1 //===-- AMDGPUInstrInfo.td - AMDGPU DAG nodes --------------*- tablegen -*-===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This file contains DAG node defintions for the AMDGPU target.
12 //===----------------------------------------------------------------------===//
14 //===----------------------------------------------------------------------===//
15 // AMDGPU DAG Profiles
16 //===----------------------------------------------------------------------===//
18 def AMDGPUDTIntTernaryOp : SDTypeProfile<1, 3, [
19 SDTCisSameAs<0, 1>, SDTCisSameAs<0, 2>, SDTCisInt<0>, SDTCisInt<3>
22 //===----------------------------------------------------------------------===//
26 // This argument to this node is a dword address.
27 def AMDGPUdwordaddr : SDNode<"AMDGPUISD::DWORDADDR", SDTIntUnaryOp>;
30 def AMDGPUfract : SDNode<"AMDGPUISD::FRACT", SDTFPUnaryOp>;
32 // out = max(a, b) a and b are floats
33 def AMDGPUfmax : SDNode<"AMDGPUISD::FMAX", SDTFPBinOp,
34 [SDNPCommutative, SDNPAssociative]
37 // out = max(a, b) a and b are signed ints
38 def AMDGPUsmax : SDNode<"AMDGPUISD::SMAX", SDTIntBinOp,
39 [SDNPCommutative, SDNPAssociative]
42 // out = max(a, b) a and b are unsigned ints
43 def AMDGPUumax : SDNode<"AMDGPUISD::UMAX", SDTIntBinOp,
44 [SDNPCommutative, SDNPAssociative]
47 // out = min(a, b) a and b are floats
48 def AMDGPUfmin : SDNode<"AMDGPUISD::FMIN", SDTFPBinOp,
49 [SDNPCommutative, SDNPAssociative]
52 // out = min(a, b) a snd b are signed ints
53 def AMDGPUsmin : SDNode<"AMDGPUISD::SMIN", SDTIntBinOp,
54 [SDNPCommutative, SDNPAssociative]
57 // out = min(a, b) a and b are unsigned ints
58 def AMDGPUumin : SDNode<"AMDGPUISD::UMIN", SDTIntBinOp,
59 [SDNPCommutative, SDNPAssociative]
63 def AMDGPUcvt_f32_ubyte0 : SDNode<"AMDGPUISD::CVT_F32_UBYTE0",
65 def AMDGPUcvt_f32_ubyte1 : SDNode<"AMDGPUISD::CVT_F32_UBYTE1",
67 def AMDGPUcvt_f32_ubyte2 : SDNode<"AMDGPUISD::CVT_F32_UBYTE2",
69 def AMDGPUcvt_f32_ubyte3 : SDNode<"AMDGPUISD::CVT_F32_UBYTE3",
73 // urecip - This operation is a helper for integer division, it returns the
74 // result of 1 / a as a fractional unsigned integer.
75 // out = (2^32 / a) + e
76 // e is rounding error
77 def AMDGPUurecip : SDNode<"AMDGPUISD::URECIP", SDTIntUnaryOp>;
79 def AMDGPUregister_load : SDNode<"AMDGPUISD::REGISTER_LOAD",
80 SDTypeProfile<1, 2, [SDTCisPtrTy<1>, SDTCisInt<2>]>,
81 [SDNPHasChain, SDNPMayLoad]>;
83 def AMDGPUregister_store : SDNode<"AMDGPUISD::REGISTER_STORE",
84 SDTypeProfile<0, 3, [SDTCisPtrTy<1>, SDTCisInt<2>]>,
85 [SDNPHasChain, SDNPMayStore]>;
87 // MSKOR instructions are atomic memory instructions used mainly for storing
88 // 8-bit and 16-bit values. The definition is:
90 // MSKOR(dst, mask, src) MEM[dst] = ((MEM[dst] & ~mask) | src)
92 // src0: vec4(src, 0, 0, mask)
93 // src1: dst - rat offset (aka pointer) in dwords
94 def AMDGPUstore_mskor : SDNode<"AMDGPUISD::STORE_MSKOR",
95 SDTypeProfile<0, 2, []>,
96 [SDNPHasChain, SDNPMayStore, SDNPMemOperand]>;
98 def AMDGPUround : SDNode<"ISD::FROUND",
99 SDTypeProfile<1, 1, [SDTCisFP<0>, SDTCisSameAs<0,1>]>>;
101 def AMDGPUbfe_u32 : SDNode<"AMDGPUISD::BFE_U32", AMDGPUDTIntTernaryOp>;
102 def AMDGPUbfe_i32 : SDNode<"AMDGPUISD::BFE_I32", AMDGPUDTIntTernaryOp>;
103 def AMDGPUbfi : SDNode<"AMDGPUISD::BFI", AMDGPUDTIntTernaryOp>;
104 def AMDGPUbfm : SDNode<"AMDGPUISD::BFM", SDTIntBinOp>;
106 // Signed and unsigned 24-bit mulitply. The highest 8-bits are ignore when
107 // performing the mulitply. The result is a 32-bit value.
108 def AMDGPUmul_u24 : SDNode<"AMDGPUISD::MUL_U24", SDTIntBinOp,
111 def AMDGPUmul_i24 : SDNode<"AMDGPUISD::MUL_I24", SDTIntBinOp,
115 def AMDGPUmad_u24 : SDNode<"AMDGPUISD::MAD_U24", AMDGPUDTIntTernaryOp,
118 def AMDGPUmad_i24 : SDNode<"AMDGPUISD::MAD_I24", AMDGPUDTIntTernaryOp,