1 //===-- AMDGPUInstrInfo.td - AMDGPU DAG nodes --------------*- tablegen -*-===//
3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4 // See https://llvm.org/LICENSE.txt for license information.
5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
7 //===----------------------------------------------------------------------===//
9 // This file contains DAG node definitions for the AMDGPU target.
11 //===----------------------------------------------------------------------===//
13 //===----------------------------------------------------------------------===//
14 // AMDGPU DAG Profiles
15 //===----------------------------------------------------------------------===//
17 def AMDGPUDTIntTernaryOp : SDTypeProfile<1, 3, [
18 SDTCisSameAs<0, 1>, SDTCisSameAs<0, 2>, SDTCisInt<0>, SDTCisInt<3>
21 def AMDGPULdExpOp : SDTypeProfile<1, 2,
22 [SDTCisSameAs<0, 1>, SDTCisFP<0>, SDTCisInt<2>]
25 def AMDGPUFPClassOp : SDTypeProfile<1, 2,
26 [SDTCisInt<0>, SDTCisFP<1>, SDTCisInt<2>]
29 def AMDGPUFPPackOp : SDTypeProfile<1, 2,
30 [SDTCisFP<1>, SDTCisSameAs<1, 2>]
33 def AMDGPUIntPackOp : SDTypeProfile<1, 2,
34 [SDTCisInt<1>, SDTCisSameAs<1, 2>]
37 def AMDGPUDivScaleOp : SDTypeProfile<2, 3,
38 [SDTCisFP<0>, SDTCisInt<1>, SDTCisSameAs<0, 2>, SDTCisSameAs<0, 3>, SDTCisSameAs<0, 4>]
41 // float, float, float, vcc
42 def AMDGPUFmasOp : SDTypeProfile<1, 4,
43 [SDTCisFP<0>, SDTCisSameAs<0, 1>, SDTCisSameAs<0, 2>, SDTCisSameAs<0, 3>, SDTCisInt<4>]
46 def AMDGPUKillSDT : SDTypeProfile<0, 1, [SDTCisInt<0>]>;
48 def AMDGPUIfOp : SDTypeProfile<1, 2,
49 [SDTCisVT<0, i1>, SDTCisVT<1, i1>, SDTCisVT<2, OtherVT>]
52 def AMDGPUElseOp : SDTypeProfile<1, 2,
53 [SDTCisVT<0, i1>, SDTCisVT<1, i1>, SDTCisVT<2, OtherVT>]
56 def AMDGPULoopOp : SDTypeProfile<0, 2,
57 [SDTCisVT<0, i1>, SDTCisVT<1, OtherVT>]
60 def AMDGPUIfBreakOp : SDTypeProfile<1, 2,
61 [SDTCisVT<0, i1>, SDTCisVT<1, i1>, SDTCisVT<2, i1>]
64 //===----------------------------------------------------------------------===//
68 def AMDGPUif : SDNode<"AMDGPUISD::IF", AMDGPUIfOp, [SDNPHasChain]>;
69 def AMDGPUelse : SDNode<"AMDGPUISD::ELSE", AMDGPUElseOp, [SDNPHasChain]>;
70 def AMDGPUloop : SDNode<"AMDGPUISD::LOOP", AMDGPULoopOp, [SDNPHasChain]>;
72 def callseq_start : SDNode<"ISD::CALLSEQ_START",
73 SDCallSeqStart<[ SDTCisVT<0, i32>, SDTCisVT<1, i32> ]>,
74 [SDNPHasChain, SDNPOutGlue]
77 def callseq_end : SDNode<"ISD::CALLSEQ_END",
78 SDCallSeqEnd<[ SDTCisVT<0, i32>, SDTCisVT<1, i32> ]>,
79 [SDNPHasChain, SDNPOptInGlue, SDNPOutGlue]
82 def AMDGPUcall : SDNode<"AMDGPUISD::CALL",
83 SDTypeProfile<0, -1, [SDTCisPtrTy<0>]>,
84 [SDNPHasChain, SDNPOptInGlue, SDNPOutGlue,
88 def AMDGPUtc_return: SDNode<"AMDGPUISD::TC_RETURN",
89 SDTypeProfile<0, 3, [SDTCisPtrTy<0>]>,
90 [SDNPHasChain, SDNPOptInGlue, SDNPVariadic]
93 def AMDGPUtrap : SDNode<"AMDGPUISD::TRAP",
94 SDTypeProfile<0, -1, [SDTCisVT<0, i16>]>,
95 [SDNPHasChain, SDNPVariadic, SDNPSideEffect, SDNPInGlue]
98 def AMDGPUconstdata_ptr : SDNode<
99 "AMDGPUISD::CONST_DATA_PTR", SDTypeProfile <1, 1, [SDTCisVT<0, iPTR>,
103 // This argument to this node is a dword address.
104 def AMDGPUdwordaddr : SDNode<"AMDGPUISD::DWORDADDR", SDTIntUnaryOp>;
106 def AMDGPUcos_impl : SDNode<"AMDGPUISD::COS_HW", SDTFPUnaryOp>;
107 def AMDGPUsin_impl : SDNode<"AMDGPUISD::SIN_HW", SDTFPUnaryOp>;
108 // out = a - floor(a)
109 def AMDGPUfract_impl : SDNode<"AMDGPUISD::FRACT", SDTFPUnaryOp>;
112 def AMDGPUrcp_impl : SDNode<"AMDGPUISD::RCP", SDTFPUnaryOp>;
114 // out = 1.0 / sqrt(a)
115 def AMDGPUrsq_impl : SDNode<"AMDGPUISD::RSQ", SDTFPUnaryOp>;
117 def AMDGPUrcp_legacy_impl : SDNode<"AMDGPUISD::RCP_LEGACY", SDTFPUnaryOp>;
119 def AMDGPUrcp_iflag : SDNode<"AMDGPUISD::RCP_IFLAG", SDTFPUnaryOp>;
121 // out = 1.0 / sqrt(a) result clamped to +/- max_float.
122 def AMDGPUrsq_clamp_impl : SDNode<"AMDGPUISD::RSQ_CLAMP", SDTFPUnaryOp>;
124 def AMDGPUldexp_impl : SDNode<"AMDGPUISD::LDEXP", AMDGPULdExpOp>;
126 def AMDGPUpkrtz_f16_f32_impl : SDNode<"AMDGPUISD::CVT_PKRTZ_F16_F32", AMDGPUFPPackOp>;
127 def AMDGPUpknorm_i16_f32_impl : SDNode<"AMDGPUISD::CVT_PKNORM_I16_F32", AMDGPUFPPackOp>;
128 def AMDGPUpknorm_u16_f32_impl : SDNode<"AMDGPUISD::CVT_PKNORM_U16_F32", AMDGPUFPPackOp>;
129 def AMDGPUpk_i16_i32_impl : SDNode<"AMDGPUISD::CVT_PK_I16_I32", AMDGPUIntPackOp>;
130 def AMDGPUpk_u16_u32_impl : SDNode<"AMDGPUISD::CVT_PK_U16_U32", AMDGPUIntPackOp>;
131 def AMDGPUfp_to_f16 : SDNode<"AMDGPUISD::FP_TO_FP16" , SDTFPToIntOp>;
134 def AMDGPUfp_class_impl : SDNode<"AMDGPUISD::FP_CLASS", AMDGPUFPClassOp>;
136 // out = max(a, b) a and b are floats, where a nan comparison fails.
137 // This is not commutative because this gives the second operand:
138 // x < nan ? x : nan -> nan
139 // nan < x ? nan : x -> x
140 def AMDGPUfmax_legacy : SDNode<"AMDGPUISD::FMAX_LEGACY", SDTFPBinOp,
144 def AMDGPUfmul_legacy_impl : SDNode<"AMDGPUISD::FMUL_LEGACY", SDTFPBinOp,
145 [SDNPCommutative, SDNPAssociative]
148 // out = min(a, b) a and b are floats, where a nan comparison fails.
149 def AMDGPUfmin_legacy : SDNode<"AMDGPUISD::FMIN_LEGACY", SDTFPBinOp,
153 // FIXME: TableGen doesn't like commutative instructions with more
155 // out = max(a, b, c) a, b and c are floats
156 def AMDGPUfmax3 : SDNode<"AMDGPUISD::FMAX3", SDTFPTernaryOp,
157 [/*SDNPCommutative, SDNPAssociative*/]
160 // out = max(a, b, c) a, b, and c are signed ints
161 def AMDGPUsmax3 : SDNode<"AMDGPUISD::SMAX3", AMDGPUDTIntTernaryOp,
162 [/*SDNPCommutative, SDNPAssociative*/]
165 // out = max(a, b, c) a, b and c are unsigned ints
166 def AMDGPUumax3 : SDNode<"AMDGPUISD::UMAX3", AMDGPUDTIntTernaryOp,
167 [/*SDNPCommutative, SDNPAssociative*/]
170 // out = min(a, b, c) a, b and c are floats
171 def AMDGPUfmin3 : SDNode<"AMDGPUISD::FMIN3", SDTFPTernaryOp,
172 [/*SDNPCommutative, SDNPAssociative*/]
175 // out = min(a, b, c) a, b and c are signed ints
176 def AMDGPUsmin3 : SDNode<"AMDGPUISD::SMIN3", AMDGPUDTIntTernaryOp,
177 [/*SDNPCommutative, SDNPAssociative*/]
180 // out = min(a, b) a and b are unsigned ints
181 def AMDGPUumin3 : SDNode<"AMDGPUISD::UMIN3", AMDGPUDTIntTernaryOp,
182 [/*SDNPCommutative, SDNPAssociative*/]
185 // out = (src0 + src1 > 0xFFFFFFFF) ? 1 : 0
186 def AMDGPUcarry : SDNode<"AMDGPUISD::CARRY", SDTIntBinOp, []>;
188 // out = (src1 > src0) ? 1 : 0
189 def AMDGPUborrow : SDNode<"AMDGPUISD::BORROW", SDTIntBinOp, []>;
191 def AMDGPUSetCCOp : SDTypeProfile<1, 3, [ // setcc
192 SDTCisInt<0>, SDTCisSameAs<1, 2>, SDTCisVT<3, OtherVT>
195 def AMDGPUsetcc : SDNode<"AMDGPUISD::SETCC", AMDGPUSetCCOp>;
197 def AMDGPUfma : SDNode<"AMDGPUISD::FMA_W_CHAIN", SDTFPTernaryOp, [
198 SDNPHasChain, SDNPOptInGlue, SDNPOutGlue]>;
200 def AMDGPUmul : SDNode<"AMDGPUISD::FMUL_W_CHAIN", SDTFPBinOp, [
201 SDNPHasChain, SDNPOptInGlue, SDNPOutGlue]>;
203 def AMDGPUcvt_f32_ubyte0 : SDNode<"AMDGPUISD::CVT_F32_UBYTE0",
205 def AMDGPUcvt_f32_ubyte1 : SDNode<"AMDGPUISD::CVT_F32_UBYTE1",
207 def AMDGPUcvt_f32_ubyte2 : SDNode<"AMDGPUISD::CVT_F32_UBYTE2",
209 def AMDGPUcvt_f32_ubyte3 : SDNode<"AMDGPUISD::CVT_F32_UBYTE3",
212 def AMDGPUcvt_pk_i16_i32 : SDNode<"AMDGPUISD::CVT_PK_I16_I32",
213 AMDGPUIntPackOp, []>;
215 // urecip - This operation is a helper for integer division, it returns the
216 // result of 1 / a as a fractional unsigned integer.
217 // out = (2^32 / a) + e
218 // e is rounding error
219 def AMDGPUurecip : SDNode<"AMDGPUISD::URECIP", SDTIntUnaryOp>;
221 // Special case divide preop and flags.
222 def AMDGPUdiv_scale : SDNode<"AMDGPUISD::DIV_SCALE", AMDGPUDivScaleOp>;
224 // Special case divide FMA with scale and flags (src0 = Quotient,
225 // src1 = Denominator, src2 = Numerator).
226 def AMDGPUdiv_fmas_impl : SDNode<"AMDGPUISD::DIV_FMAS", AMDGPUFmasOp,
229 // Single or double precision division fixup.
230 // Special case divide fixup and flags(src0 = Quotient, src1 =
231 // Denominator, src2 = Numerator).
232 def AMDGPUdiv_fixup_impl : SDNode<"AMDGPUISD::DIV_FIXUP", SDTFPTernaryOp>;
234 def AMDGPUfmad_ftz_impl : SDNode<"AMDGPUISD::FMAD_FTZ", SDTFPTernaryOp>;
236 def AMDGPUregister_load : SDNode<"AMDGPUISD::REGISTER_LOAD",
237 SDTypeProfile<1, 2, [SDTCisPtrTy<1>, SDTCisInt<2>]>,
238 [SDNPHasChain, SDNPMayLoad]>;
240 def AMDGPUregister_store : SDNode<"AMDGPUISD::REGISTER_STORE",
241 SDTypeProfile<0, 3, [SDTCisPtrTy<1>, SDTCisInt<2>]>,
242 [SDNPHasChain, SDNPMayStore]>;
244 // MSKOR instructions are atomic memory instructions used mainly for storing
245 // 8-bit and 16-bit values. The definition is:
247 // MSKOR(dst, mask, src) MEM[dst] = ((MEM[dst] & ~mask) | src)
249 // src0: vec4(src, 0, 0, mask)
250 // src1: dst - rat offset (aka pointer) in dwords
251 def AMDGPUstore_mskor : SDNode<"AMDGPUISD::STORE_MSKOR",
252 SDTypeProfile<0, 2, []>,
253 [SDNPHasChain, SDNPMayStore, SDNPMemOperand]>;
255 def AMDGPUatomic_cmp_swap : SDNode<"AMDGPUISD::ATOMIC_CMP_SWAP",
256 SDTypeProfile<1, 2, [SDTCisPtrTy<1>, SDTCisVec<2>]>,
257 [SDNPHasChain, SDNPMayStore, SDNPMayLoad,
260 def AMDGPUround : SDNode<"ISD::FROUND",
261 SDTypeProfile<1, 1, [SDTCisFP<0>, SDTCisSameAs<0,1>]>>;
263 def AMDGPUbfe_u32_impl : SDNode<"AMDGPUISD::BFE_U32", AMDGPUDTIntTernaryOp>;
264 def AMDGPUbfe_i32_impl : SDNode<"AMDGPUISD::BFE_I32", AMDGPUDTIntTernaryOp>;
265 def AMDGPUbfi : SDNode<"AMDGPUISD::BFI", AMDGPUDTIntTernaryOp>;
266 def AMDGPUbfm : SDNode<"AMDGPUISD::BFM", SDTIntBinOp>;
268 def AMDGPUffbh_u32_impl : SDNode<"AMDGPUISD::FFBH_U32", SDTIntBitCountUnaryOp>;
269 def AMDGPUffbh_i32_impl : SDNode<"AMDGPUISD::FFBH_I32", SDTIntBitCountUnaryOp>;
271 def AMDGPUffbl_b32_impl : SDNode<"AMDGPUISD::FFBL_B32", SDTIntBitCountUnaryOp>;
273 // Signed and unsigned 24-bit multiply. The highest 8-bits are ignore
274 // when performing the multiply. The result is a 32-bit value.
275 def AMDGPUmul_u24_impl : SDNode<"AMDGPUISD::MUL_U24", SDTIntBinOp,
276 [SDNPCommutative, SDNPAssociative]
278 def AMDGPUmul_i24_impl : SDNode<"AMDGPUISD::MUL_I24", SDTIntBinOp,
279 [SDNPCommutative, SDNPAssociative]
282 // mulhi24 yields the high-order 16 bits of the 48-bit result. Here's an example
283 // that shows mulhi24 is not associative:
285 // Given a = 0x10002, b = c = 0xffffff:
286 // mulhi24(mulhi24(a, b), c) = mulhi24(0x100, 0xffffff) = 0
287 // Which is not equal to:
288 // mulhi24(a, mulhi24(b, c)) = mulhi24(0x10002, 0xffff) = 1
289 def AMDGPUmulhi_u24_impl : SDNode<"AMDGPUISD::MULHI_U24", SDTIntBinOp,
292 def AMDGPUmulhi_i24_impl : SDNode<"AMDGPUISD::MULHI_I24", SDTIntBinOp,
296 def AMDGPUmad_u24 : SDNode<"AMDGPUISD::MAD_U24", AMDGPUDTIntTernaryOp,
299 def AMDGPUmad_i24 : SDNode<"AMDGPUISD::MAD_I24", AMDGPUDTIntTernaryOp,
303 def AMDGPUsmed3 : SDNode<"AMDGPUISD::SMED3", AMDGPUDTIntTernaryOp,
307 def AMDGPUumed3 : SDNode<"AMDGPUISD::UMED3", AMDGPUDTIntTernaryOp,
311 def AMDGPUfmed3_impl : SDNode<"AMDGPUISD::FMED3", SDTFPTernaryOp, []>;
313 def AMDGPUfdot2_impl : SDNode<"AMDGPUISD::FDOT2",
314 SDTypeProfile<1, 4, [SDTCisSameAs<0, 3>, SDTCisSameAs<1, 2>,
315 SDTCisFP<0>, SDTCisVec<1>,
319 def AMDGPUperm_impl : SDNode<"AMDGPUISD::PERM", AMDGPUDTIntTernaryOp, []>;
322 def AMDGPUExportOp : SDTypeProfile<0, 8, [
323 SDTCisInt<0>, // i8 tgt
324 SDTCisInt<1>, // i8 en
326 SDTCisSameAs<3, 2>, // f32 src1
327 SDTCisSameAs<4, 2>, // f32 src2
328 SDTCisSameAs<5, 2>, // f32 src3
329 SDTCisInt<6>, // i1 compr
331 SDTCisInt<1> // i1 vm
336 //===----------------------------------------------------------------------===//
337 // Flow Control Profile Types
338 //===----------------------------------------------------------------------===//
339 // Branch instruction where second and third are basic blocks
340 def SDTIL_BRCond : SDTypeProfile<0, 2, [
344 //===----------------------------------------------------------------------===//
345 // Flow Control DAG Nodes
346 //===----------------------------------------------------------------------===//
347 def IL_brcond : SDNode<"AMDGPUISD::BRANCH_COND", SDTIL_BRCond, [SDNPHasChain]>;
349 //===----------------------------------------------------------------------===//
350 // Call/Return DAG Nodes
351 //===----------------------------------------------------------------------===//
352 def AMDGPUendpgm : SDNode<"AMDGPUISD::ENDPGM", SDTNone,
353 [SDNPHasChain, SDNPOptInGlue]>;
355 def AMDGPUreturn_to_epilog : SDNode<"AMDGPUISD::RETURN_TO_EPILOG", SDTNone,
356 [SDNPHasChain, SDNPOptInGlue, SDNPVariadic]>;
358 def AMDGPUret_flag : SDNode<"AMDGPUISD::RET_FLAG", SDTNone,
359 [SDNPHasChain, SDNPOptInGlue, SDNPVariadic]
363 //===----------------------------------------------------------------------===//
364 // Intrinsic/Custom node compatibility PatFrags
365 //===----------------------------------------------------------------------===//
367 def AMDGPUrcp : PatFrags<(ops node:$src), [(int_amdgcn_rcp node:$src),
368 (AMDGPUrcp_impl node:$src)]>;
369 def AMDGPUrcp_legacy : PatFrags<(ops node:$src), [(int_amdgcn_rcp_legacy node:$src),
370 (AMDGPUrcp_legacy_impl node:$src)]>;
372 def AMDGPUrsq : PatFrags<(ops node:$src), [(int_amdgcn_rsq node:$src),
373 (AMDGPUrsq_impl node:$src)]>;
375 def AMDGPUrsq_clamp : PatFrags<(ops node:$src), [(int_amdgcn_rsq_clamp node:$src),
376 (AMDGPUrsq_clamp_impl node:$src)]>;
378 def AMDGPUsin : PatFrags<(ops node:$src), [(int_amdgcn_sin node:$src),
379 (AMDGPUsin_impl node:$src)]>;
380 def AMDGPUcos : PatFrags<(ops node:$src), [(int_amdgcn_cos node:$src),
381 (AMDGPUcos_impl node:$src)]>;
382 def AMDGPUfract : PatFrags<(ops node:$src), [(int_amdgcn_fract node:$src),
383 (AMDGPUfract_impl node:$src)]>;
385 def AMDGPUldexp : PatFrags<(ops node:$src0, node:$src1),
386 [(int_amdgcn_ldexp node:$src0, node:$src1),
387 (AMDGPUldexp_impl node:$src0, node:$src1)]>;
389 def AMDGPUfp_class : PatFrags<(ops node:$src0, node:$src1),
390 [(int_amdgcn_class node:$src0, node:$src1),
391 (AMDGPUfp_class_impl node:$src0, node:$src1)]>;
393 def AMDGPUfmed3 : PatFrags<(ops node:$src0, node:$src1, node:$src2),
394 [(int_amdgcn_fmed3 node:$src0, node:$src1, node:$src2),
395 (AMDGPUfmed3_impl node:$src0, node:$src1, node:$src2)]>;
397 def AMDGPUdiv_fixup : PatFrags<(ops node:$src0, node:$src1, node:$src2),
398 [(int_amdgcn_div_fixup node:$src0, node:$src1, node:$src2),
399 (AMDGPUdiv_fixup_impl node:$src0, node:$src1, node:$src2)]>;
401 def AMDGPUffbh_i32 : PatFrags<(ops node:$src),
402 [(int_amdgcn_sffbh node:$src),
403 (AMDGPUffbh_i32_impl node:$src)]>;
405 def AMDGPUffbh_u32 : PatFrags<(ops node:$src),
406 [(ctlz_zero_undef node:$src),
407 (AMDGPUffbh_u32_impl node:$src)]>;
409 def AMDGPUffbl_b32 : PatFrags<(ops node:$src),
410 [(cttz_zero_undef node:$src),
411 (AMDGPUffbl_b32_impl node:$src)]>;
413 def AMDGPUpkrtz_f16_f32 : PatFrags<(ops node:$src0, node:$src1),
414 [(int_amdgcn_cvt_pkrtz node:$src0, node:$src1),
415 (AMDGPUpkrtz_f16_f32_impl node:$src0, node:$src1)]>;
417 def AMDGPUpknorm_i16_f32 : PatFrags<(ops node:$src0, node:$src1),
418 [(int_amdgcn_cvt_pknorm_i16 node:$src0, node:$src1),
419 (AMDGPUpknorm_i16_f32_impl node:$src0, node:$src1)]>;
421 def AMDGPUpknorm_u16_f32 : PatFrags<(ops node:$src0, node:$src1),
422 [(int_amdgcn_cvt_pknorm_u16 node:$src0, node:$src1),
423 (AMDGPUpknorm_u16_f32_impl node:$src0, node:$src1)]>;
425 def AMDGPUpk_i16_i32 : PatFrags<(ops node:$src0, node:$src1),
426 [(int_amdgcn_cvt_pk_i16 node:$src0, node:$src1),
427 (AMDGPUpk_i16_i32_impl node:$src0, node:$src1)]>;
429 def AMDGPUpk_u16_u32 : PatFrags<(ops node:$src0, node:$src1),
430 [(int_amdgcn_cvt_pk_u16 node:$src0, node:$src1),
431 (AMDGPUpk_u16_u32_impl node:$src0, node:$src1)]>;
433 def AMDGPUfmad_ftz : PatFrags<(ops node:$src0, node:$src1, node:$src2),
434 [(int_amdgcn_fmad_ftz node:$src0, node:$src1, node:$src2),
435 (AMDGPUfmad_ftz_impl node:$src0, node:$src1, node:$src2)]>;
437 def AMDGPUmul_u24 : PatFrags<(ops node:$src0, node:$src1),
438 [(int_amdgcn_mul_u24 node:$src0, node:$src1),
439 (AMDGPUmul_u24_impl node:$src0, node:$src1)]>;
441 def AMDGPUmul_i24 : PatFrags<(ops node:$src0, node:$src1),
442 [(int_amdgcn_mul_i24 node:$src0, node:$src1),
443 (AMDGPUmul_i24_impl node:$src0, node:$src1)]>;
445 def AMDGPUmulhi_u24 : PatFrags<(ops node:$src0, node:$src1),
446 [(int_amdgcn_mulhi_u24 node:$src0, node:$src1),
447 (AMDGPUmulhi_u24_impl node:$src0, node:$src1)]>;
449 def AMDGPUmulhi_i24 : PatFrags<(ops node:$src0, node:$src1),
450 [(int_amdgcn_mulhi_i24 node:$src0, node:$src1),
451 (AMDGPUmulhi_i24_impl node:$src0, node:$src1)]>;
453 def AMDGPUbfe_i32 : PatFrags<(ops node:$src0, node:$src1, node:$src2),
454 [(int_amdgcn_sbfe node:$src0, node:$src1, node:$src2),
455 (AMDGPUbfe_i32_impl node:$src0, node:$src1, node:$src2)]>;
457 def AMDGPUbfe_u32 : PatFrags<(ops node:$src0, node:$src1, node:$src2),
458 [(int_amdgcn_ubfe node:$src0, node:$src1, node:$src2),
459 (AMDGPUbfe_u32_impl node:$src0, node:$src1, node:$src2)]>;
461 def AMDGPUfmul_legacy : PatFrags<(ops node:$src0, node:$src1),
462 [(int_amdgcn_fmul_legacy node:$src0, node:$src1),
463 (AMDGPUfmul_legacy_impl node:$src0, node:$src1)]>;
465 def AMDGPUfdot2 : PatFrags<(ops node:$src0, node:$src1, node:$src2, node:$clamp),
466 [(int_amdgcn_fdot2 node:$src0, node:$src1, node:$src2, node:$clamp),
467 (AMDGPUfdot2_impl node:$src0, node:$src1, node:$src2, node:$clamp)]>;
469 def AMDGPUdiv_fmas : PatFrags<(ops node:$src0, node:$src1, node:$src2, node:$vcc),
470 [(int_amdgcn_div_fmas node:$src0, node:$src1, node:$src2, node:$vcc),
471 (AMDGPUdiv_fmas_impl node:$src0, node:$src1, node:$src2, node:$vcc)]>;
473 def AMDGPUperm : PatFrags<(ops node:$src0, node:$src1, node:$src2),
474 [(int_amdgcn_perm node:$src0, node:$src1, node:$src2),
475 (AMDGPUperm_impl node:$src0, node:$src1, node:$src2)]>;