvm_v16i8_ty, llvm_v16i8_ty,
|
|
llvm_v16i8_ty], [IntrNoMem]>;
|
|
def int_x86_sse2_paddus_w : GCCBuiltin<"__builtin_ia32_paddusw128">,
|
|
Intrinsic<[llvm_v8i16_ty, llvm_v8i16_ty,
|
|
llvm_v8i16_ty], [IntrNoMem]>;
|
|
def int_x86_sse2_psubs_b : GCCBuiltin<"__builtin_ia32_psubsb128">,
|
|
Intrinsic<[llvm_v16i8_ty, llvm_v16i8_ty,
|
|
llvm_v16i8_ty], [IntrNoMem]>;
|
|
def int_x86_sse2_psubs_w : GCCBuiltin<"__builtin_ia32_psubsw128">,
|
|
Intrinsic<[llvm_v8i16_ty, llvm_v8i16_ty,
|
|
llvm_v8i16_ty], [IntrNoMem]>;
|
|
def int_x86_sse2_psubus_b : GCCBuiltin<"__builtin_ia32_psubusb128">,
|
|
Intrinsic<[llvm_v16i8_ty, llvm_v16i8_ty,
|
|
llvm_v16i8_ty], [IntrNoMem]>;
|
|
def int_x86_sse2_psubus_w : GCCBuiltin<"__builtin_ia32_psubusw128">,
|
|
Intrinsic<[llvm_v8i16_ty, llvm_v8i16_ty,
|
|
llvm_v8i16_ty], [IntrNoMem]>;
|
|
def int_x86_sse2_pmulhu_w : GCCBuiltin<"__builtin_ia32_pmulhuw128">,
|
|
Intrinsic<[llvm_v8i16_ty, llvm_v8i16_ty,
|
|
llvm_v8i16_ty], [IntrNoMem]>;
|
|
def int_x86_sse2_pmulh_w : GCCBuiltin<"__builtin_ia32_pmulhw128">,
|
|
Intrinsic<[llvm_v8i16_ty, llvm_v8i16_ty,
|
|
llvm_v8i16_ty], [IntrNoMem]>;
|
|
def int_x86_sse2_pmulu_dq : GCCBuiltin<"__builtin_ia32_pmuludq128">,
|
|
Intrinsic<[llvm_v2i64_ty, llvm_v4i32_ty,
|
|
llvm_v4i32_ty], [IntrNoMem]>;
|
|
def int_x86_sse2_pmadd_wd : GCCBuiltin<"__builtin_ia32_pmaddwd128">,
|
|
Intrinsic<[llvm_v4i32_ty, llvm_v8i16_ty,
|
|
llvm_v8i16_ty], [IntrNoMem]>;
|
|
def int_x86_sse2_pavg_b : GCCBuiltin<"__builtin_ia32_pavgb128">,
|
|
Intrinsic<[llvm_v16i8_ty, llvm_v16i8_ty,
|
|
llvm_v16i8_ty], [IntrNoMem]>;
|
|
def int_x86_sse2_pavg_w : GCCBuiltin<"__builtin_ia32_pavgw128">,
|
|
Intrinsic<[llvm_v8i16_ty, llvm_v8i16_ty,
|
|
llvm_v8i16_ty], [IntrNoMem]>;
|
|
def int_x86_sse2_pmaxu_b : GCCBuiltin<"__builtin_ia32_pmaxub128">,
|
|
Intrinsic<[llvm_v16i8_ty, llvm_v16i8_ty,
|
|
llvm_v16i8_ty], [IntrNoMem]>;
|
|
def int_x86_sse2_pmaxs_w : GCCBuiltin<"__builtin_ia32_pmaxsw128">,
|
|
Intrinsic<[llvm_v8i16_ty, llvm_v8i16_ty,
|
|
llvm_v8i16_ty], [IntrNoMem]>;
|
|
def int_x86_sse2_pminu_b : GCCBuiltin<"__builtin_ia32_pminub128">,
|
|
Intrinsic<[llvm_v16i8_ty, llvm_v16i8_ty,
|
|
llvm_v16i8_ty], [IntrNoMem]>;
|
|
def int_x86_sse2_pmins_w : GCCBuiltin<"__builtin_ia32_pminsw128">,
|
|
Intrinsic<[llvm_v8i16_ty, llvm_v8i16_ty,
|
|
llvm_v8i16_ty], [IntrNoMem]>;
|
|
def int_x86_sse2_psad_bw : GCCBuiltin<"__builtin_ia32_psadbw128">,
|
|
Intrinsic<[llvm_v2i64_ty, llvm_v16i8_ty,
|
|
llvm_v16i8_ty], [IntrNoMem]>;
|
|
}
|
|
|
|
// Integer shift ops.
|
|
let TargetPrefix = "x86" in { // All intrinsics start with "llvm.x86.".
|
|
def int_x86_sse2_psll_w :
|
|
Intrinsic<[llvm_v8i16_ty, llvm_v8i16_ty,
|
|
llvm_v4i32_ty], [IntrNoMem]>;
|
|
def int_x86_sse2_psll_d :
|
|
Intrinsic<[llvm_v4i32_ty, llvm_v4i32_ty,
|
|
llvm_v4i32_ty], [IntrNoMem]>;
|
|
def int_x86_sse2_psll_q :
|
|
Intrinsic<[llvm_v2i64_ty, llvm_v2i64_ty,
|
|
llvm_v4i32_ty], [IntrNoMem]>;
|
|
def int_x86_sse2_psll_dq : GCCBuiltin<"__builtin_ia32_pslldqi128">,
|
|
Intrinsic<[llvm_v2i64_ty, llvm_v2i64_ty,
|
|
llvm_int_ty], [IntrNoMem]>;
|
|
def int_x86_sse2_psrl_w :
|
|
Intrinsic<[llvm_v8i16_ty, llvm_v8i16_ty,
|
|
llvm_v4i32_ty], [IntrNoMem]>;
|
|
def int_x86_sse2_psrl_d :
|
|
Intrinsic<[llvm_v4i32_ty, llvm_v4i32_ty,
|
|
llvm_v4i32_ty], [IntrNoMem]>;
|
|
def int_x86_sse2_psrl_q :
|
|
Intrinsic<[llvm_v2i64_ty, llvm_v2i64_ty,
|
|
llvm_v4i32_ty], [IntrNoMem]>;
|
|
def int_x86_sse2_psrl_dq : GCCBuiltin<"__builtin_ia32_psrldqi128">,
|
|
Intrinsic<[llvm_v2i64_ty, llvm_v2i64_ty,
|
|
llvm_int_ty], [IntrNoMem]>;
|
|
def int_x86_sse2_psra_w :
|
|
Intrinsic<[llvm_v8i16_ty, llvm_v8i16_ty,
|
|
llvm_v4i32_ty], [IntrNoMem]>;
|
|
def int_x86_sse2_psra_d :
|
|
Intrinsic<[llvm_v4i32_ty, llvm_v4i32_ty,
|
|
llvm_v4i32_ty], [IntrNoMem]>;
|
|
}
|
|
|
|
// Integer comparison ops
|
|
let TargetPrefix = "x86" in { // All intrinsics start with "llvm.x86.".
|
|
def int_x86_sse2_pcmpeq_b : GCCBuiltin<"__builtin_ia32_pcmpeqb128">,
|
|
Intrinsic<[llvm_v16i8_ty, llvm_v16i8_ty,
|
|
llvm_v16i8_ty], [IntrNoMem]>;
|
|
def int_x86_sse2_pcmpeq_w : GCCBuiltin<"__builtin_ia32_pcmpeqw128">,
|
|
Intrinsic<[llvm_v8i16_ty, llvm_v8i16_ty,
|
|
llvm_v8i16_ty], [IntrNoMem]>;
|
|
def int_x86_sse2_pcmpeq_d : GCCBuiltin<"__builtin_ia32_pcmpeqd128">,
|
|
Intrinsic<[llvm_v4i32_ty, llvm_v4i32_ty,
|
|
llvm_v4i32_ty], [IntrNoMem]>;
|
|
def int_x86_sse2_pcmpgt_b : GCCBuiltin<"__builtin_ia32_pcmpgtb128">,
|
|
Intrinsic<[llvm_v16i8_ty, llvm_v16i8_ty,
|
|
llvm_v16i8_ty], [IntrNoMem]>;
|
|
def int_x86_sse2_pcmpgt_w : GCCBuiltin<"__builtin_ia32_pcmpgtw128">,
|
|
Intrinsic<[llvm_v8i16_ty, llvm_v8i16_ty,
|
|
llvm_v8i16_ty], [IntrNoMem]>;
|
|
def int_x86_sse2_pcmpgt_d : GCCBuiltin<"__builtin_ia32_pcmpgtd128">,
|
|
Intrinsic<[llvm_v4i32_ty, llvm_v4i32_ty,
|
|
llvm_v4i32_ty], [IntrNoMem]>;
|
|
}
|
|
|
|
// Conversion ops
|
|
let TargetPrefix = "x86" in { // All intrinsics start with "llvm.x86.".
|
|
def int_x86_sse2_cvtdq2pd : GCCBuiltin<"__builtin_ia32_cvtdq2pd">,
|
|
Intrinsic<[llvm_v2f64_ty, llvm_v4i32_ty], [IntrNoMem]>;
|
|
def int_x86_sse2_cvtdq2ps : GCCBuiltin<"__builtin_ia32_cvtdq2ps">,
|
|
Intrinsic<[llvm_v4f32_ty, llvm_v4i32_ty], [IntrNoMem]>;
|
|
def int_x86_sse2_cvtpd2dq : GCCBuiltin<"__builtin_ia32_cvtpd2dq">,
|
|
Intrinsic<[llvm_v4i32_ty, llvm_v2f64_ty], [IntrNoMem]>;
|
|
def int_x86_sse2_cvttpd2dq : GCCBuiltin<"__builtin_ia32_cvttpd2dq">,
|
|
Intrinsic<[llvm_v4i32_ty, llvm_v2f64_ty], [IntrNoMem]>;
|
|
def int_x86_sse2_cvtpd2ps : GCCBuiltin<"__builtin_ia32_cvtpd2ps">,
|
|
Intrinsic<[llvm_v4f32_ty, llvm_v2f64_ty], [IntrNoMem]>;
|
|
def int_x86_sse2_cvtps2dq : GCCBuiltin<"__builtin_ia32_cvtps2dq">,
|
|
Intrinsic<[llvm_v4i32_ty, llvm_v4f32_ty], [IntrNoMem]>;
|
|
def int_x86_sse2_cvttps2dq : GCCBuiltin<"__builtin_ia32_cvttps2dq">,
|
|
Intrinsic<[llvm_v4i32_ty, llvm_v4f32_ty], [IntrNoMem]>;
|
|
def int_x86_sse2_cvtps2pd : GCCBuiltin<"__builtin_ia32_cvtps2pd">,
|
|
Intrinsic<[llvm_v2f64_ty, llvm_v4f32_ty], [IntrNoMem]>;
|
|
def int_x86_sse2_cvtsd2si : GCCBuiltin<"__builtin_ia32_cvtsd2si">,
|
|
Intrinsic<[llvm_int_ty, llvm_v2f64_ty], [IntrNoMem]>;
|
|
def int_x86_sse2_cvttsd2si : GCCBuiltin<"__builtin_ia32_cvttsd2si">,
|
|
Intrinsic<[llvm_int_ty, llvm_v2f64_ty], [IntrNoMem]>;
|
|
def int_x86_sse2_cvtsi2sd : GCCBuiltin<"__builtin_ia32_cvtsi2sd">,
|
|
Intrinsic<[llvm_v2f64_ty, llvm_v2f64_ty,
|
|
llvm_int_ty], [IntrNoMem]>;
|
|
def int_x86_sse2_cvtsd2ss : GCCBuiltin<"__builtin_ia32_cvtsd2ss">,
|
|
Intrinsic<[llvm_v4f32_ty, llvm_v4f32_ty,
|
|
llvm_v2f64_ty], [IntrNoMem]>;
|
|
def int_x86_sse2_cvtss2sd : GCCBuiltin<"__builtin_ia32_cvtss2sd">,
|
|
Intrinsic<[llvm_v2f64_ty, llvm_v2f64_ty,
|
|
llvm_v4f32_ty], [IntrNoMem]>;
|
|
}
|
|
|
|
// SIMD load ops
|
|
let TargetPrefix = "x86" in { // All intrinsics start with "llvm.x86.".
|
|
def int_x86_sse2_loadu_pd : GCCBuiltin<"__builtin_ia32_loadupd">,
|
|
Intrinsic<[llvm_v2f64_ty, llvm_ptr_ty], [IntrReadMem]>;
|
|
def int_x86_sse2_loadu_dq : GCCBuiltin<"__builtin_ia32_loaddqu">,
|
|
Intrinsic<[llvm_v16i8_ty, llvm_ptr_ty], [IntrReadMem]>;
|
|
}
|
|
|
|
// SIMD store ops
|
|
let TargetPrefix = "x86" in { // All intrinsics start with "llvm.x86.".
|
|
def int_x86_sse2_storeu_pd : GCCBuiltin<"__builtin_ia32_storeupd">,
|
|
Intrinsic<[llvm_void_ty, llvm_ptr_ty,
|
|
llvm_v2f64_ty], [IntrWriteMem]>;
|
|
def int_x86_sse2_storeu_dq : GCCBuiltin<"__builtin_ia32_storedqu">,
|
|
Intrinsic<[llvm_void_ty, llvm_ptr_ty,
|
|
llvm_v16i8_ty], [IntrWriteMem]>;
|
|
def int_x86_sse2_storel_dq : GCCBuiltin<"__builtin_ia32_storelv4si">,
|
|
Intrinsic<[llvm_void_ty, llvm_ptr_ty,
|
|
llvm_v4i32_ty], [IntrWriteMem]>;
|
|
}
|
|
|
|
// Cacheability support ops
|
|
let TargetPrefix = "x86" in { // All intrinsics start with "llvm.x86.".
|
|
def int_x86_sse2_movnt_dq : GCCBuiltin<"__builtin_ia32_movntdq">,
|
|
Intrinsic<[llvm_void_ty, llvm_ptr_ty,
|
|
llvm_v2i64_ty], [IntrWriteMem]>;
|
|
def int_x86_sse2_movnt_pd : GCCBuiltin<"__builtin_ia32_movntpd">,
|
|
Intrinsic<[llvm_void_ty, llvm_ptr_ty,
|
|
llvm_v2f64_ty], [IntrWriteMem]>;
|
|
def int_x86_sse2_movnt_i : GCCBuiltin<"__builtin_ia32_movnti">,
|
|
Intrinsic<[llvm_void_ty, llvm_ptr_ty,
|
|
llvm_int_ty], [IntrWriteMem]>;
|
|
}
|
|
|
|
// Misc.
|
|
let TargetPrefix = "x86" in { // All intrinsics start with "llvm.x86.".
|
|
def int_x86_sse2_packsswb_128 : GCCBuiltin<"__builtin_ia32_packsswb128">,
|
|
Intrinsic<[llvm_v8i16_ty, llvm_v8i16_ty,
|
|
llvm_v8i16_ty], [IntrNoMem]>;
|
|
def int_x86_sse2_packssdw_128 : GCCBuiltin<"__builtin_ia32_packssdw128">,
|
|
Intrinsic<[llvm_v4i32_ty, llvm_v4i32_ty,
|
|
llvm_v4i32_ty], [IntrNoMem]>;
|
|
def int_x86_sse2_packuswb_128 : GCCBuiltin<"__builtin_ia32_packuswb128">,
|
|
Intrinsic<[llvm_v8i16_ty, llvm_v8i16_ty,
|
|
llvm_v8i16_ty], [IntrNoMem]>;
|
|
def int_x86_sse2_movl_dq : GCCBuiltin<"__builtin_ia32_movqv4si">,
|
|
Intrinsic<[llvm_v4i32_ty, llvm_v4i32_ty], [IntrNoMem]>;
|
|
def int_x86_sse2_movmsk_pd : GCCBuiltin<"__builtin_ia32_movmskpd">,
|
|
Intrinsic<[llvm_int_ty, llvm_v2f64_ty], [IntrNoMem]>;
|
|
def int_x86_sse2_pmovmskb_128 : GCCBuiltin<"__builtin_ia32_pmovmskb128">,
|
|
Intrinsic<[llvm_int_ty, llvm_v16i8_ty], [IntrNoMem]>;
|
|
def int_x86_sse2_maskmov_dqu : GCCBuiltin<"__builtin_ia32_maskmovdqu">,
|
|
Intrinsic<[llvm_void_ty, llvm_v16i8_ty,
|
|
llvm_v16i8_ty, llvm_ptr_ty], [IntrWriteMem]>;
|
|
def int_x86_sse2_clflush : GCCBuiltin<"__builtin_ia32_clflush">,
|
|
Intrinsic<[llvm_void_ty, llvm_ptr_ty], [IntrWriteMem]>;
|
|
def int_x86_sse2_lfence : GCCBuiltin<"__builtin_ia32_lfence">,
|
|
Intrinsic<[llvm_void_ty], [IntrWriteMem]>;
|
|
def int_x86_sse2_mfence : GCCBuiltin<"__builtin_ia32_mfence">,
|
|
Intrinsic<[llvm_void_ty], [IntrWriteMem]>;
|
|
}
|
|
|
|
// Shuffles.
|
|
// FIXME: Temporary workarounds since 2-wide shuffle is broken.
|
|
let TargetPrefix = "x86" in { // All intrinsics start with "llvm.x86.".
|
|
def int_x86_sse2_movs_d : GCCBuiltin<"__builtin_ia32_movsd">,
|
|
Intrinsic<[llvm_v2f64_ty, llvm_v2f64_ty,
|
|
llvm_v2f64_ty], [IntrNoMem]>;
|
|
def int_x86_sse2_loadh_pd : GCCBuiltin<"__builtin_ia32_loadhpd">,
|
|
Intrinsic<[llvm_v2f64_ty, llvm_v2f64_ty,
|
|
llvm_ptr_ty], [IntrReadMem]>;
|
|
def int_x86_sse2_loadl_pd : GCCBuiltin<"__builtin_ia32_loadlpd">,
|
|
Intrinsic<[llvm_v2f64_ty, llvm_v2f64_ty,
|
|
llvm_ptr_ty], [IntrReadMem]>;
|
|
def int_x86_sse2_shuf_pd : GCCBuiltin<"__builtin_ia32_shufpd">,
|
|
Intrinsic<[llvm_v2f64_ty, llvm_v2f64_ty,
|
|
llvm_v2f64_ty, llvm_int_ty], [IntrNoMem]>;
|
|
def int_x86_sse2_unpckh_pd : GCCBuiltin<"__builtin_ia32_unpckhpd">,
|
|
Intrinsic<[llvm_v2f64_ty, llvm_v2f64_ty,
|
|
llvm_v2f64_ty], [IntrNoMem]>;
|
|
def int_x86_sse2_unpckl_pd : GCCBuiltin<"__builtin_ia32_unpcklpd">,
|
|
Intrinsic<[llvm_v2f64_ty, llvm_v2f64_ty,
|
|
llvm_v2f64_ty], [IntrNoMem]>;
|
|
def int_x86_sse2_punpckh_qdq : GCCBuiltin<"__builtin_ia32_punpckhqdq128">,
|
|
Intrinsic<[llvm_v2i64_ty, llvm_v2i64_ty,
|
|
llvm_v2i64_ty], [IntrNoMem]>;
|
|
def int_x86_sse2_punpckl_qdq : GCCBuiltin<"__builtin_ia32_punpcklqdq128">,
|
|
Intrinsic<[llvm_v2i64_ty, llvm_v2i64_ty,
|
|
llvm_v2i64_ty], [IntrNoMem]>;
|
|
}
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
// SSE3
|
|
|
|
// Addition / subtraction ops.
|
|
let TargetPrefix = "x86" in { // All intrinsics start with "llvm.x86.".
|
|
def int_x86_sse3_addsub_ps : GCCBuiltin<"__builtin_ia32_addsubps">,
|
|
Intrinsic<[llvm_v4f32_ty, llvm_v4f32_ty,
|
|
llvm_v4f32_ty], [IntrNoMem]>;
|
|
def int_x86_sse3_addsub_pd : GCCBuiltin<"__builtin_ia32_addsubpd">,
|
|
Intrinsic<[llvm_v2f64_ty, llvm_v2f64_ty,
|
|
llvm_v2f64_ty], [IntrNoMem]>;
|
|
}
|
|
|
|
// Horizontal ops.
|
|
let TargetPrefix = "x86" in { // All intrinsics start with "llvm.x86.".
|
|
def int_x86_sse3_hadd_ps : GCCBuiltin<"__builtin_ia32_haddps">,
|
|
Intrinsic<[llvm_v4f32_ty, llvm_v4f32_ty,
|
|
llvm_v4f32_ty], [IntrNoMem]>;
|
|
def int_x86_sse3_hadd_pd : GCCBuiltin<"__builtin_ia32_haddpd">,
|
|
Intrinsic<[llvm_v2f64_ty, llvm_v2f64_ty,
|
|
llvm_v2f64_ty], [IntrNoMem]>;
|
|
def int_x86_sse3_hsub_ps : GCCBuiltin<"__builtin_ia32_hsubps">,
|
|
Intrinsic<[llvm_v4f32_ty, llvm_v4f32_ty,
|
|
llvm_v4f32_ty], [IntrNoMem]>;
|
|
def int_x86_sse3_hsub_pd : GCCBuiltin<"__builtin_ia32_hsubpd">,
|
|
Intrinsic<[llvm_v2f64_ty, llvm_v2f64_ty,
|
|
llvm_v2f64_ty], [IntrNoMem]>;
|
|
}
|
|
|
|
// Specialized unaligned load.
|
|
let TargetPrefix = "x86" in { // All intrinsics start with "llvm.x86.".
|
|
def int_x86_sse3_ldu_dq : GCCBuiltin<"__builtin_ia32_lddqu">,
|
|
Intrinsic<[llvm_v16i8_ty, llvm_ptr_ty], [IntrReadMem]>;
|
|
}
|
|
|
|
// Thread synchronization ops.
|
|
let TargetPrefix = "x86" in { // All intrinsics start with "llvm.x86.".
|
|
def int_x86_sse3_monitor : GCCBuiltin<"__builtin_ia32_monitor">,
|
|
Intrinsic<[llvm_void_ty, llvm_ptr_ty,
|
|
llvm_uint_ty, llvm_uint_ty], [IntrWriteMem]>;
|
|
def int_x86_sse3_mwait : GCCBuiltin<"__builtin_ia32_mwait">,
|
|
Intrinsic<[llvm_void_ty, llvm_uint_ty,
|
|
llvm_uint_ty], [IntrWriteMem]>;
|
|
}
|