diff options
Diffstat (limited to 'src/gallium/drivers/radeon/AMDILInstructions.td')
| -rw-r--r-- | src/gallium/drivers/radeon/AMDILInstructions.td | 2436 |
1 files changed, 2436 insertions, 0 deletions
diff --git a/src/gallium/drivers/radeon/AMDILInstructions.td b/src/gallium/drivers/radeon/AMDILInstructions.td new file mode 100644 index 00000000000..f824a67d7ad --- /dev/null +++ b/src/gallium/drivers/radeon/AMDILInstructions.td @@ -0,0 +1,2436 @@ +//===-- AMDILInstructions.td - TODO: Add brief description -------===// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//==-----------------------------------------------------------------------===// + +// Operations in this file are generic to all data types +// This opcode has custom swizzle pattern encoded in Swizzle Encoder +let isReMaterializable = 1, isAsCheapAsAMove = 1 in { + defm LOADCONST : ILConstant<"mov $dst, $val">; + defm MOVE : UnaryOpMC<IL_OP_MOV, IL_mov>; + defm PHIMOVE : UnaryOpMC<IL_OP_MOV, IL_phimov>; +} +defm BINARY_NOT : UnaryOpMC<IL_OP_I_NOT, IL_not>; +defm BINARY_OR : BinaryOpMC<IL_OP_I_OR, IL_or>; +defm BINARY_AND : BinaryOpMC<IL_OP_AND, IL_and>; +defm BINARY_XOR : BinaryOpMC<IL_OP_I_XOR, IL_xor>; +defm AND : BinaryOpMCInt<IL_OP_AND, and>; +defm CMOV : BinaryOpMC<IL_OP_CMOV, IL_cmov>; +defm DIV_INF : BinaryOpMC<IL_OP_DIV_INF, IL_div_inf>; +defm SMAX : BinaryOpMCInt<IL_OP_I_MAX, IL_smax>; +// This opcode has custom swizzle pattern encoded in Swizzle Encoder for 64bit +// instructions +defm CMOVLOG : TernaryOpMC<IL_OP_CMOV_LOGICAL, IL_cmov_logical>; +// This opcode has a custom swizzle pattern in the Swizzle Encoder and +// should never be selected in ISel. It should only be generated in the +// I/O expansion code. These are different from the CMOVLOG instruction +// in that the src0 argument uses a custom swizzle for the Y/Z/W +// vector channel respectively instead of the default channel. +def CMOVLOG_Y_i32 : ThreeInOneOut<IL_OP_CMOV_LOGICAL, (outs GPRI32:$dst), + (ins GPRI32:$src0, GPRI32:$src1, GPRI32:$src2), + !strconcat(IL_OP_CMOV_LOGICAL.Text, " $dst, $src0, $src1, $src2"), + []>; +def CMOVLOG_Z_i32 : ThreeInOneOut<IL_OP_CMOV_LOGICAL, (outs GPRI32:$dst), + (ins GPRI32:$src0, GPRI32:$src1, GPRI32:$src2), + !strconcat(IL_OP_CMOV_LOGICAL.Text, " $dst, $src0, $src1, $src2"), + []>; +def CMOVLOG_W_i32 : ThreeInOneOut<IL_OP_CMOV_LOGICAL, (outs GPRI32:$dst), + (ins GPRI32:$src0, GPRI32:$src1 ,GPRI32:$src2), + !strconcat(IL_OP_CMOV_LOGICAL.Text, " $dst, $src0, $src1, $src2"), + []>; +defm SELECTBIN : TernaryOpMCScalar<IL_OP_CMOV_LOGICAL, select>; +//===---------------------------------------------------------------------===// +// Signed 8bit integer math instructions start here +//===---------------------------------------------------------------------===// +def INTTOANY_i8 : OneInOneOut<IL_OP_MOV, (outs GPRI8:$dst), (ins GPRI32:$src0), + !strconcat(IL_OP_MOV.Text, " $dst, $src0"), + [(set GPRI8:$dst, (IL_inttoany GPRI32:$src0))]>; +//===---------------------------------------------------------------------===// +// Signed 16bit integer math instructions start here +//===---------------------------------------------------------------------===// +def INTTOANY_i16: OneInOneOut<IL_OP_MOV, (outs GPRI16:$dst), (ins GPRI32:$src0), + !strconcat(IL_OP_MOV.Text," $dst, $src0"), + [(set GPRI16:$dst, (IL_inttoany GPRI32:$src0))]>; +//===---------------------------------------------------------------------===// +// Signed 32bit integer math instructions start here +//===---------------------------------------------------------------------===// +defm NEGATE : UnaryOpMCi32<IL_OP_I_NEGATE, IL_inegate>; +defm SMUL : BinaryOpMCi32<IL_OP_I_MUL, mul>; +defm SMULHI : BinaryOpMCi32<IL_OP_I_MUL_HIGH, mulhs>; +defm SHL : BinaryOpMCi32Const<IL_OP_I_SHL, shl>; +defm SHR : BinaryOpMCi32Const<IL_OP_I_SHR, sra>; +defm SHLVEC : BinaryOpMCi32<IL_OP_I_SHL, shl>; +defm SHRVEC : BinaryOpMCi32<IL_OP_I_SHR, sra>; +defm ADD : BinaryOpMCi32<IL_OP_I_ADD, add>; +defm CUSTOM_XOR : BinaryOpMCInt<IL_OP_I_XOR, xor>; +// get rid of the addri via the tablegen instead of custom lowered instruction +defm CUSTOM_ADD : BinaryOpMCi32<IL_OP_I_ADD, IL_add>; +defm EADD : BinaryOpMCi32<IL_OP_I_ADD, adde>; +def INTTOANY_i32: OneInOneOut<IL_OP_MOV, (outs GPRI32:$dst), (ins GPRI32:$src0), + !strconcat(IL_OP_MOV.Text, " $dst, $src0"), + [(set GPRI32:$dst, (IL_inttoany GPRI32:$src0))]>; +// Integer offsets for addressing +// This opcode has custom swizzle pattern encoded in Swizzle Encoder +def ADDir : TwoInOneOut<IL_OP_I_ADD, (outs GPRI32:$dst), + (ins MEMI32:$ptr, GPRI32:$offset), + !strconcat(IL_OP_I_ADD.Text, " $dst, $ptr, $offset"), + [(set GPRI32:$dst, + (IL_addaddrri ADDR:$ptr, + (i32 GPRI32:$offset)))]>; +// This opcode has custom swizzle pattern encoded in Swizzle Encoder +def ADDri : TwoInOneOut<IL_OP_I_ADD, (outs GPRI32:$dst), + (ins GPRI32:$offset, MEMI32:$ptr), + !strconcat(IL_OP_I_ADD.Text, " $dst, $offset, $ptr"), + [(set GPRI32:$dst, + (IL_addaddrir + (i32 GPRI32:$offset), ADDR:$ptr))]>; + +defm IFFB_HI : UnaryOpMCi32<IL_OP_I_FFB_HI, IL_ffb_hi>; +defm IFFB_LO : UnaryOpMCi32<IL_OP_I_FFB_LO, IL_ffb_lo>; +let mayLoad = 0, mayStore = 0 in { +defm ABS : UnaryIntrinsicInt<IL_OP_ABS, int_AMDIL_abs>; +defm BITCOUNT : UnaryIntrinsicInt<IL_OP_IBIT_COUNT, int_AMDIL_bit_count_i32>; +defm FFB_LO : UnaryIntrinsicInt<IL_OP_I_FFB_LO, int_AMDIL_bit_find_first_lo>; +defm FFB_HI : UnaryIntrinsicInt<IL_OP_I_FFB_HI, int_AMDIL_bit_find_first_hi>; +defm FFB_SGN : UnaryIntrinsicInt<IL_OP_I_FFB_SGN, + int_AMDIL_bit_find_first_sgn>; +defm IMULHI : BinaryIntrinsicInt<IL_OP_I_MUL_HIGH, int_AMDIL_mulhi_i32>; +let Predicates = [HasHWSign24Bit] in { +defm IMUL24 : BinaryIntrinsicInt<IL_OP_I_MUL24, int_AMDIL_mul24_i32>; +defm IMULHI24 : BinaryIntrinsicInt<IL_OP_I_MULHI24, int_AMDIL_mulhi24_i32>; +defm IMAD24 : TernaryIntrinsicInt<IL_OP_I_MAD24, int_AMDIL_mad24_i32>; +} +defm CARRY : BinaryIntrinsicInt<IL_OP_I_CARRY, int_AMDIL_carry_i32>; +defm BORROW : BinaryIntrinsicInt<IL_OP_I_BORROW, int_AMDIL_borrow_i32>; +defm IMIN : BinaryIntrinsicInt<IL_OP_I_MIN, int_AMDIL_min_i32>; +defm IMAX : BinaryIntrinsicInt<IL_OP_I_MAX, int_AMDIL_max_i32>; +defm CMOV_LOG : TernaryIntrinsicInt<IL_OP_CMOV_LOGICAL, + int_AMDIL_cmov_logical>; +defm IBIT_EXTRACT : TernaryIntrinsicInt<IL_OP_IBIT_EXTRACT, + int_AMDIL_bit_extract_i32>; +defm IMAD : TernaryIntrinsicInt<IL_OP_I_MAD, int_AMDIL_mad_i32>; +defm SAD : TernaryIntrinsicInt<IL_OP_SAD, int_AMDIL_media_sad>; +defm SADHI : TernaryIntrinsicInt<IL_OP_SAD_HI, + int_AMDIL_media_sad_hi>; +} +def SAD4_i32 : ThreeInOneOut<IL_OP_SAD4, (outs GPRI32:$dst), + (ins GPRV4I32:$src, GPRV4I32:$src1, GPRI32:$src2), + !strconcat(IL_OP_SAD4.Text, " $dst, $src, $src1, $src2"), + [(set GPRI32:$dst, + (int_AMDIL_media_sad4 GPRV4I32:$src, GPRV4I32:$src1, + GPRI32:$src2))]>; +def FTOV4U8_i32 : OneInOneOut<IL_OP_F2U4, (outs GPRI32:$dst), + (ins GPRV4F32:$src), + !strconcat(IL_OP_F2U4.Text, " $dst, $src"), + [(set GPRI32:$dst, + (int_AMDIL_media_convert_f2v4u8 GPRV4F32:$src))]>; +//===---------------------------------------------------------------------===// +// Unsigned 32bit integer math instructions start here +//===---------------------------------------------------------------------===// +defm UMUL : BinaryOpMCi32<IL_OP_U_MUL, IL_umul>; +defm UMULHI : BinaryOpMCi32<IL_OP_U_MUL_HIGH, mulhu>; +defm USHR : BinaryOpMCi32Const<IL_OP_U_SHR, srl>; +defm USHRVEC : BinaryOpMCi32<IL_OP_U_SHR, srl>; +defm UDIV : BinaryOpMCi32<IL_OP_U_DIV, udiv>; +defm NATIVE_UDIV : BinaryIntrinsicInt<IL_OP_U_DIV, int_AMDIL_udiv>; +let mayLoad=0, mayStore=0 in { +defm UBIT_REVERSE : UnaryIntrinsicInt<IL_OP_UBIT_REVERSE, + int_AMDIL_bit_reverse_u32>; +defm UMULHI_INT : BinaryIntrinsicInt<IL_OP_U_MUL_HIGH, int_AMDIL_mulhi_u32>; +defm UMULHI24 : BinaryIntrinsicInt<IL_OP_U_MULHI24, int_AMDIL_mulhi24_u32>; +defm UMUL24 : BinaryIntrinsicInt<IL_OP_U_MUL24, int_AMDIL_mul24_u32>; +defm UMIN : BinaryIntrinsicInt<IL_OP_U_MIN, int_AMDIL_min_u32>; +defm UMAX : BinaryIntrinsicInt<IL_OP_U_MAX, int_AMDIL_max_u32>; +defm UBIT_EXTRACT : TernaryIntrinsicInt<IL_OP_UBIT_EXTRACT, + int_AMDIL_bit_extract_u32>; +defm UBIT_INSERT : QuaternaryIntrinsicInt<IL_OP_UBIT_INSERT, + int_AMDIL_bit_insert_u32>; +defm BFI : TernaryIntrinsicInt<IL_OP_BFI, int_AMDIL_bfi>; +defm BFM : BinaryIntrinsicInt<IL_OP_BFM, int_AMDIL_bfm>; +defm UMAD : TernaryIntrinsicInt<IL_OP_U_MAD, int_AMDIL_mad_u32>; +defm UMAD24 : TernaryIntrinsicInt<IL_OP_U_MAD24, int_AMDIL_mad24_u32>; +defm U4LERP : TernaryIntrinsicInt<IL_OP_U4_LERP, + int_AMDIL_media_lerp_u4>; +defm BITALIGN : TernaryIntrinsicInt<IL_OP_BIT_ALIGN, int_AMDIL_media_bitalign>; +defm BYTEALIGN : TernaryIntrinsicInt<IL_OP_BYTE_ALIGN, int_AMDIL_media_bytealign>; +} +//===---------------------------------------------------------------------===// +// Signed 64bit integer math instructions start here +//===---------------------------------------------------------------------===// +// This opcode has custom swizzle pattern encoded in Swizzle Encoder +def LNEGATE : OneInOneOut<IL_OP_MOV, (outs GPRI64:$dst), (ins GPRI64:$src), + !strconcat(IL_OP_MOV.Text, " $dst, $src"), + [(set GPRI64:$dst, (IL_inegate GPRI64:$src))]>; +// This opcode has custom swizzle pattern encoded in Swizzle Encoder +def LNEGATE_v2i64: OneInOneOut<IL_OP_MOV, (outs GPRV2I64:$dst), + (ins GPRV2I64:$src), + !strconcat(IL_OP_MOV.Text, " $dst, $src"), + [(set GPRV2I64:$dst, (IL_inegate GPRV2I64:$src))]>; +let Predicates = [HasHW64Bit] in { +def LADD : TwoInOneOut<IL_OP_I64_ADD, (outs GPRI64:$dst), + (ins GPRI64:$src1, GPRI64:$src2), + !strconcat(IL_OP_I64_ADD.Text, " $dst, $src1, $src2"), + [(set GPRI64:$dst, (IL_add GPRI64:$src1, GPRI64:$src2))]>; +defm IMIN64 : BinaryIntrinsicLong<IL_OP_I64_MIN, int_AMDIL_min_i32>; +defm UMIN64 : BinaryIntrinsicLong<IL_OP_U64_MIN, int_AMDIL_min_u32>; +defm IMAX64 : BinaryIntrinsicLong<IL_OP_I64_MAX, int_AMDIL_max_i32>; +defm UMAX64 : BinaryIntrinsicLong<IL_OP_U64_MAX, int_AMDIL_max_u32>; +} +let Predicates = [HasHW64Bit] in { +def LSHR : TwoInOneOut<IL_OP_I64_SHR, (outs GPRI64:$dst), + (ins GPRI64:$src1, GPRI32:$src2), + !strconcat(IL_OP_I64_SHR.Text, " $dst, $src1, $src2"), + [(set GPRI64:$dst, (sra GPRI64:$src1, GPRI32:$src2))]>; +def LSHL : TwoInOneOut<IL_OP_I64_SHL, (outs GPRI64:$dst), + (ins GPRI64:$src1, GPRI32:$src2), + !strconcat(IL_OP_I64_SHL.Text, " $dst, $src1, $src2"), + [(set GPRI64:$dst, (shl GPRI64:$src1, GPRI32:$src2))]>; +} + + +//===---------------------------------------------------------------------===// +// Unsigned 64bit integer math instructions start here +//===---------------------------------------------------------------------===// +let Predicates = [HasTmrRegister] in { + def Tmr : ILFormat<IL_OP_MOV, (outs GPRI64:$tmr), + (ins), !strconcat(IL_OP_MOV.Text, " $tmr, Tmr"), + [(set GPRI64:$tmr, (int_AMDIL_get_cycle_count))]>; +} +let Predicates = [HasDeviceIDInst] in { +def CU_ID : ILFormat<IL_OP_CU_ID, (outs GPRI32:$id), (ins), + !strconcat(IL_OP_CU_ID.Text, " $id"), + [(set GPRI32:$id, (int_AMDIL_compute_unit_id))]>; +def WAVE_ID : ILFormat<IL_OP_WAVE_ID, (outs GPRI32:$id), (ins), + !strconcat(IL_OP_WAVE_ID.Text, " $id"), + [(set GPRI32:$id, (int_AMDIL_wavefront_id))]>; +} +let Predicates = [HasHW64Bit] in { +def LUSHR : TwoInOneOut<IL_OP_U64_SHR, (outs GPRI64:$dst), + (ins GPRI64:$src1, GPRI32:$src2), + !strconcat(IL_OP_U64_SHR.Text, " $dst, $src1, $src2"), + [(set GPRI64:$dst, (srl GPRI64:$src1, GPRI32:$src2))]>; +} + + +//===---------------------------------------------------------------------===// +// Generic Float Instructions +//===---------------------------------------------------------------------===// +let hasIEEEFlag = 1 in { +defm MUL_IEEE : BinaryOpMCFloat<IL_OP_MUL_IEEE, IL_OP_D_MUL, fmul>; +} +defm ADD : BinaryOpMCFloat<IL_OP_ADD, IL_OP_D_ADD, fadd>; +//===---------------------------------------------------------------------===// +// float math instructions start here +//===---------------------------------------------------------------------===// +let mayLoad=0, mayStore=0 in { +defm ABS : UnaryIntrinsicFloat<IL_OP_ABS, int_AMDIL_fabs>; +defm FRAC : UnaryIntrinsicFloat<IL_OP_FRC, int_AMDIL_fraction>; +defm PIREDUCE : UnaryIntrinsicFloat<IL_OP_PI_REDUCE, int_AMDIL_pireduce>; +defm ROUND_NEAREST : UnaryIntrinsicFloat<IL_OP_ROUND_NEAR, + int_AMDIL_round_nearest>; +defm ROUND_NEGINF : UnaryIntrinsicFloat<IL_OP_ROUND_NEG_INF, + int_AMDIL_round_neginf>; +defm ROUND_POSINF : UnaryIntrinsicFloat<IL_OP_ROUND_POS_INF, + int_AMDIL_round_posinf>; +defm ROUND_ZERO : UnaryIntrinsicFloat<IL_OP_ROUND_ZERO, + int_AMDIL_round_zero>; +defm ACOS : UnaryIntrinsicFloatScalar<IL_OP_ACOS, int_AMDIL_acos>; +defm ATAN : UnaryIntrinsicFloatScalar<IL_OP_ATAN, int_AMDIL_atan>; +defm ASIN : UnaryIntrinsicFloatScalar<IL_OP_ASIN, int_AMDIL_asin>; +defm TAN : UnaryIntrinsicFloatScalar<IL_OP_TAN, int_AMDIL_tan>; +defm SIN : UnaryIntrinsicFloatScalar<IL_OP_SIN, int_AMDIL_sin>; +defm COS : UnaryIntrinsicFloatScalar<IL_OP_COS, int_AMDIL_cos>; +defm SQRT : UnaryIntrinsicFloatScalar<IL_OP_SQRT, int_AMDIL_sqrt>; +defm EXP : UnaryIntrinsicFloatScalar<IL_OP_EXP, int_AMDIL_exp>; +defm EXPVEC : UnaryIntrinsicFloat<IL_OP_EXP_VEC, int_AMDIL_exp_vec>; +defm SQRTVEC : UnaryIntrinsicFloat<IL_OP_SQRT_VEC, int_AMDIL_sqrt_vec>; +defm COSVEC : UnaryIntrinsicFloat<IL_OP_COS_VEC, int_AMDIL_cos_vec>; +defm SINVEC : UnaryIntrinsicFloat<IL_OP_SIN_VEC, int_AMDIL_sin_vec>; +defm LOGVEC : UnaryIntrinsicFloat<IL_OP_LOG_VEC, int_AMDIL_log_vec>; +defm RSQVEC : UnaryIntrinsicFloat<IL_OP_RSQ_VEC, int_AMDIL_rsq_vec>; +defm EXN : UnaryIntrinsicFloatScalar<IL_OP_EXN, int_AMDIL_exn>; +defm SIGN : UnaryIntrinsicFloat<IL_OP_SGN, int_AMDIL_sign>; +defm LENGTH : UnaryIntrinsicFloat<IL_OP_LEN, int_AMDIL_length>; +defm POW : BinaryIntrinsicFloat<IL_OP_POW, int_AMDIL_pow>; +} + +let hasIEEEFlag = 1 in { + let mayLoad = 0, mayStore=0 in { +defm MIN : BinaryIntrinsicFloat<IL_OP_MIN, int_AMDIL_min>; +defm MAX : BinaryIntrinsicFloat<IL_OP_MAX, int_AMDIL_max>; +defm MAD : TernaryIntrinsicFloat<IL_OP_MAD, int_AMDIL_mad>; + } +defm MOD : BinaryOpMCf32<IL_OP_MOD, frem>; +} +let hasZeroOpFlag = 1 in { + let mayLoad = 0, mayStore=0 in { +defm LN : UnaryIntrinsicFloatScalar<IL_OP_LN, int_AMDIL_ln>; +defm LOG : UnaryIntrinsicFloatScalar<IL_OP_LOG, int_AMDIL_log>; +defm RSQ : UnaryIntrinsicFloatScalar<IL_OP_RSQ, int_AMDIL_rsq>; +defm DIV : BinaryIntrinsicFloat<IL_OP_DIV, int_AMDIL_div>; + } +} + let mayLoad = 0, mayStore=0 in { +defm CLAMP : TernaryIntrinsicFloat<IL_OP_CLAMP, int_AMDIL_clamp>; +defm FMA : TernaryIntrinsicFloat<IL_OP_FMA, int_AMDIL_fma>; +defm LERP : TernaryIntrinsicFloat<IL_OP_LERP, int_AMDIL_lerp>; + } +defm SUB : BinaryOpMCf32<IL_OP_SUB, fsub>; +defm FABS : UnaryOpMCf32<IL_OP_ABS, fabs>; +defm FMAD : TernaryOpMCf32<IL_OP_MAD, IL_mad>; +defm NEAR : UnaryOpMCf32<IL_OP_ROUND_NEAR, fnearbyint>; +defm RND_Z : UnaryOpMCf32<IL_OP_ROUND_ZERO, ftrunc>; + +// This opcode has custom swizzle pattern encoded in Swizzle Encoder +def NEG_f32 : OneInOneOut<IL_OP_MOV, (outs GPRF32:$dst), + (ins GPRF32:$src0), + !strconcat(IL_OP_MOV.Text, " $dst, $src0"), + [(set GPRF32:$dst, (fneg GPRF32:$src0))]>; +def INTTOANY_f32 : OneInOneOut<IL_OP_MOV, (outs GPRF32:$dst), + (ins GPRI32:$src0), + !strconcat(IL_OP_MOV.Text, " $dst, $src0"), + [(set GPRF32:$dst, (IL_inttoany GPRI32:$src0))]>; +let hasIEEEFlag = 1 in { +def DP2ADD_f32 : ThreeInOneOut<IL_OP_DP2_ADD, (outs GPRF32:$dst), + (ins GPRV2F32:$src0, GPRV2F32:$src1, GPRF32:$src2), + !strconcat(IL_OP_DP2_ADD.Text, " $dst, $src0, $src1, $src2"), + [(set GPRF32:$dst, + (int_AMDIL_dp2_add GPRV2F32:$src0, + GPRV2F32:$src1, GPRF32:$src2))]>; +def DP2_f32 : TwoInOneOut<IL_OP_DP2, (outs GPRF32:$dst), + (ins GPRV2F32:$src0, GPRV2F32:$src1), + !strconcat(IL_OP_DP2.Text, " $dst, $src0, $src1"), + [(set GPRF32:$dst, + (int_AMDIL_dp2 GPRV2F32:$src0, GPRV2F32:$src1))]>; +def DP3_f32 : TwoInOneOut<IL_OP_DP3, (outs GPRF32:$dst), + (ins GPRV4F32:$src0, GPRV4F32:$src1), + !strconcat(IL_OP_DP3.Text, " $dst, $src0, $src1"), + [(set GPRF32:$dst, + (int_AMDIL_dp3 GPRV4F32:$src0, GPRV4F32:$src1))]>; +def DP4_f32 : TwoInOneOut<IL_OP_DP4, (outs GPRF32:$dst), + (ins GPRV4F32:$src0, GPRV4F32:$src1), + !strconcat(IL_OP_DP4.Text, " $dst, $src0, $src1"), + [(set GPRF32:$dst, + (int_AMDIL_dp4 GPRV4F32:$src0, GPRV4F32:$src1))]>; +} +defm UNPACK_B0 : IntrConvertI32TOF32<IL_OP_UNPACK_0, int_AMDIL_media_unpack_byte_0>; +defm UNPACK_B1 : IntrConvertI32TOF32<IL_OP_UNPACK_1, int_AMDIL_media_unpack_byte_1>; +defm UNPACK_B2 : IntrConvertI32TOF32<IL_OP_UNPACK_2, int_AMDIL_media_unpack_byte_2>; +defm UNPACK_B3 : IntrConvertI32TOF32<IL_OP_UNPACK_3, int_AMDIL_media_unpack_byte_3>; +defm FTOI_FLR : IntrConvertF32TOI32<IL_OP_FTOI_FLR, int_AMDIL_convert_f32_i32_flr>; +defm FTOI_RPI : IntrConvertF32TOI32<IL_OP_FTOI_RPI, int_AMDIL_convert_f32_i32_rpi>; +defm HTOF : IntrConvertF16TOF32<IL_OP_F16_TO_F32, int_AMDIL_convert_f16_f32>; +defm FTOH : IntrConvertF32TOF16<IL_OP_F32_TO_F16, int_AMDIL_convert_f32_f16>; +defm FTOH_NEAR : IntrConvertF32TOF16<IL_OP_F32_TO_F16_NEAR, int_AMDIL_convert_f32_f16_near>; +defm FTOH_NEG_INF : IntrConvertF32TOF16<IL_OP_F32_TO_F16_NEG_INF, int_AMDIL_convert_f32_f16_neg_inf>; +defm FTOH_PLUS_INF : IntrConvertF32TOF16<IL_OP_F32_TO_F16_PLUS_INF, int_AMDIL_convert_f32_f16_plus_inf>; +//===---------------------------------------------------------------------===// +// float math instructions end here +//===---------------------------------------------------------------------===// + +//===---------------------------------------------------------------------===// +// float2 math instructions start here +//===---------------------------------------------------------------------===// +// This opcode has custom swizzle pattern encoded in Swizzle Encoder +def NEG_v2f32 : OneInOneOut<IL_OP_MOV, (outs GPRV2F32:$dst), + (ins GPRV2F32:$src0), + !strconcat(IL_OP_MOV.Text, " $dst, $src0"), + [(set GPRV2F32:$dst, (fneg GPRV2F32:$src0))]>; +//===---------------------------------------------------------------------===// +// float2 math instructions end here +//===---------------------------------------------------------------------===// + +//===---------------------------------------------------------------------===// +// float4 math instructions start here +//===---------------------------------------------------------------------===// +// This opcode has custom swizzle pattern encoded in Swizzle Encoder +def NEG_v4f32 : OneInOneOut<IL_OP_MOV, (outs GPRV4F32:$dst), + (ins GPRV4F32:$src0), + !strconcat(IL_OP_MOV.Text, " $dst, $src0"), + [(set GPRV4F32:$dst, (fneg GPRV4F32:$src0))]>; +//===---------------------------------------------------------------------===// +// float4 math instructions end here +//===---------------------------------------------------------------------===// + +//===---------------------------------------------------------------------===// +// double math instructions start here +//===---------------------------------------------------------------------===// +// This opcode has custom swizzle pattern encoded in Swizzle Encoder +def SUB_f64 : TwoInOneOut<IL_OP_D_ADD, (outs GPRF64:$dst), + (ins GPRF64:$src0, GPRF64:$src1), + !strconcat(IL_OP_D_ADD.Text, " $dst, $src0, $src1"), + [(set GPRF64:$dst, (fsub GPRF64:$src0, GPRF64:$src1))]>; +// This opcode has custom swizzle pattern encoded in Swizzle Encoder +def SUB_v2f64 : TwoInOneOut<IL_OP_D_ADD, (outs GPRV2F64:$dst), + (ins GPRV2F64:$src0, GPRV2F64:$src1), + !strconcat(IL_OP_D_ADD.Text, " $dst, $src0, $src1"), + [(set GPRV2F64:$dst, (fsub GPRV2F64:$src0, GPRV2F64:$src1))]>; +// This opcode has custom swizzle pattern encoded in Swizzle Encoder +def NEG_f64 : OneInOneOut<IL_OP_MOV, (outs GPRF64:$dst), + (ins GPRF64:$src0), + !strconcat(IL_OP_MOV.Text, " $dst, $src0"), + [(set GPRF64:$dst, (fneg GPRF64:$src0))]>; +// This opcode has custom swizzle pattern encoded in Swizzle Encoder +def NEG_v2f64 : OneInOneOut<IL_OP_MOV, (outs GPRV2F64:$dst), + (ins GPRV2F64:$src0), + !strconcat(IL_OP_MOV.Text, " $dst, $src0"), + [(set GPRV2F64:$dst, (fneg GPRV2F64:$src0))]>; + let mayLoad = 0, mayStore=0 in { +defm MIN : BinaryIntrinsicDouble<IL_OP_D_MIN, int_AMDIL_min>; +defm MAX : BinaryIntrinsicDouble<IL_OP_D_MAX, int_AMDIL_max>; +defm DIV : BinaryIntrinsicDouble<IL_OP_D_DIV, int_AMDIL_div>; +defm MAD : TernaryIntrinsicDouble<IL_OP_D_MAD, int_AMDIL_mad>; +defm DFMA : TernaryIntrinsicDouble<IL_OP_D_MAD, int_AMDIL_fma>; +defm FRAC : UnaryIntrinsicDouble<IL_OP_D_FRC, int_AMDIL_fraction>; +defm SQRT : UnaryIntrinsicDouble<IL_OP_D_SQRT, int_AMDIL_sqrt>; +defm RSQ : UnaryIntrinsicDoubleScalar<IL_OP_D_RSQ, int_AMDIL_rsq>; +defm RCP : UnaryIntrinsicDoubleScalar<IL_OP_D_RCP, int_AMDIL_drcp>; +defm DMAD : TernaryOpMCf64<IL_OP_D_MAD, IL_mad>; + } +def FREXP_f64 : OneInOneOut<IL_OP_D_FREXP, (outs GPRV2I64:$dst), + (ins GPRF64:$src), + !strconcat(IL_OP_D_FREXP.Text," $dst, $src"), + [(set GPRV2I64:$dst, + (int_AMDIL_frexp_f64 GPRF64:$src))]>; +def LDEXP_f64 : TwoInOneOut<IL_OP_D_LDEXP, (outs GPRF64:$dst), + (ins GPRF64:$src, GPRI32:$src1), + !strconcat(IL_OP_D_LDEXP.Text, " $dst, $src, $src1"), + [(set GPRF64:$dst, + (int_AMDIL_ldexp GPRF64:$src, GPRI32:$src1))]>; +def LDEXP_v2f64 : TwoInOneOut<IL_OP_D_LDEXP, (outs GPRV2F64:$dst), + (ins GPRV2F64:$src, GPRV2I32:$src1), + !strconcat(IL_OP_D_LDEXP.Text, " $dst, $src, $src1"), + [(set GPRV2F64:$dst, + (int_AMDIL_ldexp GPRV2F64:$src, GPRV2I32:$src1))]>; +//===---------------------------------------------------------------------===// +// double math instructions end here +//===---------------------------------------------------------------------===// + +//===---------------------------------------------------------------------===// +// Various Macros +//===---------------------------------------------------------------------===// +def MACRO__sdiv_i8 : BinaryMacro< GPRI8, GPRI8, GPRI8, sdiv>; +def MACRO__sdiv_i16 : BinaryMacro<GPRI16, GPRI16, GPRI16, sdiv>; +def MACRO__sdiv_i32 : BinaryMacro<GPRI32, GPRI32, GPRI32, sdiv>; +def MACRO__udiv_i8 : BinaryMacro< GPRI8, GPRI8, GPRI8, udiv>; +def MACRO__udiv_i16 : BinaryMacro<GPRI16, GPRI16, GPRI16, udiv>; +def MACRO__udiv_i32 : BinaryMacro<GPRI32, GPRI32, GPRI32, udiv>; +def MACRO__smod_i8 : BinaryMacro< GPRI8, GPRI8, GPRI8, srem>; +def MACRO__smod_i16 : BinaryMacro<GPRI16, GPRI16, GPRI16, srem>; +def MACRO__smod_i32 : BinaryMacro<GPRI32, GPRI32, GPRI32, srem>; +def MACRO__umod_i8 : BinaryMacro< GPRI8, GPRI8, GPRI8, urem>; +def MACRO__umod_i16 : BinaryMacro<GPRI16, GPRI16, GPRI16, urem>; +def MACRO__umod_i32 : BinaryMacro<GPRI32, GPRI32, GPRI32, urem>; +let Predicates = [HasSWDDiv] in { + def MACRO__ddiv_f64: BinaryMacro<GPRF64, GPRF64, GPRF64, fdiv>; +} +let Predicates = [HasHWDDiv] in { + def MACRO__ddiv_f64_fma: BinaryMacro<GPRF64, GPRF64, GPRF64, fdiv>; +} +def MACRO__ftol_i64 : UnaryMacro<GPRI64, GPRF32, fp_to_sint>; +def MACRO__ftoul_i64 : UnaryMacro<GPRI64, GPRF32, fp_to_uint>; +def MACRO__ultof_f32 : UnaryMacro<GPRF32, GPRI64, uint_to_fp>; +def MACRO__ltof_f32 : UnaryMacro<GPRF32, GPRI64, sint_to_fp>; +let Predicates = [HasSW64Mul] in { +def MACRO__mul_i64 : BinaryMacro<GPRI64, GPRI64, GPRI64, mul>; +def MACRO__mul_v2i64 : BinaryMacro<GPRV2I64, GPRV2I64, GPRV2I64, mul>; +} +let Predicates = [HasSW64DivMod] in { +def MACRO__sdiv_i64 : BinaryMacro<GPRI64, GPRI64, GPRI64, sdiv>; +def MACRO__udiv_i64 : BinaryMacro<GPRI64, GPRI64, GPRI64, udiv>; +def MACRO__smod_i64 : BinaryMacro<GPRI64, GPRI64, GPRI64, srem>; +def MACRO__umod_i64 : BinaryMacro<GPRI64, GPRI64, GPRI64, urem>; +} +let Predicates = [HasHW64DivMod] in { + defm SDIV : BinaryOpMCi64<IL_OP_I64_DIV, sdiv>; + defm UDIV : BinaryOpMCi64<IL_OP_U64_DIV, udiv>; + defm SMOD : BinaryOpMCi64<IL_OP_I64_MOD, srem>; + defm UMOD : BinaryOpMCi64<IL_OP_U64_MOD, urem>; +} +let Predicates = [HasHW64Mul] in { + defm SMUL : BinaryOpMCi64<IL_OP_I64_MUL, mul>; + defm UMUL : BinaryOpMCi64<IL_OP_U64_MUL, IL_umul>; +} +def MACRO__shr_v2i64 : BinaryMacro<GPRV2I64, GPRV2I64, GPRV2I32, srl>; +def MACRO__shl_v2i64 : BinaryMacro<GPRV2I64, GPRV2I64, GPRV2I32, shl>; +def MACRO__sra_v2i64 : BinaryMacro<GPRV2I64, GPRV2I64, GPRV2I32, sra>; + +let Predicates = [HasSW64Bit] in { +def MACRO__shr_i64 : BinaryMacro<GPRI64, GPRI64, GPRI32, srl>; +def MACRO__shl_i64 : BinaryMacro<GPRI64, GPRI64, GPRI32, shl>; +def MACRO__sra_i64 : BinaryMacro<GPRI64, GPRI64, GPRI32, sra>; +} +//===---------------------------------------------------------------------===// +// Comparison Instructions +//===---------------------------------------------------------------------===// +let usesCustomInserter = 1 in { + defm CMP : Compare<"Pseudo comparison instr">; +} +//===---------------------------------------------------------------------===// +// 32-bit floating point operations +//===---------------------------------------------------------------------===// +def FEQ : TwoInOneOut<IL_OP_EQ, (outs GPRF32:$dst), + (ins GPRF32:$lhs, GPRF32:$rhs), + !strconcat(IL_OP_EQ.Text, " $dst, $lhs, $rhs") + , []>; +def FGE : TwoInOneOut<IL_OP_GE, (outs GPRF32:$dst), + (ins GPRF32:$lhs, GPRF32:$rhs), + !strconcat(IL_OP_GE.Text, " $dst, $lhs, $rhs") + , []>; +def FLT : TwoInOneOut<IL_OP_LT, (outs GPRF32:$dst), + (ins GPRF32:$lhs, GPRF32:$rhs), + !strconcat(IL_OP_LT.Text, " $dst, $lhs, $rhs") + , []>; +def FLT_v2f32 : TwoInOneOut<IL_OP_LT, (outs GPRV2F32:$dst), + (ins GPRV2F32:$lhs, GPRV2F32:$rhs), + !strconcat(IL_OP_LT.Text, " $dst, $lhs, $rhs") + , []>; +def FLT_v4f32 : TwoInOneOut<IL_OP_LT, (outs GPRV4F32:$dst), + (ins GPRV4F32:$lhs, GPRV4F32:$rhs), + !strconcat(IL_OP_LT.Text, " $dst, $lhs, $rhs") + , []>; +def FNE : TwoInOneOut<IL_OP_NE, (outs GPRF32:$dst), + (ins GPRF32:$lhs, GPRF32:$rhs), + !strconcat(IL_OP_NE.Text, " $dst, $lhs, $rhs") + , []>; + +//===---------------------------------------------------------------------===// +//TODO: need to correctly define comparison instructions +//===---------------------------------------------------------------------===// +def DEQ : TwoInOneOut<IL_OP_D_EQ, (outs GPRF64:$dst), + (ins GPRF64:$lhs, GPRF64:$rhs), + !strconcat(IL_OP_D_EQ.Text, " $dst, $lhs, $rhs") + , []>; +def DEQ_v2f64 : TwoInOneOut<IL_OP_D_EQ, (outs GPRV2F64:$dst), + (ins GPRV2F64:$lhs, GPRV2F64:$rhs), + !strconcat(IL_OP_D_EQ.Text, " $dst, $lhs, $rhs") + , []>; +def DGE : TwoInOneOut<IL_OP_D_GE, (outs GPRF64:$dst), + (ins GPRF64:$lhs, GPRF64:$rhs), + !strconcat(IL_OP_D_GE.Text, " $dst, $lhs, $rhs") + , []>; +def DLT : TwoInOneOut<IL_OP_D_LT, (outs GPRF64:$dst), + (ins GPRF64:$lhs, GPRF64:$rhs), + !strconcat(IL_OP_D_LT.Text, " $dst, $lhs, $rhs") + , []>; +def DNE : TwoInOneOut<IL_OP_D_NE, (outs GPRF64:$dst), + (ins GPRF64:$lhs, GPRF64:$rhs), + !strconcat(IL_OP_D_NE.Text, " $dst, $lhs, $rhs") + , []>; + +//===---------------------------------------------------------------------===// +//TODO: need to correctly define comparison instructions +//===---------------------------------------------------------------------===// +def IEQ : TwoInOneOut<IL_OP_I_EQ, (outs GPRI32:$dst), + (ins GPRI32:$lhs, GPRI32:$rhs), + !strconcat(IL_OP_I_EQ.Text, " $dst, $lhs, $rhs") + , []>; +def IEQ_v2i32 : TwoInOneOut<IL_OP_I_EQ, (outs GPRV2I32:$dst), + (ins GPRV2I32:$lhs, GPRV2I32:$rhs), + !strconcat(IL_OP_I_EQ.Text, " $dst, $lhs, $rhs") + , []>; +def IEQ_v4i32 : TwoInOneOut<IL_OP_I_EQ, (outs GPRV4I32:$dst), + (ins GPRV4I32:$lhs, GPRV4I32:$rhs), + !strconcat(IL_OP_I_EQ.Text, " $dst, $lhs, $rhs") + , []>; +def IGE : TwoInOneOut<IL_OP_I_GE, (outs GPRI32:$dst), + (ins GPRI32:$lhs, GPRI32:$rhs), + !strconcat(IL_OP_I_GE.Text, " $dst, $lhs, $rhs") + , []>; +def IGE_v2i32 : TwoInOneOut<IL_OP_I_GE, (outs GPRV2I32:$dst), + (ins GPRV2I32:$lhs, GPRV2I32:$rhs), + !strconcat(IL_OP_I_GE.Text, " $dst, $lhs, $rhs") + , []>; +def IGE_v4i32 : TwoInOneOut<IL_OP_I_GE, (outs GPRV4I32:$dst), + (ins GPRV4I32:$lhs, GPRV4I32:$rhs), + !strconcat(IL_OP_I_GE.Text, " $dst, $lhs, $rhs") + , []>; +def ILT : TwoInOneOut<IL_OP_I_LT, (outs GPRI32:$dst), + (ins GPRI32:$lhs, GPRI32:$rhs), + !strconcat(IL_OP_I_LT.Text, " $dst, $lhs, $rhs") + , []>; +def ILT_v2i32 : TwoInOneOut<IL_OP_I_LT, (outs GPRV2I32:$dst), + (ins GPRV2I32:$lhs, GPRV2I32:$rhs), + !strconcat(IL_OP_I_LT.Text, " $dst, $lhs, $rhs") + , []>; +def ILT_v4i32 : TwoInOneOut<IL_OP_I_LT, (outs GPRV4I32:$dst), + (ins GPRV4I32:$lhs, GPRV4I32:$rhs), + !strconcat(IL_OP_I_LT.Text, " $dst, $lhs, $rhs") + , []>; +def INE : TwoInOneOut<IL_OP_I_NE, (outs GPRI32:$dst), + (ins GPRI32:$lhs, GPRI32:$rhs), + !strconcat(IL_OP_I_NE.Text, " $dst, $lhs, $rhs") + , []>; +def INE_v2i32 : TwoInOneOut<IL_OP_I_NE, (outs GPRV2I32:$dst), + (ins GPRV2I32:$lhs, GPRV2I32:$rhs), + !strconcat(IL_OP_I_NE.Text, " $dst, $lhs, $rhs") + , []>; +def INE_v4i32 : TwoInOneOut<IL_OP_I_NE, (outs GPRV4I32:$dst), + (ins GPRV4I32:$lhs, GPRV4I32:$rhs), + !strconcat(IL_OP_I_NE.Text, " $dst, $lhs, $rhs") + , []>; +let Predicates = [HasHW64Bit] in { +def LEQ : TwoInOneOut<IL_OP_I64_EQ, (outs GPRI64:$dst), + (ins GPRI64:$lhs, GPRI64:$rhs), + !strconcat(IL_OP_I64_EQ.Text, " $dst, $lhs, $rhs") + , []>; +def LGE : TwoInOneOut<IL_OP_I64_GE, (outs GPRI64:$dst), + (ins GPRI64:$lhs, GPRI64:$rhs), + !strconcat(IL_OP_I64_GE.Text, " $dst, $lhs, $rhs") + , []>; +def LLE : TwoInOneOut<IL_OP_I64_GE, (outs GPRI64:$dst), + (ins GPRI64:$lhs, GPRI64:$rhs), + !strconcat(IL_OP_I64_GE.Text, " $dst, $rhs, $lhs") + , []>; +def LGT : TwoInOneOut<IL_OP_I64_LT, (outs GPRI64:$dst), + (ins GPRI64:$lhs, GPRI64:$rhs), + !strconcat(IL_OP_I64_LT.Text, " $dst, $rhs, $lhs") + , []>; +def LLT : TwoInOneOut<IL_OP_I64_LT, (outs GPRI64:$dst), + (ins GPRI64:$lhs, GPRI64:$rhs), + !strconcat(IL_OP_I64_LT.Text, " $dst, $lhs, $rhs") + , []>; +def LNE : TwoInOneOut<IL_OP_I64_NE, (outs GPRI64:$dst), + (ins GPRI64:$lhs, GPRI64:$rhs), + !strconcat(IL_OP_I64_NE.Text, " $dst, $lhs, $rhs") + , []>; +} + +//===---------------------------------------------------------------------===// +// Unsigned Integer Operations +//===---------------------------------------------------------------------===// + +//===---------------------------------------------------------------------===// +//TODO: need to correctly define comparison instructions +//===---------------------------------------------------------------------===// +def UEQ : TwoInOneOut<IL_OP_I_EQ, (outs GPRI32:$dst), + (ins GPRI32:$lhs, GPRI32:$rhs), + !strconcat(IL_OP_I_EQ.Text, " $dst, $lhs, $rhs") + , []>; +def UEQ_v2i32 : TwoInOneOut<IL_OP_I_EQ, (outs GPRV2I32:$dst), + (ins GPRV2I32:$lhs, GPRV2I32:$rhs), + !strconcat(IL_OP_I_EQ.Text, " $dst, $lhs, $rhs") + , []>; +def UEQ_v4i32 : TwoInOneOut<IL_OP_I_EQ, (outs GPRV4I32:$dst), + (ins GPRV4I32:$lhs, GPRV4I32:$rhs), + !strconcat(IL_OP_I_EQ.Text, " $dst, $lhs, $rhs") + , []>; +def ULE : TwoInOneOut<IL_OP_U_GE, (outs GPRI32:$dst), + (ins GPRI32:$lhs, GPRI32:$rhs), + !strconcat(IL_OP_U_GE.Text, " $dst, $lhs, $rhs") + , []>; +def ULE_v2i32 : TwoInOneOut<IL_OP_U_GE, (outs GPRV2I32:$dst), + (ins GPRV2I32:$lhs, GPRV2I32:$rhs), + !strconcat(IL_OP_U_GE.Text, " $dst, $lhs, $rhs") + , []>; +def ULE_v4i32 : TwoInOneOut<IL_OP_U_GE, (outs GPRV4I32:$dst), + (ins GPRV4I32:$lhs, GPRV4I32:$rhs), + !strconcat(IL_OP_U_GE.Text, " $dst, $lhs, $rhs") + , []>; +def UGT : TwoInOneOut<IL_OP_U_LT, (outs GPRI32:$dst), + (ins GPRI32:$lhs, GPRI32:$rhs), + !strconcat(IL_OP_U_LT.Text, " $dst, $lhs, $rhs") + , []>; +def UGT_v2i32 : TwoInOneOut<IL_OP_U_LT, (outs GPRV2I32:$dst), + (ins GPRV2I32:$lhs, GPRV2I32:$rhs), + !strconcat(IL_OP_U_LT.Text, " $dst, $lhs, $rhs") + , []>; +def UGT_v4i32 : TwoInOneOut<IL_OP_U_LT, (outs GPRV4I32:$dst), + (ins GPRV4I32:$lhs, GPRV4I32:$rhs), + !strconcat(IL_OP_U_LT.Text, " $dst, $lhs, $rhs") + , []>; +def UGE : TwoInOneOut<IL_OP_U_GE, (outs GPRI32:$dst), + (ins GPRI32:$lhs, GPRI32:$rhs), + !strconcat(IL_OP_U_GE.Text, " $dst, $lhs, $rhs") + , []>; +def UGE_v2i32 : TwoInOneOut<IL_OP_U_GE, (outs GPRV2I32:$dst), + (ins GPRV2I32:$lhs, GPRV2I32:$rhs), + !strconcat(IL_OP_U_GE.Text, " $dst, $lhs, $rhs") + , []>; +def UGE_v4i32 : TwoInOneOut<IL_OP_U_GE, (outs GPRV4I32:$dst), + (ins GPRV4I32:$lhs, GPRV4I32:$rhs), + !strconcat(IL_OP_U_GE.Text, " $dst, $lhs, $rhs") + , []>; +def ULT : TwoInOneOut<IL_OP_U_LT, (outs GPRI32:$dst), + (ins GPRI32:$lhs, GPRI32:$rhs), + !strconcat(IL_OP_U_LT.Text, " $dst, $lhs, $rhs") + , []>; +def ULT_v2i32 : TwoInOneOut<IL_OP_U_LT, (outs GPRV2I32:$dst), + (ins GPRV2I32:$lhs, GPRV2I32:$rhs), + !strconcat(IL_OP_U_LT.Text, " $dst, $lhs, $rhs") + , []>; +def ULT_v4i32 : TwoInOneOut<IL_OP_U_LT, (outs GPRV4I32:$dst), + (ins GPRV4I32:$lhs, GPRV4I32:$rhs), + !strconcat(IL_OP_U_LT.Text, " $dst, $lhs, $rhs") + , []>; +def UNE : TwoInOneOut<IL_OP_I_NE, (outs GPRI32:$dst), + (ins GPRI32:$lhs, GPRI32:$rhs), + !strconcat(IL_OP_I_NE.Text, " $dst, $lhs, $rhs") + , []>; +def UNE_v2i32 : TwoInOneOut<IL_OP_I_NE, (outs GPRV2I32:$dst), + (ins GPRV2I32:$lhs, GPRV2I32:$rhs), + !strconcat(IL_OP_I_NE.Text, " $dst, $lhs, $rhs") + , []>; +def UNE_v4i32 : TwoInOneOut<IL_OP_I_NE, (outs GPRV4I32:$dst), + (ins GPRV4I32:$lhs, GPRV4I32:$rhs), + !strconcat(IL_OP_I_NE.Text, " $dst, $lhs, $rhs") + , []>; +let Predicates = [HasHW64Bit] in { +def ULLE : TwoInOneOut<IL_OP_U64_GE, (outs GPRI64:$dst), + (ins GPRI64:$lhs, GPRI64:$rhs), + !strconcat(IL_OP_U64_GE.Text, " $dst, $rhs, $lhs") + , []>; +def ULGT : TwoInOneOut<IL_OP_U64_LT, (outs GPRI64:$dst), + (ins GPRI64:$lhs, GPRI64:$rhs), + !strconcat(IL_OP_U64_LT.Text, " $dst, $rhs, $lhs") + , []>; +def ULGE : TwoInOneOut<IL_OP_U64_GE, (outs GPRI64:$dst), + (ins GPRI64:$lhs, GPRI64:$rhs), + !strconcat(IL_OP_U64_GE.Text, " $dst, $lhs, $rhs") + , []>; +def ULLT : TwoInOneOut<IL_OP_U64_LT, (outs GPRI64:$dst), + (ins GPRI64:$lhs, GPRI64:$rhs), + !strconcat(IL_OP_U64_LT.Text, " $dst, $lhs, $rhs") + , []>; +} +//===---------------------------------------------------------------------===// +// Scalar ==> Scalar conversion functions +//===---------------------------------------------------------------------===// +// f32 ==> f64 +def FTOD : UnaryOp<IL_OP_F_2_D, fextend, GPRF64, GPRF32>; +// f64 ==> f32 +def DTOF : UnaryOp<IL_OP_D_2_F, IL_d2f, GPRF32, GPRF64>; +// f32 ==> i32 signed +def FTOI : UnaryOp<IL_OP_FTOI, fp_to_sint, GPRI32, GPRF32>; +def FTOI_v2i32 : UnaryOp<IL_OP_FTOI, fp_to_sint, GPRV2I32, GPRV2F32>; +def FTOI_v4i32 : UnaryOp<IL_OP_FTOI, fp_to_sint, GPRV4I32, GPRV4F32>; +// i32 ==> f32 signed +def ITOF : UnaryOp<IL_OP_ITOF, sint_to_fp, GPRF32, GPRI32>; +def ITOF_v2f32 : UnaryOp<IL_OP_ITOF, sint_to_fp, GPRV2F32, GPRV2I32>; +def ITOF_v4f32 : UnaryOp<IL_OP_ITOF, sint_to_fp, GPRV4F32, GPRV4I32>; +// f32 ==> i32 unsigned +def FTOU : UnaryOp<IL_OP_FTOU, fp_to_uint, GPRI32, GPRF32>; +def FTOU_v2i32 : UnaryOp<IL_OP_FTOU, fp_to_uint, GPRV2I32, GPRV2F32>; +def FTOU_v4i32 : UnaryOp<IL_OP_FTOU, fp_to_uint, GPRV4I32, GPRV4F32>; +// i32 ==> f32 unsigned +def UTOF : UnaryOp<IL_OP_UTOF, uint_to_fp, GPRF32, GPRI32>; +def UTOF_v2f32 : UnaryOp<IL_OP_UTOF, uint_to_fp, GPRV2F32, GPRV2I32>; +def UTOF_v4f32 : UnaryOp<IL_OP_UTOF, uint_to_fp, GPRV4F32, GPRV4I32>; +// Get upper 32 bits of f64 +// This opcode has custom swizzle pattern encoded in Swizzle Encoder +def DHI : OneInOneOut<IL_OP_MOV, (outs GPRI32:$dst), + (ins GPRF64:$src), + !strconcat(IL_OP_MOV.Text, " $dst, $src"), + [(set GPRI32:$dst, (IL_dcomphi GPRF64:$src))]>; +// This opcode has custom swizzle pattern encoded in Swizzle Encoder +def DHI_v2f64 : OneInOneOut<IL_OP_MOV, (outs GPRV2I32:$dst), + (ins GPRV2F64:$src), + !strconcat(IL_OP_MOV.Text, " $dst, $src"), + [(set GPRV2I32:$dst, (IL_dcomphi2 GPRV2F64:$src))]>; +// Get lower 32 bits of f64 +// This opcode has custom swizzle pattern encoded in Swizzle Encoder +def DLO : OneInOneOut<IL_OP_MOV, (outs GPRI32:$dst), + (ins GPRF64:$src), + !strconcat(IL_OP_MOV.Text, " $dst, $src"), + [(set GPRI32:$dst, (IL_dcomplo GPRF64:$src))]>; +// This opcode has custom swizzle pattern encoded in Swizzle Encoder +def DLO_v2f64 : OneInOneOut<IL_OP_MOV, (outs GPRV2I32:$dst), + (ins GPRV2F64:$src), + !strconcat(IL_OP_MOV.Text, " $dst, $src"), + [(set GPRV2I32:$dst, (IL_dcomplo2 GPRV2F64:$src))]>; +// Convert two 32 bit integers into a f64 +// This opcode has custom swizzle pattern encoded in Swizzle Encoder +def DCREATE : TwoInOneOut<IL_OP_I_ADD, (outs GPRF64:$dst), + (ins GPRI32:$src0, GPRI32:$src1), + !strconcat(IL_OP_I_ADD.Text, " $dst, $src0, $src1"), + [(set GPRF64:$dst, (IL_dcreate GPRI32:$src0, GPRI32:$src1))]>; +// This opcode has custom swizzle pattern encoded in Swizzle Encoder +def DCREATE_v2f64 : TwoInOneOut<IL_OP_I_ADD, (outs GPRV2F64:$dst), + (ins GPRV2I32:$src0, GPRV2I32:$src1), + !strconcat(IL_OP_I_ADD.Text, " $dst, $src0, $src1"), + [(set GPRV2F64:$dst, + (IL_dcreate2 GPRV2I32:$src0, GPRV2I32:$src1))]>; +// Get upper 32 bits of i64 +// This opcode has custom swizzle pattern encoded in Swizzle Encoder +def LHI : OneInOneOut<IL_OP_MOV, (outs GPRI32:$dst), + (ins GPRI64:$src), + !strconcat(IL_OP_MOV.Text, " $dst, $src"), + [(set GPRI32:$dst, (IL_lcomphi GPRI64:$src))]>; +// This opcode has custom swizzle pattern encoded in Swizzle Encoder +def LHI_v2i64 : OneInOneOut<IL_OP_MOV, (outs GPRV2I32:$dst), + (ins GPRV2I64:$src), + !strconcat(IL_OP_MOV.Text, " $dst, $src"), + [(set GPRV2I32:$dst, (IL_lcomphi2 GPRV2I64:$src))]>; +// Get lower 32 bits of i64 +// This opcode has custom swizzle pattern encoded in Swizzle Encoder +def LLO : OneInOneOut<IL_OP_MOV, (outs GPRI32:$dst), + (ins GPRI64:$src), + !strconcat(IL_OP_MOV.Text, " $dst, $src"), + [(set GPRI32:$dst, (IL_lcomplo GPRI64:$src))]>; +// This opcode has custom swizzle pattern encoded in Swizzle Encoder +def LLO_v2i64 : OneInOneOut<IL_OP_MOV, (outs GPRV2I32:$dst), + (ins GPRV2I64:$src), + !strconcat(IL_OP_MOV.Text, " $dst, $src"), + [(set GPRV2I32:$dst, (IL_lcomplo2 GPRV2I64:$src))]>; +// This opcode has custom swizzle pattern encoded in Swizzle Encoder +def HILO_BITOR_v4i16 : TwoInOneOut<IL_OP_I_OR, (outs GPRI32:$dst), + (ins GPRI32:$src, GPRI32:$src2), + !strconcat(IL_OP_I_OR.Text, " $dst, $src, $src2"), []>; +// This opcode has custom swizzle pattern encoded in Swizzle Encoder +def HILO_BITOR_v2i32 : TwoInOneOut<IL_OP_I_OR, (outs GPRI32:$dst), + (ins GPRI32:$src, GPRI32:$src2), + !strconcat(IL_OP_I_OR.Text, " $dst, $src, $src2"), []>; +// This opcode has custom swizzle pattern encoded in Swizzle Encoder +def HILO_BITOR_v2i64 : TwoInOneOut<IL_OP_I_OR, (outs GPRI64:$dst), + (ins GPRI64:$src, GPRI64:$src2), + !strconcat(IL_OP_I_OR.Text, " $dst, $src, $src2"), []>; +// Convert two 32 bit integers into a i64 +// This opcode has custom swizzle pattern encoded in Swizzle Encoder +def LCREATE : TwoInOneOut<IL_OP_I_ADD, (outs GPRI64:$dst), + (ins GPRI32:$src0, GPRI32:$src1), + !strconcat(IL_OP_I_ADD.Text, " $dst, $src0, $src1"), + [(set GPRI64:$dst, (IL_lcreate GPRI32:$src0, GPRI32:$src1))]>; +// This opcode has custom swizzle pattern encoded in Swizzle Encoder +def LCREATE_v2i64 : TwoInOneOut<IL_OP_I_ADD, (outs GPRV2I64:$dst), + (ins GPRV2I32:$src0, GPRV2I32:$src1), + !strconcat(IL_OP_I_ADD.Text, " $dst, $src0, $src1"), + [(set GPRV2I64:$dst, + (IL_lcreate2 GPRV2I32:$src0, GPRV2I32:$src1))]>; +//===---------------------------------------------------------------------===// +// Scalar ==> Vector conversion functions +//===---------------------------------------------------------------------===// +// This opcode has custom swizzle pattern encoded in Swizzle Encoder +defm VCREATE : UnaryOpMCVec<IL_OP_MOV, IL_vbuild>; + +//===---------------------------------------------------------------------===// +// Vector ==> Scalar conversion functions +//===---------------------------------------------------------------------===// + +// This opcode has custom swizzle pattern encoded in Swizzle Encoder +defm VEXTRACT : VectorExtract<IL_vextract>; + +//===---------------------------------------------------------------------===// +// Vector ==> Vector conversion functions +//===---------------------------------------------------------------------===// +// This opcode has custom swizzle pattern encoded in Swizzle Encoder +defm VINSERT : VectorInsert<IL_vinsert>; +// This opcode has custom swizzle pattern encoded in Swizzle Encoder +defm VCONCAT : VectorConcat<IL_vconcat>; + +//===---------------------------------------------------------------------===// +// Bit conversion functions +//===---------------------------------------------------------------------===// +defm IL_ASCHAR : BitConversion<IL_OP_MOV, GPRI8, IL_bitconv>; +defm IL_ASSHORT : BitConversion<IL_OP_MOV, GPRI16, IL_bitconv>; +defm IL_ASINT : BitConversion<IL_OP_MOV, GPRI32, IL_bitconv>; +defm IL_ASFLOAT : BitConversion<IL_OP_MOV, GPRF32, IL_bitconv>; +defm IL_ASDOUBLE : BitConversion<IL_OP_MOV, GPRF64, IL_bitconv>; +defm IL_ASLONG : BitConversion<IL_OP_MOV, GPRI64, IL_bitconv>; +defm IL_ASV2CHAR : BitConversion<IL_OP_MOV, GPRV2I8, IL_bitconv>; +defm IL_ASV2SHORT : BitConversion<IL_OP_MOV, GPRV2I16, IL_bitconv>; +defm IL_ASV2INT : BitConversion<IL_OP_MOV, GPRV2I32, IL_bitconv>; +defm IL_ASV2FLOAT : BitConversion<IL_OP_MOV, GPRV2F32, IL_bitconv>; +defm IL_ASV2DOUBLE : BitConversion<IL_OP_MOV, GPRV2F64, IL_bitconv>; +defm IL_ASV2LONG : BitConversion<IL_OP_MOV, GPRV2I64, IL_bitconv>; +defm IL_ASV4CHAR : BitConversion<IL_OP_MOV, GPRV4I8, IL_bitconv>; +defm IL_ASV4SHORT : BitConversion<IL_OP_MOV, GPRV4I16, IL_bitconv>; +defm IL_ASV4INT : BitConversion<IL_OP_MOV, GPRV4I32, IL_bitconv>; +defm IL_ASV4FLOAT : BitConversion<IL_OP_MOV, GPRV4F32, IL_bitconv>; + +let Predicates = [Has32BitPtr] in { + let isCodeGenOnly=1 in { + //===----------------------------------------------------------------------===// + // Store Memory Operations + //===----------------------------------------------------------------------===// + defm GLOBALTRUNCSTORE : GTRUNCSTORE<"!global trunc store">; + defm GLOBALSTORE : STORE<"!global store" , global_store>; + defm LOCALTRUNCSTORE : LTRUNCSTORE<"!local trunc store">; + defm LOCALSTORE : STORE<"!local store" , local_store>; + defm PRIVATETRUNCSTORE : PTRUNCSTORE<"!private trunc store">; + defm PRIVATESTORE : STORE<"!private store" , private_store>; + defm REGIONTRUNCSTORE : RTRUNCSTORE<"!region trunc store">; + defm REGIONSTORE : STORE<"!region hw store" , region_store>; + + + //===---------------------------------------------------------------------===// + // Load Memory Operations + //===---------------------------------------------------------------------===// + defm GLOBALLOAD : LOAD<"!global load" , global_load>; + defm GLOBALZEXTLOAD : LOAD<"!global zext load" , global_zext_load>; + defm GLOBALSEXTLOAD : LOAD<"!global sext load" , global_sext_load>; + defm GLOBALAEXTLOAD : LOAD<"!global aext load" , global_aext_load>; + defm PRIVATELOAD : LOAD<"!private load" , private_load>; + defm PRIVATEZEXTLOAD : LOAD<"!private zext load" , private_zext_load>; + defm PRIVATESEXTLOAD : LOAD<"!private sext load" , private_sext_load>; + defm PRIVATEAEXTLOAD : LOAD<"!private aext load" , private_aext_load>; + defm CPOOLLOAD : LOAD<"!constant pool load" , cp_load>; + defm CPOOLZEXTLOAD : LOAD<"!constant pool zext load", cp_zext_load>; + defm CPOOLSEXTLOAD : LOAD<"!constant pool sext load", cp_sext_load>; + defm CPOOLAEXTLOAD : LOAD<"!constant aext pool load", cp_aext_load>; + defm CONSTANTLOAD : LOAD<"!constant load" , constant_load>; + defm CONSTANTZEXTLOAD : LOAD<"!constant zext load" , constant_zext_load>; + defm CONSTANTSEXTLOAD : LOAD<"!constant sext load" , constant_sext_load>; + defm CONSTANTAEXTLOAD : LOAD<"!constant aext load" , constant_aext_load>; + defm LOCALLOAD : LOAD<"!local load" , local_load>; + defm LOCALZEXTLOAD : LOAD<"!local zext load" , local_zext_load>; + defm LOCALSEXTLOAD : LOAD<"!local sext load" , local_sext_load>; + defm LOCALAEXTLOAD : LOAD<"!local aext load" , local_aext_load>; + defm REGIONLOAD : LOAD<"!region load" , region_load>; + defm REGIONZEXTLOAD : LOAD<"!region zext load" , region_zext_load>; + defm REGIONSEXTLOAD : LOAD<"!region sext load" , region_sext_load>; + defm REGIONAEXTLOAD : LOAD<"!region aext load" , region_aext_load>; + } + + + //===---------------------------------------------------------------------===// + // IO Expansion Load/Store Instructions + //===---------------------------------------------------------------------===// + let mayLoad = 1 in { + // This opcode has custom swizzle patterns for some of the arguments. + def SCRATCHLOAD : TwoInOneOut<IL_OP_MOV, (outs GPRV4I32:$dst), + (ins GPRI32:$addy, i32imm:$id), + !strconcat(IL_OP_MOV.Text, " $dst, x$id[$addy]"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def CBLOAD : TwoInOneOut<IL_OP_MOV, (outs GPRV4I32:$dst), + (ins GPRI32:$addy, i32imm:$id), + !strconcat(IL_OP_MOV.Text, " $dst, cb$id[$addy]"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def GDSLOAD : TwoInOneOut<IL_OP_GDS_LOAD, (outs GPRI32:$dst), + (ins GPRI32:$addy, i32imm:$id), + !strconcat(IL_OP_GDS_LOAD.Text, "_id($id) $dst, $addy"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def GDSLOAD_Y : TwoInOneOut<IL_OP_GDS_LOAD, (outs GPRI32:$dst), + (ins GPRI32:$addy, i32imm:$id), + !strconcat(IL_OP_GDS_LOAD.Text, "_id($id) $dst, $addy"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def GDSLOAD_Z : TwoInOneOut<IL_OP_GDS_LOAD, (outs GPRI32:$dst), + (ins GPRI32:$addy, i32imm:$id), + !strconcat(IL_OP_GDS_LOAD.Text, "_id($id) $dst, $addy"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def GDSLOAD_W : TwoInOneOut<IL_OP_GDS_LOAD, (outs GPRI32:$dst), + (ins GPRI32:$addy, i32imm:$id), + !strconcat(IL_OP_GDS_LOAD.Text, "_id($id) $dst, $addy"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def LDSLOADVEC : TwoInOneOut<IL_OP_LDS_LOAD_VEC, (outs GPRI32:$dst), + (ins GPRI32:$addy, i32imm:$id), + !strconcat(IL_OP_LDS_LOAD_VEC.Text, "_id($id) $dst, $addy, $addy"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def LDSLOADVEC_v2i32 : TwoInOneOut<IL_OP_LDS_LOAD_VEC, (outs GPRV2I32:$dst), + (ins GPRI32:$addy, i32imm:$id), + !strconcat(IL_OP_LDS_LOAD_VEC.Text, "_id($id) $dst, $addy, $addy"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def LDSLOADVEC_v4i32 : TwoInOneOut<IL_OP_LDS_LOAD_VEC, (outs GPRV4I32:$dst), + (ins GPRI32:$addy, i32imm:$id), + !strconcat(IL_OP_LDS_LOAD_VEC.Text, "_id($id) $dst, $addy, $addy"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def LDSLOAD : TwoInOneOut<IL_OP_LDS_LOAD, (outs GPRI32:$dst), + (ins GPRI32:$addy, i32imm:$id), + !strconcat(IL_OP_LDS_LOAD.Text, "_id($id) $dst, $addy"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def LDSLOAD_i8 : TwoInOneOut<IL_OP_LDS_LOAD_BYTE, (outs GPRI32:$dst), + (ins GPRI32:$addy, i32imm:$id), + !strconcat(IL_OP_LDS_LOAD_BYTE.Text, "_id($id) $dst, $addy"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def LDSLOAD_u8 : TwoInOneOut<IL_OP_LDS_LOAD_UBYTE, (outs GPRI32:$dst), + (ins GPRI32:$addy, i32imm:$id), + !strconcat(IL_OP_LDS_LOAD_UBYTE.Text, "_id($id) $dst, $addy"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def LDSLOAD_i16 : TwoInOneOut<IL_OP_LDS_LOAD_SHORT, (outs GPRI32:$dst), + (ins GPRI32:$addy, i32imm:$id), + !strconcat(IL_OP_LDS_LOAD_SHORT.Text, "_id($id) $dst, $addy"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def LDSLOAD_u16 : TwoInOneOut<IL_OP_LDS_LOAD_USHORT, (outs GPRI32:$dst), + (ins GPRI32:$addy, i32imm:$id), + !strconcat(IL_OP_LDS_LOAD_USHORT.Text, "_id($id) $dst, $addy"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def LDSLOAD_Y : TwoInOneOut<IL_OP_LDS_LOAD, (outs GPRI32:$dst), + (ins GPRI32:$addy, i32imm:$id), + !strconcat(IL_OP_LDS_LOAD.Text, "_id($id) $dst, $addy"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def LDSLOAD_Z : TwoInOneOut<IL_OP_LDS_LOAD, (outs GPRI32:$dst), + (ins GPRI32:$addy, i32imm:$id), + !strconcat(IL_OP_LDS_LOAD.Text, "_id($id) $dst, $addy"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def LDSLOAD_W : TwoInOneOut<IL_OP_LDS_LOAD, (outs GPRI32:$dst), + (ins GPRI32:$addy, i32imm:$id), + !strconcat(IL_OP_LDS_LOAD.Text, "_id($id) $dst, $addy"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def UAVARENALOAD_i8 : TwoInOneOut<IL_OP_ARENA_UAV_LOAD, (outs GPRI32:$dst), + (ins GPRI32:$addy, i32imm:$id), + !strconcat(IL_OP_ARENA_UAV_LOAD.Text, "_id($id)_size(byte) $dst, $addy"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def UAVARENALOAD_i16 : TwoInOneOut<IL_OP_ARENA_UAV_LOAD, (outs GPRI32:$dst), + (ins GPRI32:$addy, i32imm:$id), + !strconcat(IL_OP_ARENA_UAV_LOAD.Text, "_id($id)_size(short) $dst, $addy"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def UAVARENALOAD_i32 : TwoInOneOut<IL_OP_ARENA_UAV_LOAD, (outs GPRI32:$dst), + (ins GPRI32:$addy, i32imm:$id), + !strconcat(IL_OP_ARENA_UAV_LOAD.Text, "_id($id)_size(dword) $dst, $addy"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def UAVARENALOAD_Y_i32 : TwoInOneOut<IL_OP_ARENA_UAV_LOAD, (outs GPRI32:$dst), + (ins GPRI32:$addy, i32imm:$id), + !strconcat(IL_OP_ARENA_UAV_LOAD.Text, "_id($id)_size(dword) $dst, $addy"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def UAVARENALOAD_Z_i32 : TwoInOneOut<IL_OP_ARENA_UAV_LOAD, (outs GPRI32:$dst), + (ins GPRI32:$addy, i32imm:$id), + !strconcat(IL_OP_ARENA_UAV_LOAD.Text, "_id($id)_size(dword) $dst, $addy"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def UAVARENALOAD_W_i32 : TwoInOneOut<IL_OP_ARENA_UAV_LOAD, (outs GPRI32:$dst), + (ins GPRI32:$addy, i32imm:$id), + !strconcat(IL_OP_ARENA_UAV_LOAD.Text, "_id($id)_size(dword) $dst, $addy"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def UAVRAWLOAD_i32 : TwoInOneOut<IL_OP_RAW_UAV_LOAD, (outs GPRI32:$dst), + (ins GPRI32:$addy, i32imm:$id), + !strconcat(IL_OP_RAW_UAV_LOAD.Text, "_id($id) $dst, $addy"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def UAVRAWLOAD_v2i32 : TwoInOneOut<IL_OP_RAW_UAV_LOAD, (outs GPRV2I32:$dst), + (ins GPRI32:$addy, i32imm:$id), + !strconcat(IL_OP_RAW_UAV_LOAD.Text, "_id($id) $dst, $addy"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def UAVRAWLOAD_v4i32 : TwoInOneOut<IL_OP_RAW_UAV_LOAD, (outs GPRV4I32:$dst), + (ins GPRI32:$addy, i32imm:$id), + !strconcat(IL_OP_RAW_UAV_LOAD.Text, "_id($id) $dst, $addy"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def UAVRAWLOADCACHED_i32 : TwoInOneOut<IL_OP_RAW_UAV_LOAD, (outs GPRI32:$dst), + (ins GPRI32:$addy, i32imm:$id), + !strconcat(IL_OP_RAW_UAV_LOAD.Text, "_id($id)_cached $dst, $addy"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def UAVRAWLOADCACHED_v2i32 : TwoInOneOut<IL_OP_RAW_UAV_LOAD, (outs GPRV2I32:$dst), + (ins GPRI32:$addy, i32imm:$id), + !strconcat(IL_OP_RAW_UAV_LOAD.Text, "_id($id)_cached $dst, $addy"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def UAVRAWLOADCACHED_v4i32 : TwoInOneOut<IL_OP_RAW_UAV_LOAD, (outs GPRV4I32:$dst), + (ins GPRI32:$addy, i32imm:$id), + !strconcat(IL_OP_RAW_UAV_LOAD.Text, "_id($id)_cached $dst, $addy"), []>; + } + let mayStore = 1 in { + // This opcode has custom swizzle patterns for some of the arguments. + def SCRATCHSTORE : TwoInOneOut<IL_OP_MOV, (outs GPRI32:$addy), + (ins GPRV4I32:$data, i32imm:$id), + !strconcat(IL_OP_MOV.Text, " x$id[$addy], $data"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def SCRATCHSTORE_X : TwoInOneOut<IL_OP_MOV, (outs GPRI32:$addy), + (ins GPRI32:$data, i32imm:$id), + !strconcat(IL_OP_MOV.Text, " x$id[$addy].x___, $data"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def SCRATCHSTORE_Y : TwoInOneOut<IL_OP_MOV, (outs GPRI32:$addy), + (ins GPRI32:$data, i32imm:$id), + !strconcat(IL_OP_MOV.Text, " x$id[$addy]._y__, $data"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def SCRATCHSTORE_Z : TwoInOneOut<IL_OP_MOV, (outs GPRI32:$addy), + (ins GPRI32:$data, i32imm:$id), + !strconcat(IL_OP_MOV.Text, " x$id[$addy].__z_, $data"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def SCRATCHSTORE_W : TwoInOneOut<IL_OP_MOV, (outs GPRI32:$addy), + (ins GPRI32:$data, i32imm:$id), + !strconcat(IL_OP_MOV.Text, " x$id[$addy].___w, $data"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def SCRATCHSTORE_XY : TwoInOneOut<IL_OP_MOV, (outs GPRI32:$addy), + (ins GPRV2I32:$data, i32imm:$id), + !strconcat(IL_OP_MOV.Text, " x$id[$addy].xy__, $data"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def SCRATCHSTORE_ZW : TwoInOneOut<IL_OP_MOV, (outs GPRI32:$addy), + (ins GPRV2I32:$data, i32imm:$id), + !strconcat(IL_OP_MOV.Text, " x$id[$addy].__zw, $data"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def GDSSTORE : TwoInOneOut<IL_OP_GDS_STORE, (outs GPRI32:$addy), + (ins GPRI32:$src, i32imm:$id), + !strconcat(IL_OP_GDS_STORE.Text, "_id($id) $addy, $src"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def GDSSTORE_Y : TwoInOneOut<IL_OP_GDS_STORE, (outs GPRI32:$addy), + (ins GPRI32:$src, i32imm:$id), + !strconcat(IL_OP_GDS_STORE.Text, "_id($id) $addy, $src"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def GDSSTORE_Z : TwoInOneOut<IL_OP_GDS_STORE, (outs GPRI32:$addy), + (ins GPRI32:$src, i32imm:$id), + !strconcat(IL_OP_GDS_STORE.Text, "_id($id) $addy, $src"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def GDSSTORE_W : TwoInOneOut<IL_OP_GDS_STORE, (outs GPRI32:$addy), + (ins GPRI32:$src, i32imm:$id), + !strconcat(IL_OP_GDS_STORE.Text, "_id($id) $addy, $src"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def LDSSTOREVEC : ThreeInOneOut<IL_OP_LDS_STORE_VEC, (outs GPRI32:$mem), + (ins GPRI32:$addy, GPRI32:$src, i32imm:$id), + !strconcat(IL_OP_LDS_STORE_VEC.Text, "_id($id) $mem, $addy, $src, $src"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def LDSSTOREVEC_v2i32 : ThreeInOneOut<IL_OP_LDS_STORE_VEC, (outs GPRV2I32:$mem), + (ins GPRI32:$addy, GPRV2I32:$src, i32imm:$id), + !strconcat(IL_OP_LDS_STORE_VEC.Text, "_id($id) $mem, $addy, $src, $src"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def LDSSTOREVEC_v4i32 : ThreeInOneOut<IL_OP_LDS_STORE_VEC, (outs GPRV4I32:$mem), + (ins GPRI32:$addy, GPRV4I32:$src, i32imm:$id), + !strconcat(IL_OP_LDS_STORE_VEC.Text, "_id($id) $mem, $addy, $src, $src"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def LDSSTORE : TwoInOneOut<IL_OP_LDS_STORE, (outs GPRI32:$addy), + (ins GPRI32:$src, i32imm:$id), + !strconcat(IL_OP_LDS_STORE.Text, "_id($id) $addy, $src"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def LDSSTORE_i8 : TwoInOneOut<IL_OP_LDS_STORE_BYTE, (outs GPRI32:$addy), + (ins GPRI32:$src, i32imm:$id), + !strconcat(IL_OP_LDS_STORE_BYTE.Text, "_id($id) $addy, $src"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def LDSSTORE_i16 : TwoInOneOut<IL_OP_LDS_STORE_SHORT, (outs GPRI32:$addy), + (ins GPRI32:$src, i32imm:$id), + !strconcat(IL_OP_LDS_STORE_SHORT.Text, "_id($id) $addy, $src"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def LDSSTORE_Y : TwoInOneOut<IL_OP_LDS_STORE, (outs GPRI32:$addy), + (ins GPRI32:$src, i32imm:$id), + !strconcat(IL_OP_LDS_STORE.Text, "_id($id) $addy, $src"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def LDSSTORE_Z : TwoInOneOut<IL_OP_LDS_STORE, (outs GPRI32:$addy), + (ins GPRI32:$src, i32imm:$id), + !strconcat(IL_OP_LDS_STORE.Text, "_id($id) $addy, $src"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def LDSSTORE_W : TwoInOneOut<IL_OP_LDS_STORE, (outs GPRI32:$addy), + (ins GPRI32:$src, i32imm:$id), + !strconcat(IL_OP_LDS_STORE.Text, "_id($id) $addy, $src"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def UAVARENASTORE_i8 : TwoInOneOut<IL_OP_ARENA_UAV_STORE, (outs GPRI32:$addy), + (ins GPRI8:$src, i32imm:$id), + !strconcat(IL_OP_ARENA_UAV_STORE.Text, + "_id($id)_size(byte) $addy, $src"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def UAVARENASTORE_i16 : TwoInOneOut<IL_OP_ARENA_UAV_STORE, (outs GPRI32:$addy), + (ins GPRI16:$src, i32imm:$id), + !strconcat(IL_OP_ARENA_UAV_STORE.Text, + "_id($id)_size(short) $addy, $src"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def UAVARENASTORE_i32 : TwoInOneOut<IL_OP_ARENA_UAV_STORE, (outs GPRI32:$addy), + (ins GPRI32:$src, i32imm:$id), + !strconcat(IL_OP_ARENA_UAV_STORE.Text, + "_id($id)_size(dword) $addy, $src"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def UAVARENASTORE_Y_i32 : TwoInOneOut<IL_OP_ARENA_UAV_STORE, (outs GPRI32:$addy), + (ins GPRI32:$src, i32imm:$id), + !strconcat(IL_OP_ARENA_UAV_STORE.Text, + "_id($id)_size(dword) $addy, $src"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def UAVARENASTORE_Z_i32 : TwoInOneOut<IL_OP_ARENA_UAV_STORE, (outs GPRI32:$addy), + (ins GPRI32:$src, i32imm:$id), + !strconcat(IL_OP_ARENA_UAV_STORE.Text, + "_id($id)_size(dword) $addy, $src"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def UAVARENASTORE_W_i32 : TwoInOneOut<IL_OP_ARENA_UAV_STORE, (outs GPRI32:$addy), + (ins GPRI32:$src, i32imm:$id), + !strconcat(IL_OP_ARENA_UAV_STORE.Text, + "_id($id)_size(dword) $addy, $src"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def UAVRAWSTORE_i32 : TwoInOneOut<IL_OP_RAW_UAV_STORE, (outs GPRI32:$mem), + (ins GPRI32:$addy, GPRI32:$src, i32imm:$id), + !strconcat(IL_OP_RAW_UAV_STORE.Text, "_id($id) $mem, $addy, $src"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def UAVRAWSTORE_v2i32 : TwoInOneOut<IL_OP_RAW_UAV_STORE, (outs GPRV2I32:$mem), + (ins GPRI32:$addy, GPRV2I32:$src, i32imm:$id), + !strconcat(IL_OP_RAW_UAV_STORE.Text, "_id($id) $mem, $addy, $src"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def UAVRAWSTORE_v4i32 : TwoInOneOut<IL_OP_RAW_UAV_STORE, (outs GPRV4I32:$mem), + (ins GPRI32:$addy, GPRV4I32:$src, i32imm:$id), + !strconcat(IL_OP_RAW_UAV_STORE.Text, "_id($id) $mem, $addy, $src"), []>; + } +} +let Predicates = [Has64BitPtr] in { + let isCodeGenOnly=1 in { + //===----------------------------------------------------------------------===// + // Store Memory Operations + //===----------------------------------------------------------------------===// + defm GLOBALTRUNCSTORE64 : GTRUNCSTORE64<"!global trunc store">; + defm GLOBALSTORE64 : STORE64<"!global store" , global_store>; + defm LOCALTRUNCSTORE64 : LTRUNCSTORE64<"!local trunc store">; + defm LOCALSTORE64 : STORE64<"!local store" , local_store>; + defm PRIVATETRUNCSTORE64 : PTRUNCSTORE64<"!private trunc store">; + defm PRIVATESTORE64 : STORE64<"!private store" , private_store>; + defm REGIONTRUNCSTORE64 : RTRUNCSTORE64<"!region trunc store">; + defm REGIONSTORE64 : STORE64<"!region hw store" , region_store>; + + + //===---------------------------------------------------------------------===// + // Load Memory Operations + //===---------------------------------------------------------------------===// + defm GLOBALLOAD64 : LOAD64<"!global load" , global_load>; + defm GLOBALZEXTLOAD64 : LOAD64<"!global zext load" , global_zext_load>; + defm GLOBALSEXTLOAD64 : LOAD64<"!global sext load" , global_sext_load>; + defm GLOBALAEXTLOAD64 : LOAD64<"!global aext load" , global_aext_load>; + defm PRIVATELOAD64 : LOAD64<"!private load" , private_load>; + defm PRIVATEZEXTLOAD64 : LOAD64<"!private zext load" , private_zext_load>; + defm PRIVATESEXTLOAD64 : LOAD64<"!private sext load" , private_sext_load>; + defm PRIVATEAEXTLOAD64 : LOAD64<"!private aext load" , private_aext_load>; + defm CPOOLLOAD64 : LOAD64<"!constant pool load" , cp_load>; + defm CPOOLZEXTLOAD64 : LOAD64<"!constant pool zext load", cp_zext_load>; + defm CPOOLSEXTLOAD64 : LOAD64<"!constant pool sext load", cp_sext_load>; + defm CPOOLAEXTLOAD64 : LOAD64<"!constant aext pool load", cp_aext_load>; + defm CONSTANTLOAD64 : LOAD64<"!constant load" , constant_load>; + defm CONSTANTZEXTLOAD64 : LOAD64<"!constant zext load" , constant_zext_load>; + defm CONSTANTSEXTLOAD64 : LOAD64<"!constant sext load" , constant_sext_load>; + defm CONSTANTAEXTLOAD64 : LOAD64<"!constant aext load" , constant_aext_load>; + defm LOCALLOAD64 : LOAD64<"!local load" , local_load>; + defm LOCALZEXTLOAD64 : LOAD64<"!local zext load" , local_zext_load>; + defm LOCALSEXTLOAD64 : LOAD64<"!local sext load" , local_sext_load>; + defm LOCALAEXTLOAD64 : LOAD64<"!local aext load" , local_aext_load>; + defm REGIONLOAD64 : LOAD64<"!region load" , region_load>; + defm REGIONZEXTLOAD64 : LOAD64<"!region zext load" , region_zext_load>; + defm REGIONSEXTLOAD64 : LOAD64<"!region sext load" , region_sext_load>; + defm REGIONAEXTLOAD64 : LOAD64<"!region aext load" , region_aext_load>; + } + + + //===---------------------------------------------------------------------===// + // IO Expansion Load/Store Instructions + //===---------------------------------------------------------------------===// + let mayLoad = 1 in { + // This opcode has custom swizzle patterns for some of the arguments. + def SCRATCHLOAD64 : TwoInOneOut<IL_OP_MOV, (outs GPRV4I32:$dst), + (ins GPRI32:$addy, i64imm:$id), + !strconcat(IL_OP_MOV.Text, " $dst, x$id[$addy]"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def CBLOAD64 : TwoInOneOut<IL_OP_MOV, (outs GPRV4I32:$dst), + (ins GPRI32:$addy, i64imm:$id), + !strconcat(IL_OP_MOV.Text, " $dst, cb$id[$addy]"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def GDSLOAD64 : TwoInOneOut<IL_OP_GDS_LOAD, (outs GPRI32:$dst), + (ins GPRI32:$addy, i64imm:$id), + !strconcat(IL_OP_GDS_LOAD.Text, "_id($id) $dst, $addy"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def GDSLOAD64_Y : TwoInOneOut<IL_OP_GDS_LOAD, (outs GPRI32:$dst), + (ins GPRI32:$addy, i64imm:$id), + !strconcat(IL_OP_GDS_LOAD.Text, "_id($id) $dst, $addy"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def GDSLOAD64_Z : TwoInOneOut<IL_OP_GDS_LOAD, (outs GPRI32:$dst), + (ins GPRI32:$addy, i64imm:$id), + !strconcat(IL_OP_GDS_LOAD.Text, "_id($id) $dst, $addy"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def GDSLOAD64_W : TwoInOneOut<IL_OP_GDS_LOAD, (outs GPRI32:$dst), + (ins GPRI32:$addy, i64imm:$id), + !strconcat(IL_OP_GDS_LOAD.Text, "_id($id) $dst, $addy"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def LDSLOADVEC64 : TwoInOneOut<IL_OP_LDS_LOAD_VEC, (outs GPRI32:$dst), + (ins GPRI32:$addy, i64imm:$id), + !strconcat(IL_OP_LDS_LOAD_VEC.Text, "_id($id) $dst, $addy, $addy"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def LDSLOADVEC64_v2i32 : TwoInOneOut<IL_OP_LDS_LOAD_VEC, (outs GPRV2I32:$dst), + (ins GPRI32:$addy, i64imm:$id), + !strconcat(IL_OP_LDS_LOAD_VEC.Text, "_id($id) $dst, $addy, $addy"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def LDSLOADVEC64_v4i32 : TwoInOneOut<IL_OP_LDS_LOAD_VEC, (outs GPRV4I32:$dst), + (ins GPRI32:$addy, i64imm:$id), + !strconcat(IL_OP_LDS_LOAD_VEC.Text, "_id($id) $dst, $addy, $addy"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def LDSLOAD64 : TwoInOneOut<IL_OP_LDS_LOAD, (outs GPRI32:$dst), + (ins GPRI32:$addy, i64imm:$id), + !strconcat(IL_OP_LDS_LOAD.Text, "_id($id) $dst, $addy"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def LDSLOAD64_i8 : TwoInOneOut<IL_OP_LDS_LOAD_BYTE, (outs GPRI32:$dst), + (ins GPRI32:$addy, i64imm:$id), + !strconcat(IL_OP_LDS_LOAD_BYTE.Text, "_id($id) $dst, $addy"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def LDSLOAD64_u8 : TwoInOneOut<IL_OP_LDS_LOAD_UBYTE, (outs GPRI32:$dst), + (ins GPRI32:$addy, i64imm:$id), + !strconcat(IL_OP_LDS_LOAD_UBYTE.Text, "_id($id) $dst, $addy"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def LDSLOAD64_i16 : TwoInOneOut<IL_OP_LDS_LOAD_SHORT, (outs GPRI32:$dst), + (ins GPRI32:$addy, i64imm:$id), + !strconcat(IL_OP_LDS_LOAD_SHORT.Text, "_id($id) $dst, $addy"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def LDSLOAD64_u16 : TwoInOneOut<IL_OP_LDS_LOAD_USHORT, (outs GPRI32:$dst), + (ins GPRI32:$addy, i64imm:$id), + !strconcat(IL_OP_LDS_LOAD_USHORT.Text, "_id($id) $dst, $addy"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def LDSLOAD64_Y : TwoInOneOut<IL_OP_LDS_LOAD, (outs GPRI32:$dst), + (ins GPRI32:$addy, i64imm:$id), + !strconcat(IL_OP_LDS_LOAD.Text, "_id($id) $dst, $addy"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def LDSLOAD64_Z : TwoInOneOut<IL_OP_LDS_LOAD, (outs GPRI32:$dst), + (ins GPRI32:$addy, i64imm:$id), + !strconcat(IL_OP_LDS_LOAD.Text, "_id($id) $dst, $addy"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def LDSLOAD64_W : TwoInOneOut<IL_OP_LDS_LOAD, (outs GPRI32:$dst), + (ins GPRI32:$addy, i64imm:$id), + !strconcat(IL_OP_LDS_LOAD.Text, "_id($id) $dst, $addy"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def UAVARENALOAD64_i8 : TwoInOneOut<IL_OP_ARENA_UAV_LOAD, (outs GPRI32:$dst), + (ins GPRI32:$addy, i64imm:$id), + !strconcat(IL_OP_ARENA_UAV_LOAD.Text, "_id($id)_size(byte) $dst, $addy"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def UAVARENALOAD64_i16 : TwoInOneOut<IL_OP_ARENA_UAV_LOAD, (outs GPRI32:$dst), + (ins GPRI32:$addy, i64imm:$id), + !strconcat(IL_OP_ARENA_UAV_LOAD.Text, "_id($id)_size(short) $dst, $addy"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def UAVARENALOAD64_i32 : TwoInOneOut<IL_OP_ARENA_UAV_LOAD, (outs GPRI32:$dst), + (ins GPRI32:$addy, i64imm:$id), + !strconcat(IL_OP_ARENA_UAV_LOAD.Text, "_id($id)_size(dword) $dst, $addy"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def UAVARENALOAD64_Y_i32 : TwoInOneOut<IL_OP_ARENA_UAV_LOAD, (outs GPRI32:$dst), + (ins GPRI32:$addy, i64imm:$id), + !strconcat(IL_OP_ARENA_UAV_LOAD.Text, "_id($id)_size(dword) $dst, $addy"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def UAVARENALOAD64_Z_i32 : TwoInOneOut<IL_OP_ARENA_UAV_LOAD, (outs GPRI32:$dst), + (ins GPRI32:$addy, i64imm:$id), + !strconcat(IL_OP_ARENA_UAV_LOAD.Text, "_id($id)_size(dword) $dst, $addy"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def UAVARENALOAD64_W_i32 : TwoInOneOut<IL_OP_ARENA_UAV_LOAD, (outs GPRI32:$dst), + (ins GPRI32:$addy, i64imm:$id), + !strconcat(IL_OP_ARENA_UAV_LOAD.Text, "_id($id)_size(dword) $dst, $addy"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def UAVRAWLOAD64_i32 : TwoInOneOut<IL_OP_RAW_UAV_LOAD, (outs GPRI32:$dst), + (ins GPRI32:$addy, i64imm:$id), + !strconcat(IL_OP_RAW_UAV_LOAD.Text, "_id($id) $dst, $addy"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def UAVRAWLOAD64_v2i32 : TwoInOneOut<IL_OP_RAW_UAV_LOAD, (outs GPRV2I32:$dst), + (ins GPRI32:$addy, i64imm:$id), + !strconcat(IL_OP_RAW_UAV_LOAD.Text, "_id($id) $dst, $addy"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def UAVRAWLOAD64_v4i32 : TwoInOneOut<IL_OP_RAW_UAV_LOAD, (outs GPRV4I32:$dst), + (ins GPRI32:$addy, i64imm:$id), + !strconcat(IL_OP_RAW_UAV_LOAD.Text, "_id($id) $dst, $addy"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def UAVRAWLOADCACHED64_i32 : TwoInOneOut<IL_OP_RAW_UAV_LOAD, (outs GPRI32:$dst), + (ins GPRI32:$addy, i64imm:$id), + !strconcat(IL_OP_RAW_UAV_LOAD.Text, "_id($id)_cached $dst, $addy"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def UAVRAWLOADCACHED64_v2i32 : TwoInOneOut<IL_OP_RAW_UAV_LOAD, (outs GPRV2I32:$dst), + (ins GPRI32:$addy, i64imm:$id), + !strconcat(IL_OP_RAW_UAV_LOAD.Text, "_id($id)_cached $dst, $addy"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def UAVRAWLOADCACHED64_v4i32 : TwoInOneOut<IL_OP_RAW_UAV_LOAD, (outs GPRV4I32:$dst), + (ins GPRI32:$addy, i64imm:$id), + !strconcat(IL_OP_RAW_UAV_LOAD.Text, "_id($id)_cached $dst, $addy"), []>; + } + let mayStore = 1 in { + // This opcode has custom swizzle patterns for some of the arguments. + def SCRATCHSTORE64 : TwoInOneOut<IL_OP_MOV, (outs GPRI32:$addy), + (ins GPRV4I32:$data, i64imm:$id), + !strconcat(IL_OP_MOV.Text, " x$id[$addy], $data"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def SCRATCHSTORE64_X : TwoInOneOut<IL_OP_MOV, (outs GPRI32:$addy), + (ins GPRI32:$data, i64imm:$id), + !strconcat(IL_OP_MOV.Text, " x$id[$addy].x___, $data"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def SCRATCHSTORE64_Y : TwoInOneOut<IL_OP_MOV, (outs GPRI32:$addy), + (ins GPRI32:$data, i64imm:$id), + !strconcat(IL_OP_MOV.Text, " x$id[$addy]._y__, $data"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def SCRATCHSTORE64_Z : TwoInOneOut<IL_OP_MOV, (outs GPRI32:$addy), + (ins GPRI32:$data, i64imm:$id), + !strconcat(IL_OP_MOV.Text, " x$id[$addy].__z_, $data"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def SCRATCHSTORE64_W : TwoInOneOut<IL_OP_MOV, (outs GPRI32:$addy), + (ins GPRI32:$data, i64imm:$id), + !strconcat(IL_OP_MOV.Text, " x$id[$addy].___w, $data"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def SCRATCHSTORE64_XY : TwoInOneOut<IL_OP_MOV, (outs GPRI32:$addy), + (ins GPRV2I32:$data, i64imm:$id), + !strconcat(IL_OP_MOV.Text, " x$id[$addy].xy__, $data"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def SCRATCHSTORE64_ZW : TwoInOneOut<IL_OP_MOV, (outs GPRI32:$addy), + (ins GPRV2I32:$data, i64imm:$id), + !strconcat(IL_OP_MOV.Text, " x$id[$addy].__zw, $data"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def GDSSTORE64 : TwoInOneOut<IL_OP_GDS_STORE, (outs GPRI32:$addy), + (ins GPRI32:$src, i64imm:$id), + !strconcat(IL_OP_GDS_STORE.Text, "_id($id) $addy, $src"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def GDSSTORE64_Y : TwoInOneOut<IL_OP_GDS_STORE, (outs GPRI32:$addy), + (ins GPRI32:$src, i64imm:$id), + !strconcat(IL_OP_GDS_STORE.Text, "_id($id) $addy, $src"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def GDSSTORE64_Z : TwoInOneOut<IL_OP_GDS_STORE, (outs GPRI32:$addy), + (ins GPRI32:$src, i64imm:$id), + !strconcat(IL_OP_GDS_STORE.Text, "_id($id) $addy, $src"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def GDSSTORE64_W : TwoInOneOut<IL_OP_GDS_STORE, (outs GPRI32:$addy), + (ins GPRI32:$src, i64imm:$id), + !strconcat(IL_OP_GDS_STORE.Text, "_id($id) $addy, $src"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def LDSSTOREVEC64 : ThreeInOneOut<IL_OP_LDS_STORE_VEC, (outs GPRI32:$mem), + (ins GPRI32:$addy, GPRI32:$src, i64imm:$id), + !strconcat(IL_OP_LDS_STORE_VEC.Text, "_id($id) $mem, $addy, $src, $src"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def LDSSTOREVEC64_v2i32 : ThreeInOneOut<IL_OP_LDS_STORE_VEC, (outs GPRV2I32:$mem), + (ins GPRI32:$addy, GPRV2I32:$src, i64imm:$id), + !strconcat(IL_OP_LDS_STORE_VEC.Text, "_id($id) $mem, $addy, $src, $src"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def LDSSTOREVEC64_v4i32 : ThreeInOneOut<IL_OP_LDS_STORE_VEC, (outs GPRV4I32:$mem), + (ins GPRI32:$addy, GPRV4I32:$src, i64imm:$id), + !strconcat(IL_OP_LDS_STORE_VEC.Text, "_id($id) $mem, $addy, $src, $src"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def LDSSTORE64 : TwoInOneOut<IL_OP_LDS_STORE, (outs GPRI32:$addy), + (ins GPRI32:$src, i64imm:$id), + !strconcat(IL_OP_LDS_STORE.Text, "_id($id) $addy, $src"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def LDSSTORE64_i8 : TwoInOneOut<IL_OP_LDS_STORE_BYTE, (outs GPRI32:$addy), + (ins GPRI32:$src, i64imm:$id), + !strconcat(IL_OP_LDS_STORE_BYTE.Text, "_id($id) $addy, $src"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def LDSSTORE64_i16 : TwoInOneOut<IL_OP_LDS_STORE_SHORT, (outs GPRI32:$addy), + (ins GPRI32:$src, i64imm:$id), + !strconcat(IL_OP_LDS_STORE_SHORT.Text, "_id($id) $addy, $src"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def LDSSTORE64_Y : TwoInOneOut<IL_OP_LDS_STORE, (outs GPRI32:$addy), + (ins GPRI32:$src, i64imm:$id), + !strconcat(IL_OP_LDS_STORE.Text, "_id($id) $addy, $src"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def LDSSTORE64_Z : TwoInOneOut<IL_OP_LDS_STORE, (outs GPRI32:$addy), + (ins GPRI32:$src, i64imm:$id), + !strconcat(IL_OP_LDS_STORE.Text, "_id($id) $addy, $src"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def LDSSTORE64_W : TwoInOneOut<IL_OP_LDS_STORE, (outs GPRI32:$addy), + (ins GPRI32:$src, i64imm:$id), + !strconcat(IL_OP_LDS_STORE.Text, "_id($id) $addy, $src"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def UAVARENASTORE64_i8 : TwoInOneOut<IL_OP_ARENA_UAV_STORE, (outs GPRI32:$addy), + (ins GPRI8:$src, i64imm:$id), + !strconcat(IL_OP_ARENA_UAV_STORE.Text, + "_id($id)_size(byte) $addy, $src"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def UAVARENASTORE64_i16 : TwoInOneOut<IL_OP_ARENA_UAV_STORE, (outs GPRI32:$addy), + (ins GPRI16:$src, i64imm:$id), + !strconcat(IL_OP_ARENA_UAV_STORE.Text, + "_id($id)_size(short) $addy, $src"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def UAVARENASTORE64_i32 : TwoInOneOut<IL_OP_ARENA_UAV_STORE, (outs GPRI32:$addy), + (ins GPRI32:$src, i64imm:$id), + !strconcat(IL_OP_ARENA_UAV_STORE.Text, + "_id($id)_size(dword) $addy, $src"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def UAVARENASTORE64_Y_i32 : TwoInOneOut<IL_OP_ARENA_UAV_STORE, (outs GPRI32:$addy), + (ins GPRI32:$src, i64imm:$id), + !strconcat(IL_OP_ARENA_UAV_STORE.Text, + "_id($id)_size(dword) $addy, $src"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def UAVARENASTORE64_Z_i32 : TwoInOneOut<IL_OP_ARENA_UAV_STORE, (outs GPRI32:$addy), + (ins GPRI32:$src, i64imm:$id), + !strconcat(IL_OP_ARENA_UAV_STORE.Text, + "_id($id)_size(dword) $addy, $src"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def UAVARENASTORE64_W_i32 : TwoInOneOut<IL_OP_ARENA_UAV_STORE, (outs GPRI32:$addy), + (ins GPRI32:$src, i64imm:$id), + !strconcat(IL_OP_ARENA_UAV_STORE.Text, + "_id($id)_size(dword) $addy, $src"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def UAVRAWSTORE64_i32 : TwoInOneOut<IL_OP_RAW_UAV_STORE, (outs GPRI32:$mem), + (ins GPRI32:$addy, GPRI32:$src, i64imm:$id), + !strconcat(IL_OP_RAW_UAV_STORE.Text, "_id($id) $mem, $addy, $src"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def UAVRAWSTORE64_v2i32 : TwoInOneOut<IL_OP_RAW_UAV_STORE, (outs GPRV2I32:$mem), + (ins GPRI32:$addy, GPRV2I32:$src, i64imm:$id), + !strconcat(IL_OP_RAW_UAV_STORE.Text, "_id($id) $mem, $addy, $src"), []>; + // This opcode has custom swizzle patterns for some of the arguments. + def UAVRAWSTORE64_v4i32 : TwoInOneOut<IL_OP_RAW_UAV_STORE, (outs GPRV4I32:$mem), + (ins GPRI32:$addy, GPRV4I32:$src, i64imm:$id), + !strconcat(IL_OP_RAW_UAV_STORE.Text, "_id($id) $mem, $addy, $src"), []>; + } +} +//===---------------------------------------------------------------------===// +// Custom Inserter for Branches and returns, this eventually will be a +// seperate pass +//===---------------------------------------------------------------------===// +let isTerminator = 1 in { + def BRANCH : ILFormat<IL_PSEUDO_INST, (outs), (ins brtarget:$target), + "; Pseudo unconditional branch instruction", + [(br bb:$target)]>; + defm BRANCH_COND : BranchConditional<IL_brcond>; +} +//===---------------------------------------------------------------------===// +// return instructions +//===---------------------------------------------------------------------===// +let isTerminator = 1, isReturn = 1, isBarrier = 1, hasCtrlDep = 1 in { + def RETURN : ILFormat<IL_OP_RET,(outs), (ins variable_ops), + IL_OP_RET.Text, [(IL_retflag)]>; +} +//===---------------------------------------------------------------------===// +// Lower and raise the stack x amount +//===---------------------------------------------------------------------===// +def ADJCALLSTACKDOWN : ILFormat<IL_PSEUDO_INST, (outs), (ins i32imm:$amt), + "; begin of call sequence $amt", + [(IL_callseq_start timm:$amt)]>; +def ADJCALLSTACKUP : ILFormat<IL_PSEUDO_INST, (outs), (ins i32imm:$amt1, + i32imm:$amt2), + "; end of call sequence $amt1 $amt2", + [(IL_callseq_end timm:$amt1, timm:$amt2)]>; + +//===---------------------------------------------------------------------===// +// Handle a function call +//===---------------------------------------------------------------------===// +let isCall = 1, + Defs = [ + R110, R111, + R112, R113, R114, R115, R116, R117, R118, R119, R120, R121, R122, R123, R124, + R125, R126, R127, + R128, R129, R130, R131, R132, R133, R134, R135, R136, R137, R138, R139, R140, + R141, R142, R143, + R144, R145, R146, R147, R148, R149, R150, R151, R152, R153, R154, R155, R156, + R157, R158, R159, + R160, R161, R162, R163, R164, R165, R166, R167, R168, R169, R170, R171, R172, + R173, R174, R175, + R176, R177, R178, R179, R180, R181, R182, R183, R184, R185, R186, R187, R188, + R189, R190, R191, + R192, R193, R194, R195, R196, R197, R198, R199, R200, R201, R202, R203, R204, + R205, R206, R207, + R208, R209, R210, R211, R212, R213, R214, R215, R216, R217, R218, R219, R220, + R221, R222, R223, + R224, R225, R226, R227, R228, R229, R230, R231, R232, R233, R234, R235, R236, + R237, R238, R239, + R240, R241, R242, R243, R244, R245, R246, R247, R248, R249, R250, R251, R252, + R253, R254, R255 + ] + , + Uses = [ + R1, R2, R3, R4, R5, R6, R7, R8, R9, R10, R11, R12, R13, R14, R15, + R16, R17, R18, R19, R20, R21, R22, R23, R24, R25, R26, R27, R28, R29, R30, R31, + R32, R33, R34, R35, R36, R37, R38, R39, R40, R41, R42, R43, R44, R45, R46, R47, + R48, R49, R50, R51, R52, R53, R54, R55, R56, R57, R58, R59, R60, R61, R62, R63, + R64, R65, R66, R67, R68, R69, R70, R71, R72, R73, R74, R75, R76, R77, R78, R79, + R80, R81, R82, R83, R84, R85, R86, R87, R88, R89, R90, R91, R92, R93, R94, R95, + R96, R97, R98, R99, R100, R101, R102, R103, R104, R105, R106, R107, R108, R109 + ] + in { + def CALL : UnaryOpNoRet<IL_OP_CALL, (outs), + (ins calltarget:$dst, variable_ops), + !strconcat(IL_OP_CALL.Text, " $dst"), []>; + } + + +//===---------------------------------------------------------------------===// +// Flow and Program control Instructions +//===---------------------------------------------------------------------===// +let isTerminator=1 in { + def SWITCH : ILFormat<IL_OP_SWITCH, (outs), (ins GPRI32:$src), + !strconcat(IL_OP_SWITCH.Text, " $src"), []>; + def CASE : ILFormat<IL_OP_CASE, (outs), (ins GPRI32:$src), + !strconcat(IL_OP_CASE.Text, " $src"), []>; + def BREAK : ILFormat<IL_OP_BREAK, (outs), (ins), + IL_OP_BREAK.Text, []>; + def CONTINUE : ILFormat<IL_OP_CONTINUE, (outs), (ins), + IL_OP_CONTINUE.Text, []>; + def DEFAULT : ILFormat<IL_OP_DEFAULT, (outs), (ins), + IL_OP_DEFAULT.Text, []>; + def ELSE : ILFormat<IL_OP_ELSE, (outs), (ins), + IL_OP_ELSE.Text, []>; + def ENDSWITCH : ILFormat<IL_OP_ENDSWITCH, (outs), (ins), + IL_OP_ENDSWITCH.Text, []>; + def ENDMAIN : ILFormat<IL_OP_ENDMAIN, (outs), (ins), + IL_OP_ENDMAIN.Text, []>; + def END : ILFormat<IL_OP_END, (outs), (ins), + IL_OP_END.Text, []>; + def ENDFUNC : ILFormat<IL_OP_ENDFUNC, (outs), (ins), + IL_OP_ENDFUNC.Text, []>; + def ENDIF : ILFormat<IL_OP_ENDIF, (outs), (ins), + IL_OP_ENDIF.Text, []>; + def WHILELOOP : ILFormat<IL_OP_WHILE, (outs), (ins), + IL_OP_WHILE.Text, []>; + def ENDLOOP : ILFormat<IL_OP_ENDLOOP, (outs), (ins), + IL_OP_ENDLOOP.Text, []>; + def FUNC : ILFormat<IL_OP_FUNC, (outs), (ins), + IL_OP_FUNC.Text, []>; + def RETDYN : ILFormat<IL_OP_RET_DYN, (outs), (ins), + IL_OP_RET_DYN.Text, []>; + // This opcode has custom swizzle pattern encoded in Swizzle Encoder + defm IF_LOGICALNZ : BranchInstr<IL_OP_IF_LOGICALNZ>; + // This opcode has custom swizzle pattern encoded in Swizzle Encoder + defm IF_LOGICALZ : BranchInstr<IL_OP_IF_LOGICALZ>; + // This opcode has custom swizzle pattern encoded in Swizzle Encoder + defm BREAK_LOGICALNZ : BranchInstr<IL_OP_BREAK_LOGICALNZ>; + // This opcode has custom swizzle pattern encoded in Swizzle Encoder + defm BREAK_LOGICALZ : BranchInstr<IL_OP_BREAK_LOGICALZ>; + // This opcode has custom swizzle pattern encoded in Swizzle Encoder + defm CONTINUE_LOGICALNZ : BranchInstr<IL_OP_CONTINUE_LOGICALNZ>; + // This opcode has custom swizzle pattern encoded in Swizzle Encoder + defm CONTINUE_LOGICALZ : BranchInstr<IL_OP_CONTINUE_LOGICALZ>; + defm IFC : BranchInstr2<IL_OP_IFC>; + defm BREAKC : BranchInstr2<IL_OP_BREAKC>; + defm CONTINUEC : BranchInstr2<IL_OP_CONTINUEC>; +} +let isTerminator = 1, isBarrier = 1, hasCtrlDep = 1 in { + def TRAP : ILFormat<IL_OP_NOP, (outs), (ins), + IL_OP_NOP.Text, [(trap)]>; +} + +//===---------------------------------------------------------------------===// +//----------------- Work Item Functions - OpenCL 6.11.1 ---------------------// +//===---------------------------------------------------------------------===// +let isCall = 1, isAsCheapAsAMove = 1 in { + def GET_WORK_DIM : ILFormat<IL_OP_MOV, (outs GPRI32:$dst), (ins), + !strconcat(IL_OP_MOV.Text, " $dst, cb0[0].w"), + [(set GPRI32:$dst, (int_AMDIL_get_work_dim))]>; + + def GET_GLOBAL_ID : ILFormat<IL_OP_MOV, (outs GPRV4I32:$dst), + (ins), !strconcat(IL_OP_MOV.Text, " $dst, r1021.xyz0"), + [(set GPRV4I32:$dst, (int_AMDIL_get_global_id))]>; + + def GET_LOCAL_ID : ILFormat<IL_OP_MOV, (outs GPRV4I32:$dst), + (ins), !strconcat(IL_OP_MOV.Text, " $dst, r1022.xyz0"), + [(set GPRV4I32:$dst, (int_AMDIL_get_local_id))]>; + + def GET_GROUP_ID : ILFormat<IL_OP_MOV, (outs GPRV4I32:$dst), + (ins), !strconcat(IL_OP_MOV.Text, " $dst, r1023.xyz0"), + [(set GPRV4I32:$dst, (int_AMDIL_get_group_id))]>; + + def GET_GLOBAL_SIZE : ILFormat<IL_OP_MOV, (outs GPRV4I32:$dst), + (ins), !strconcat(IL_OP_MOV.Text, " $dst, cb0[0].xyz0"), + [(set GPRV4I32:$dst, (int_AMDIL_get_global_size))]>; + + def GET_LOCAL_SIZE : ILFormat<IL_OP_MOV, (outs GPRV4I32:$dst), + (ins), !strconcat(IL_OP_MOV.Text, " $dst, cb0[1].xyz0"), + [(set GPRV4I32:$dst, (int_AMDIL_get_local_size))]>; + + def GET_NUM_GROUPS : ILFormat<IL_OP_MOV, (outs GPRV4I32:$dst), + (ins), !strconcat(IL_OP_MOV.Text, " $dst, cb0[2].xyz0"), + [(set GPRV4I32:$dst, (int_AMDIL_get_num_groups))]>; + + def GET_GLOBAL_OFFSET : ILFormat<IL_OP_MOV, (outs GPRV4I32:$dst), + (ins), !strconcat(IL_OP_MOV.Text, " $dst, cb0[9].xyz0"), + [(set GPRV4I32:$dst, (int_AMDIL_get_global_offset))]>; + + let Predicates = [Has64BitPtr] in { + def GET_PRINTF_OFFSET_i64: ILFormat<IL_OP_MOV, (outs GPRI32:$dst), + (ins), !strconcat(IL_OP_MOV.Text, " $dst, cb0[8].zw"), + [(set GPRI32:$dst, (int_AMDIL_get_printf_offset))]>; + def GET_PRINTF_SIZE_i64 : ILFormat<IL_OP_MOV, (outs GPRI32:$dst), + (ins), !strconcat(IL_OP_MOV.Text, " $dst, cb0[9].x0"), + [(set GPRI32:$dst, (int_AMDIL_get_printf_size))]>; + } + let Predicates = [Has32BitPtr] in { + def GET_PRINTF_OFFSET_i32 : ILFormat<IL_OP_MOV, (outs GPRI32:$dst), + (ins), !strconcat(IL_OP_MOV.Text, " $dst, cb0[8].y0"), + [(set GPRI32:$dst, (int_AMDIL_get_printf_offset))]>; + def GET_PRINTF_SIZE_i32 : ILFormat<IL_OP_MOV, (outs GPRI32:$dst), + (ins), !strconcat(IL_OP_MOV.Text, " $dst, cb0[8].z0"), + [(set GPRI32:$dst, (int_AMDIL_get_printf_size))]>; + } +} +//===---------------------------------------------------------------------===// +//------------- Synchronization Functions - OpenCL 6.11.9 -------------------// +//===---------------------------------------------------------------------===// +let isCall=1 in { + + def FENCE : BinaryOpNoRet<IL_OP_FENCE, (outs), (ins GPRI32:$flag), + "fence_lds_memory_gds", + [(int_AMDIL_fence GPRI32:$flag)]>; + + def FENCE_LOCAL : BinaryOpNoRet<IL_OP_FENCE, (outs), (ins GPRI32:$flag), + "fence_lds", + [(int_AMDIL_fence_local GPRI32:$flag)]>; + + def FENCE_GLOBAL : BinaryOpNoRet<IL_OP_FENCE, (outs), (ins GPRI32:$flag), + "fence_memory", + [(int_AMDIL_fence_global GPRI32:$flag)]>; + + def FENCE_REGION : BinaryOpNoRet<IL_OP_FENCE, (outs), (ins GPRI32:$flag), + "fence_gds", + [(int_AMDIL_fence_region GPRI32:$flag)]>; + + def FENCE_READ_ONLY : BinaryOpNoRet<IL_OP_FENCE_READ_ONLY, (outs), + (ins GPRI32:$flag), + "fence_lds_gds_memory_mem_read_only", + [(int_AMDIL_fence_read_only GPRI32:$flag)]>; + + def FENCE_READ_ONLY_LOCAL : BinaryOpNoRet<IL_OP_FENCE_READ_ONLY, (outs), + (ins GPRI32:$flag), + "fence_lds_mem_read_only", + [(int_AMDIL_fence_read_only_local GPRI32:$flag)]>; + + def FENCE_READ_ONLY_GLOBAL : BinaryOpNoRet<IL_OP_FENCE_READ_ONLY, (outs), + (ins GPRI32:$flag), + "fence_mem_read_only", + [(int_AMDIL_fence_read_only_global GPRI32:$flag)]>; + + def FENCE_READ_ONLY_REGION : BinaryOpNoRet<IL_OP_FENCE_READ_ONLY, (outs), + (ins GPRI32:$flag), + "fence_gds_mem_read_only", + [(int_AMDIL_fence_read_only_region GPRI32:$flag)]>; + + def FENCE_WRITE_ONLY : BinaryOpNoRet<IL_OP_FENCE_WRITE_ONLY, (outs), + (ins GPRI32:$flag), + "fence_lds_gds_memory_mem_write_only", + [(int_AMDIL_fence_write_only GPRI32:$flag)]>; + + def FENCE_WRITE_ONLY_LOCAL : BinaryOpNoRet<IL_OP_FENCE_WRITE_ONLY, (outs), + (ins GPRI32:$flag), + "fence_lds_mem_write_only", + [(int_AMDIL_fence_write_only_local GPRI32:$flag)]>; + + def FENCE_WRITE_ONLY_GLOBAL : BinaryOpNoRet<IL_OP_FENCE_WRITE_ONLY, (outs), + (ins GPRI32:$flag), + "fence_mem_write_only", + [(int_AMDIL_fence_write_only_global GPRI32:$flag)]>; + + def FENCE_WRITE_ONLY_REGION : BinaryOpNoRet<IL_OP_FENCE_WRITE_ONLY, (outs), + (ins GPRI32:$flag), + "fence_gds_mem_write_only", + [(int_AMDIL_fence_write_only_region GPRI32:$flag)]>; +} +let isReturn = 1 in { + def EARLY_EXIT : UnaryOpNoRet<IL_OP_RET_LOGICALNZ, (outs), + (ins GPRI32:$flag), + !strconcat(IL_OP_RET_LOGICALNZ.Text, " $flag"), + [(int_AMDIL_early_exit GPRI32:$flag)]>; +} +def MEDIA_UNPACK_0 : OneInOneOut<IL_OP_UNPACK_0, (outs GPRV4F32:$dst), + (ins GPRV4I32:$src), + !strconcat(IL_OP_UNPACK_0.Text, " $dst, $src"), + [(set GPRV4F32:$dst, + (v4f32 (int_AMDIL_media_unpack_byte_0 GPRV4I32:$src)))]>; +def MEDIA_UNPACK_1 : OneInOneOut<IL_OP_UNPACK_1, (outs GPRV4F32:$dst), + (ins GPRV4I32:$src), + !strconcat(IL_OP_UNPACK_1.Text, " $dst, $src"), + [(set GPRV4F32:$dst, + (v4f32 (int_AMDIL_media_unpack_byte_1 GPRV4I32:$src)))]>; +def MEDIA_UNPACK_2 : OneInOneOut<IL_OP_UNPACK_2, (outs GPRV4F32:$dst), + (ins GPRV4I32:$src), + !strconcat(IL_OP_UNPACK_2.Text, " $dst, $src"), + [(set GPRV4F32:$dst, + (v4f32 (int_AMDIL_media_unpack_byte_2 GPRV4I32:$src)))]>; +def MEDIA_UNPACK_3 : OneInOneOut<IL_OP_UNPACK_3, (outs GPRV4F32:$dst), + (ins GPRV4I32:$src), + !strconcat(IL_OP_UNPACK_3.Text, " $dst, $src"), + [(set GPRV4F32:$dst, + (v4f32 (int_AMDIL_media_unpack_byte_3 GPRV4I32:$src)))]>; +let Predicates = [Has32BitPtr] in { +// All of the image functions +def IMAGE1D_READ : ILFormat<IL_OP_SAMPLE, (outs GPRV4I32:$dst), + (ins MEMI32:$ptr, GPRI32:$sampler, GPRV4F32:$addy), + !strconcat(IL_OP_SAMPLE.Text, + "_resource($ptr)_sampler($sampler)_coordtype(normalized) $dst, $addy"), + [(set GPRV4I32:$dst, + (int_AMDIL_image1d_read_norm ADDR:$ptr, GPRI32:$sampler, GPRV4F32:$addy))]>; +def IMAGE1DA_READ : ILFormat<IL_OP_SAMPLE, (outs GPRV4I32:$dst), + (ins MEMI32:$ptr, GPRI32:$sampler, GPRV4F32:$addy), + !strconcat(IL_OP_SAMPLE.Text, + "_resource($ptr)_sampler($sampler)_coordtype(normalized) $dst, $addy"), + [(set GPRV4I32:$dst, + (int_AMDIL_image1d_array_read_norm ADDR:$ptr, GPRI32:$sampler, GPRV4F32:$addy))]>; +def IMAGE2D_READ : ILFormat<IL_OP_SAMPLE, (outs GPRV4I32:$dst), + (ins MEMI32:$ptr, GPRI32:$sampler, GPRV4F32:$addy), + !strconcat(IL_OP_SAMPLE.Text, + "_resource($ptr)_sampler($sampler)_coordtype(normalized) $dst, $addy"), + [(set GPRV4I32:$dst, + (int_AMDIL_image2d_read_norm ADDR:$ptr, GPRI32:$sampler, GPRV4F32:$addy))]>; +def IMAGE2DA_READ : ILFormat<IL_OP_SAMPLE, (outs GPRV4I32:$dst), + (ins MEMI32:$ptr, GPRI32:$sampler, GPRV4F32:$addy), + !strconcat(IL_OP_SAMPLE.Text, + "_resource($ptr)_sampler($sampler)_coordtype(normalized) $dst, $addy"), + [(set GPRV4I32:$dst, + (int_AMDIL_image2d_array_read_norm ADDR:$ptr, GPRI32:$sampler, GPRV4F32:$addy))]>; +def IMAGE3D_READ : ILFormat<IL_OP_SAMPLE, (outs GPRV4I32:$dst), + (ins MEMI32:$ptr, GPRI32:$sampler, GPRV4F32:$addy), + !strconcat(IL_OP_SAMPLE.Text, + "_resource($ptr)_sampler($sampler)_coordtype(normalized) $dst, $addy"), + [(set GPRV4I32:$dst, + (int_AMDIL_image3d_read_norm ADDR:$ptr, GPRI32:$sampler, GPRV4F32:$addy))]>; +def IMAGE1D_READ_UNNORM : ILFormat<IL_OP_SAMPLE, (outs GPRV4I32:$dst), + (ins MEMI32:$ptr, GPRI32:$sampler, GPRV4F32:$addy), + !strconcat(IL_OP_SAMPLE.Text, + "_resource($ptr)_sampler($sampler)_coordtype(unnormalized) $dst, $addy"), + [(set GPRV4I32:$dst, + (int_AMDIL_image1d_read_unnorm ADDR:$ptr, GPRI32:$sampler, GPRV4F32:$addy))]>; +def IMAGE1DA_READ_UNNORM : ILFormat<IL_OP_SAMPLE, (outs GPRV4I32:$dst), + (ins MEMI32:$ptr, GPRI32:$sampler, GPRV4F32:$addy), + !strconcat(IL_OP_SAMPLE.Text, + "_resource($ptr)_sampler($sampler)_coordtype(unnormalized) $dst, $addy"), + [(set GPRV4I32:$dst, + (int_AMDIL_image1d_array_read_unnorm ADDR:$ptr, GPRI32:$sampler, GPRV4F32:$addy))]>; +def IMAGE2D_READ_UNNORM : ILFormat<IL_OP_SAMPLE, (outs GPRV4I32:$dst), + (ins MEMI32:$ptr, GPRI32:$sampler, GPRV4F32:$addy), + !strconcat(IL_OP_SAMPLE.Text, + "_resource($ptr)_sampler($sampler)_coordtype(unnormalized) $dst, $addy"), + [(set GPRV4I32:$dst, + (int_AMDIL_image2d_read_unnorm ADDR:$ptr, GPRI32:$sampler, GPRV4F32:$addy))]>; +def IMAGE2DA_READ_UNNORM : ILFormat<IL_OP_SAMPLE, (outs GPRV4I32:$dst), + (ins MEMI32:$ptr, GPRI32:$sampler, GPRV4F32:$addy), + !strconcat(IL_OP_SAMPLE.Text, + "_resource($ptr)_sampler($sampler)_coordtype(unnormalized) $dst, $addy"), + [(set GPRV4I32:$dst, + (int_AMDIL_image2d_array_read_unnorm ADDR:$ptr, GPRI32:$sampler, GPRV4F32:$addy))]>; +def IMAGE3D_READ_UNNORM : ILFormat<IL_OP_SAMPLE, (outs GPRV4I32:$dst), + (ins MEMI32:$ptr, GPRI32:$sampler, GPRV4F32:$addy), + !strconcat(IL_OP_SAMPLE.Text, + "_resource($ptr)_sampler($sampler)_coordtype(unnormalized) $dst, $addy"), + [(set GPRV4I32:$dst, + (int_AMDIL_image3d_read_unnorm ADDR:$ptr, GPRI32:$sampler, GPRV4F32:$addy))]>; +def IMAGE1D_INFO0 : ILFormat<IL_OP_MOV, (outs GPRV4I32:$dst), + (ins MEMI32:$ptr), + !strconcat(IL_OP_MOV.Text, " $dst, $ptr"), + [(set GPRV4I32:$dst, (int_AMDIL_image1d_info0 ADDR:$ptr))]>; +def IMAGE1D_INFO1 : ILFormat<IL_OP_MOV, (outs GPRV4I32:$dst), + (ins MEMI32:$ptr), + !strconcat(IL_OP_MOV.Text, " $dst, $ptr"), + [(set GPRV4I32:$dst, (int_AMDIL_image1d_info1 ADDR:$ptr))]>; +def IMAGE1DA_INFO0 : ILFormat<IL_OP_MOV, (outs GPRV4I32:$dst), + (ins MEMI32:$ptr), + !strconcat(IL_OP_MOV.Text, " $dst, $ptr"), + [(set GPRV4I32:$dst, (int_AMDIL_image1d_array_info0 ADDR:$ptr))]>; +def IMAGE1DA_INFO1 : ILFormat<IL_OP_MOV, (outs GPRV4I32:$dst), + (ins MEMI32:$ptr), + !strconcat(IL_OP_MOV.Text, " $dst, $ptr"), + [(set GPRV4I32:$dst, (int_AMDIL_image1d_array_info1 ADDR:$ptr))]>; +def IMAGE2D_INFO0 : ILFormat<IL_OP_MOV, (outs GPRV4I32:$dst), + (ins MEMI32:$ptr), + !strconcat(IL_OP_MOV.Text, " $dst, $ptr"), + [(set GPRV4I32:$dst, (int_AMDIL_image2d_info0 ADDR:$ptr))]>; +def IMAGE2D_INFO1 : ILFormat<IL_OP_MOV, (outs GPRV4I32:$dst), + (ins MEMI32:$ptr), + !strconcat(IL_OP_MOV.Text, " $dst, $ptr"), + [(set GPRV4I32:$dst, (int_AMDIL_image2d_info1 ADDR:$ptr))]>; +def IMAGE2DA_INFO0 : ILFormat<IL_OP_MOV, (outs GPRV4I32:$dst), + (ins MEMI32:$ptr), + !strconcat(IL_OP_MOV.Text, " $dst, $ptr"), + [(set GPRV4I32:$dst, (int_AMDIL_image2d_array_info0 ADDR:$ptr))]>; +def IMAGE2DA_INFO1 : ILFormat<IL_OP_MOV, (outs GPRV4I32:$dst), + (ins MEMI32:$ptr), + !strconcat(IL_OP_MOV.Text, " $dst, $ptr"), + [(set GPRV4I32:$dst, (int_AMDIL_image2d_array_info1 ADDR:$ptr))]>; +def IMAGE3D_INFO0 : ILFormat<IL_OP_MOV, (outs GPRV4I32:$dst), + (ins MEMI32:$ptr), + !strconcat(IL_OP_MOV.Text, " $dst, $ptr"), + [(set GPRV4I32:$dst, (int_AMDIL_image3d_info0 ADDR:$ptr))]>; +def IMAGE3D_INFO1 : ILFormat<IL_OP_MOV, (outs GPRV4I32:$dst), + (ins MEMI32:$ptr), + !strconcat(IL_OP_MOV.Text, " $dst, $ptr"), + [(set GPRV4I32:$dst, (int_AMDIL_image3d_info1 ADDR:$ptr))]>; +def IMAGE1D_WRITE : ILFormat<IL_OP_UAV_STORE, (outs), + (ins MEMI32:$ptr, GPRV2I32:$addy, GPRV4I32:$data), + !strconcat(IL_OP_UAV_STORE.Text, + "_id($ptr) $addy, $data"), + [(int_AMDIL_image1d_write ADDR:$ptr, GPRV2I32:$addy, GPRV4I32:$data)]>; +def IMAGE1DA_WRITE : ILFormat<IL_OP_UAV_STORE, (outs), + (ins MEMI32:$ptr, GPRV2I32:$addy, GPRV4I32:$data), + !strconcat(IL_OP_UAV_STORE.Text, + "_id($ptr) $addy, $data"), + [(int_AMDIL_image1d_array_write ADDR:$ptr, GPRV2I32:$addy, GPRV4I32:$data)]>; +def IMAGE2D_WRITE : ILFormat<IL_OP_UAV_STORE, (outs), + (ins MEMI32:$ptr, GPRV2I32:$addy, GPRV4I32:$data), + !strconcat(IL_OP_UAV_STORE.Text, + "_id($ptr) $addy, $data"), + [(int_AMDIL_image2d_write ADDR:$ptr, GPRV2I32:$addy, GPRV4I32:$data)]>; +def IMAGE2DA_WRITE : ILFormat<IL_OP_UAV_STORE, (outs), + (ins MEMI32:$ptr, GPRV2I32:$addy, GPRV4I32:$data), + !strconcat(IL_OP_UAV_STORE.Text, + "_id($ptr) $addy, $data"), + [(int_AMDIL_image2d_array_write ADDR:$ptr, GPRV2I32:$addy, GPRV4I32:$data)]>; +def IMAGE3D_WRITE : ILFormat<IL_OP_UAV_STORE, (outs), + (ins MEMI32:$ptr, GPRV4I32:$addy, GPRV4I32:$data), + !strconcat(IL_OP_UAV_STORE.Text, + "_id($ptr) $addy, $data"), + [(int_AMDIL_image3d_write ADDR:$ptr, GPRV4I32:$addy, GPRV4I32:$data)]>; +let hasSideEffects = 1, isNotDuplicable = 1 in { + // All of the noret atomic functions + def ATOM_G_ADD_NORET : BinAtomNoRet<IL_OP_UAV_ADD, + "_id($id)", atom_g_add_noret>; + def ATOM_G_AND_NORET : BinAtomNoRet<IL_OP_UAV_AND, + "_id($id)", atom_g_and_noret>; + def ATOM_G_MAX_NORET : BinAtomNoRet<IL_OP_UAV_MAX, + "_id($id)", atom_g_max_noret>; + def ATOM_G_MIN_NORET : BinAtomNoRet<IL_OP_UAV_MIN, + "_id($id)", atom_g_min_noret>; + def ATOM_G_UMAX_NORET : BinAtomNoRet<IL_OP_UAV_UMAX, + "_id($id)", atom_g_umax_noret>; + def ATOM_G_UMIN_NORET : BinAtomNoRet<IL_OP_UAV_UMIN, + "_id($id)", atom_g_umin_noret>; + def ATOM_G_OR_NORET : BinAtomNoRet<IL_OP_UAV_OR, + "_id($id)", atom_g_or_noret>; + def ATOM_G_RSUB_NORET : BinAtomNoRet<IL_OP_UAV_RSUB, + "_id($id)", atom_g_rsub_noret>; + def ATOM_G_SUB_NORET : BinAtomNoRet<IL_OP_UAV_SUB, + "_id($id)", atom_g_sub_noret>; + def ATOM_G_XOR_NORET : BinAtomNoRet<IL_OP_UAV_XOR, + "_id($id)", atom_g_xor_noret>; + def ATOM_G_INC_NORET : BinAtomNoRet<IL_OP_UAV_INC, + "_id($id)", atom_g_inc_noret>; + def ATOM_G_DEC_NORET : BinAtomNoRet<IL_OP_UAV_DEC, + "_id($id)", atom_g_dec_noret>; + def ATOM_G_CMPXCHG_NORET : CmpXChgNoRet<IL_OP_UAV_CMP, + "_id($id)", atom_g_cmpxchg_noret>; + def ATOM_A_ADD_NORET : BinAtomNoRet<IL_OP_UAV_ADD, + "_id($id)_arena", atom_g_add_noret>; + def ATOM_A_AND_NORET : BinAtomNoRet<IL_OP_UAV_AND, + "_id($id)_arena", atom_g_and_noret>; + def ATOM_A_MAX_NORET : BinAtomNoRet<IL_OP_UAV_MAX, + "_id($id)_arena", atom_g_max_noret>; + def ATOM_A_MIN_NORET : BinAtomNoRet<IL_OP_UAV_MIN, + "_id($id)_arena", atom_g_min_noret>; + def ATOM_A_UMAX_NORET : BinAtomNoRet<IL_OP_UAV_UMAX, + "_id($id)_arena", atom_g_umax_noret>; + def ATOM_A_UMIN_NORET : BinAtomNoRet<IL_OP_UAV_UMIN, + "_id($id)_arena", atom_g_umin_noret>; + def ATOM_A_OR_NORET : BinAtomNoRet<IL_OP_UAV_OR, + "_id($id)_arena", atom_g_or_noret>; + def ATOM_A_RSUB_NORET : BinAtomNoRet<IL_OP_UAV_RSUB, + "_id($id)_arena", atom_g_rsub_noret>; + def ATOM_A_SUB_NORET : BinAtomNoRet<IL_OP_UAV_SUB, + "_id($id)_arena", atom_g_sub_noret>; + def ATOM_A_XOR_NORET : BinAtomNoRet<IL_OP_UAV_XOR, + "_id($id)_arena", atom_g_xor_noret>; + def ATOM_A_INC_NORET : BinAtomNoRet<IL_OP_UAV_INC, + "_id($id)_arena", atom_g_inc_noret>; + def ATOM_A_DEC_NORET : BinAtomNoRet<IL_OP_UAV_DEC, + "_id($id)_arena", atom_g_dec_noret>; + def ATOM_A_CMPXCHG_NORET : CmpXChgNoRet<IL_OP_UAV_CMP, + "_id($id)_arena", atom_g_cmpxchg_noret>; + def ATOM_L_ADD_NORET : BinAtomNoRet<IL_OP_LDS_ADD, + "_resource($id)", atom_l_add_noret>; + def ATOM_L_AND_NORET : BinAtomNoRet<IL_OP_LDS_AND, + "_resource($id)", atom_l_and_noret>; + def ATOM_L_MAX_NORET : BinAtomNoRet<IL_OP_LDS_MAX, + "_resource($id)", atom_l_max_noret>; + def ATOM_L_MIN_NORET : BinAtomNoRet<IL_OP_LDS_MIN, + "_resource($id)", atom_l_min_noret>; + def ATOM_L_UMAX_NORET : BinAtomNoRet<IL_OP_LDS_UMAX, + "_resource($id)", atom_l_umax_noret>; + def ATOM_L_UMIN_NORET : BinAtomNoRet<IL_OP_LDS_UMIN, + "_resource($id)", atom_l_umin_noret>; + def ATOM_L_MSKOR_NORET : TriAtomNoRet<IL_OP_LDS_MSKOR, + "_resource($id)", atom_l_mskor_noret>; + def ATOM_L_OR_NORET : BinAtomNoRet<IL_OP_LDS_OR, + "_resource($id)", atom_l_or_noret>; + def ATOM_L_RSUB_NORET : BinAtomNoRet<IL_OP_LDS_RSUB, + "_resource($id)", atom_l_rsub_noret>; + def ATOM_L_SUB_NORET : BinAtomNoRet<IL_OP_LDS_SUB, + "_resource($id)", atom_l_sub_noret>; + def ATOM_L_XOR_NORET : BinAtomNoRet<IL_OP_LDS_XOR, + "_resource($id)", atom_l_xor_noret>; + def ATOM_L_INC_NORET : BinAtomNoRet<IL_OP_LDS_INC, + "_resource($id)", atom_l_inc_noret>; + def ATOM_L_DEC_NORET : BinAtomNoRet<IL_OP_LDS_DEC, + "_resource($id)", atom_l_dec_noret>; + def ATOM_L_CMPXCHG_NORET : TriAtomNoRet<IL_OP_LDS_CMP, + "_resource($id)", atom_l_cmpxchg_noret>; + def ATOM_R_ADD_NORET : BinAtomNoRet<IL_OP_GDS_ADD, + "_resource($id)", atom_r_add_noret>; + def ATOM_R_AND_NORET : BinAtomNoRet<IL_OP_GDS_AND, + "_resource($id)", atom_r_and_noret>; + def ATOM_R_MAX_NORET : BinAtomNoRet<IL_OP_GDS_MAX, + "_resource($id)", atom_r_max_noret>; + def ATOM_R_MIN_NORET : BinAtomNoRet<IL_OP_GDS_MIN, + "_resource($id)", atom_r_min_noret>; + def ATOM_R_UMAX_NORET : BinAtomNoRet<IL_OP_GDS_UMAX, + "_resource($id)", atom_r_umax_noret>; + def ATOM_R_UMIN_NORET : BinAtomNoRet<IL_OP_GDS_UMIN, + "_resource($id)", atom_r_umin_noret>; + def ATOM_R_MSKOR_NORET : TriAtomNoRet<IL_OP_GDS_MSKOR, + "_resource($id)", atom_r_mskor_noret>; + def ATOM_R_OR_NORET : BinAtomNoRet<IL_OP_GDS_OR, + "_resource($id)", atom_r_or_noret>; + def ATOM_R_RSUB_NORET : BinAtomNoRet<IL_OP_GDS_RSUB, + "_resource($id)", atom_r_rsub_noret>; + def ATOM_R_SUB_NORET : BinAtomNoRet<IL_OP_GDS_SUB, + "_resource($id)", atom_r_sub_noret>; + def ATOM_R_XOR_NORET : BinAtomNoRet<IL_OP_GDS_XOR, + "_resource($id)", atom_r_xor_noret>; + def ATOM_R_INC_NORET : BinAtomNoRet<IL_OP_GDS_INC, + "_resource($id)", atom_r_inc_noret>; + def ATOM_R_DEC_NORET : BinAtomNoRet<IL_OP_GDS_DEC, + "_resource($id)", atom_r_dec_noret>; + def ATOM_R_CMPXCHG_NORET : CmpXChgNoRet<IL_OP_GDS_CMP, + "_resource($id)", atom_r_cmpxchg_noret>; + def APPEND_ALLOC_NORET : AppendNoRet<IL_OP_APPEND_BUF_ALLOC, + "_id($id)", append_alloc_noret>; + def APPEND_CONSUME_NORET : AppendNoRet<IL_OP_APPEND_BUF_CONSUME, + "_id($id)", append_consume_noret>; + // All of the atomic functions that return + def ATOM_G_ADD : BinAtom<IL_OP_UAV_READ_ADD, + "_id($id)", atom_g_add>; + def ATOM_G_AND : BinAtom<IL_OP_UAV_READ_AND, + "_id($id)", atom_g_and>; + def ATOM_G_MAX : BinAtom<IL_OP_UAV_READ_MAX, + "_id($id)", atom_g_max>; + def ATOM_G_MIN : BinAtom<IL_OP_UAV_READ_MIN, + "_id($id)", atom_g_min>; + def ATOM_G_UMAX : BinAtom<IL_OP_UAV_READ_UMAX, + "_id($id)", atom_g_umax>; + def ATOM_G_UMIN : BinAtom<IL_OP_UAV_READ_UMIN, + "_id($id)", atom_g_umin>; + def ATOM_G_OR : BinAtom<IL_OP_UAV_READ_OR, + "_id($id)", atom_g_or>; + def ATOM_G_RSUB : BinAtom<IL_OP_UAV_READ_RSUB, + "_id($id)", atom_g_rsub>; + def ATOM_G_SUB : BinAtom<IL_OP_UAV_READ_SUB, + "_id($id)", atom_g_sub>; + def ATOM_G_XOR : BinAtom<IL_OP_UAV_READ_XOR, + "_id($id)", atom_g_xor>; + def ATOM_G_INC : BinAtom<IL_OP_UAV_READ_INC, + "_id($id)", atom_g_inc>; + def ATOM_G_DEC : BinAtom<IL_OP_UAV_READ_DEC, + "_id($id)", atom_g_dec>; + def ATOM_G_XCHG : BinAtom<IL_OP_UAV_READ_XCHG, + "_id($id)", atom_g_xchg>; + def ATOM_G_CMPXCHG : CmpXChg<IL_OP_UAV_READ_CMPXCHG, + "_id($id)", atom_g_cmpxchg>; + // Arena atomic accesses + def ATOM_A_ADD : BinAtom<IL_OP_UAV_READ_ADD, + "_id($id)_arena", atom_g_add>; + def ATOM_A_AND : BinAtom<IL_OP_UAV_READ_AND, + "_id($id)_arena", atom_g_and>; + def ATOM_A_MAX : BinAtom<IL_OP_UAV_READ_MAX, + "_id($id)_arena", atom_g_max>; + def ATOM_A_MIN : BinAtom<IL_OP_UAV_READ_MIN, + "_id($id)_arena", atom_g_min>; + def ATOM_A_UMAX : BinAtom<IL_OP_UAV_READ_UMAX, + "_id($id)_arena", atom_g_umax>; + def ATOM_A_UMIN : BinAtom<IL_OP_UAV_READ_UMIN, + "_id($id)_arena", atom_g_umin>; + def ATOM_A_OR : BinAtom<IL_OP_UAV_READ_OR, + "_id($id)_arena", atom_g_or>; + def ATOM_A_RSUB : BinAtom<IL_OP_UAV_READ_RSUB, + "_id($id)_arena", atom_g_rsub>; + def ATOM_A_SUB : BinAtom<IL_OP_UAV_READ_SUB, + "_id($id)_arena", atom_g_sub>; + def ATOM_A_XOR : BinAtom<IL_OP_UAV_READ_XOR, + "_id($id)_arena", atom_g_xor>; + def ATOM_A_INC : BinAtom<IL_OP_UAV_READ_INC, + "_id($id)_arena", atom_g_inc>; + def ATOM_A_DEC : BinAtom<IL_OP_UAV_READ_DEC, + "_id($id)_arena", atom_g_dec>; + def ATOM_A_XCHG : BinAtom<IL_OP_UAV_READ_XCHG, + "_id($id)_arena", atom_g_xchg>; + def ATOM_A_CMPXCHG : CmpXChg<IL_OP_UAV_READ_CMPXCHG, + "_id($id)_arena", atom_g_cmpxchg>; + def ATOM_L_ADD : BinAtom<IL_OP_LDS_READ_ADD, + "_resource($id)", atom_l_add>; + def ATOM_L_AND : BinAtom<IL_OP_LDS_READ_AND, + "_resource($id)", atom_l_and>; + def ATOM_L_MAX : BinAtom<IL_OP_LDS_READ_MAX, + "_resource($id)", atom_l_max>; + def ATOM_L_MIN : BinAtom<IL_OP_LDS_READ_MIN, + "_resource($id)", atom_l_min>; + def ATOM_L_UMAX : BinAtom<IL_OP_LDS_READ_UMAX, + "_resource($id)", atom_l_umax>; + def ATOM_L_UMIN : BinAtom<IL_OP_LDS_READ_UMIN, + "_resource($id)", atom_l_umin>; + def ATOM_L_OR : BinAtom<IL_OP_LDS_READ_OR, + "_resource($id)", atom_l_or>; + def ATOM_L_MSKOR : TriAtom<IL_OP_LDS_READ_MSKOR, + "_resource($id)", atom_l_mskor>; + def ATOM_L_RSUB : BinAtom<IL_OP_LDS_READ_RSUB, + "_resource($id)", atom_l_rsub>; + def ATOM_L_SUB : BinAtom<IL_OP_LDS_READ_SUB, + "_resource($id)", atom_l_sub>; + def ATOM_L_XOR : BinAtom<IL_OP_LDS_READ_XOR, + "_resource($id)", atom_l_xor>; + def ATOM_L_INC : BinAtom<IL_OP_LDS_READ_INC, + "_resource($id)", atom_l_inc>; + def ATOM_L_DEC : BinAtom<IL_OP_LDS_READ_DEC, + "_resource($id)", atom_l_dec>; + def ATOM_L_XCHG : BinAtom<IL_OP_LDS_READ_XCHG, + "_resource($id)", atom_l_xchg>; + def ATOM_L_CMPXCHG : TriAtom<IL_OP_LDS_READ_CMPXCHG, + "_resource($id)", atom_l_cmpxchg>; + def ATOM_R_ADD : BinAtom<IL_OP_GDS_READ_ADD, + "_resource($id)", atom_r_add>; + def ATOM_R_AND : BinAtom<IL_OP_GDS_READ_AND, + "_resource($id)", atom_r_and>; + def ATOM_R_MAX : BinAtom<IL_OP_GDS_READ_MAX, + "_resource($id)", atom_r_max>; + def ATOM_R_MIN : BinAtom<IL_OP_GDS_READ_MIN, + "_resource($id)", atom_r_min>; + def ATOM_R_UMAX : BinAtom<IL_OP_GDS_READ_UMAX, + "_resource($id)", atom_r_umax>; + def ATOM_R_UMIN : BinAtom<IL_OP_GDS_READ_UMIN, + "_resource($id)", atom_r_umin>; + def ATOM_R_OR : BinAtom<IL_OP_GDS_READ_OR, + "_resource($id)", atom_r_or>; + def ATOM_R_MSKOR : TriAtom<IL_OP_GDS_READ_MSKOR, + "_resource($id)", atom_r_mskor>; + def ATOM_R_RSUB : BinAtom<IL_OP_GDS_READ_RSUB, + "_resource($id)", atom_r_rsub>; + def ATOM_R_SUB : BinAtom<IL_OP_GDS_READ_SUB, + "_resource($id)", atom_r_sub>; + def ATOM_R_XOR : BinAtom<IL_OP_GDS_READ_XOR, + "_resource($id)", atom_r_xor>; + def ATOM_R_INC : BinAtom<IL_OP_GDS_READ_INC, + "_resource($id)", atom_r_inc>; + def ATOM_R_DEC : BinAtom<IL_OP_GDS_READ_DEC, + "_resource($id)", atom_r_dec>; + def ATOM_R_XCHG : BinAtom<IL_OP_GDS_READ_XCHG, + "_resource($id)", atom_r_xchg>; + def ATOM_R_CMPXCHG : CmpXChg<IL_OP_GDS_READ_CMPXCHG, + "_resource($id)", atom_r_cmpxchg>; + def APPEND_ALLOC : Append<IL_OP_APPEND_BUF_ALLOC, + "_id($id)", append_alloc>; + def APPEND_CONSUME : Append<IL_OP_APPEND_BUF_CONSUME, + "_id($id)", append_consume>; +} +} +let Predicates = [Has64BitPtr] in { +// All of the image functions +def IMAGE1D64_READ : ILFormat<IL_OP_SAMPLE, (outs GPRV4I32:$dst), + (ins MEMI64:$ptr, GPRI32:$sampler, GPRV4F32:$addy), + !strconcat(IL_OP_SAMPLE.Text, + "_resource($ptr)_sampler($sampler)_coordtype(normalized) $dst, $addy"), + [(set GPRV4I32:$dst, + (int_AMDIL_image1d_read_norm ADDR64:$ptr, GPRI32:$sampler, GPRV4F32:$addy))]>; +def IMAGE1DA64_READ : ILFormat<IL_OP_SAMPLE, (outs GPRV4I32:$dst), + (ins MEMI64:$ptr, GPRI32:$sampler, GPRV4F32:$addy), + !strconcat(IL_OP_SAMPLE.Text, + "_resource($ptr)_sampler($sampler)_coordtype(normalized) $dst, $addy"), + [(set GPRV4I32:$dst, + (int_AMDIL_image1d_array_read_norm ADDR64:$ptr, GPRI32:$sampler, GPRV4F32:$addy))]>; +def IMAGE2D64_READ : ILFormat<IL_OP_SAMPLE, (outs GPRV4I32:$dst), + (ins MEMI64:$ptr, GPRI32:$sampler, GPRV4F32:$addy), + !strconcat(IL_OP_SAMPLE.Text, + "_resource($ptr)_sampler($sampler)_coordtype(normalized) $dst, $addy"), + [(set GPRV4I32:$dst, + (int_AMDIL_image2d_read_norm ADDR64:$ptr, GPRI32:$sampler, GPRV4F32:$addy))]>; +def IMAGE2DA64_READ : ILFormat<IL_OP_SAMPLE, (outs GPRV4I32:$dst), + (ins MEMI64:$ptr, GPRI32:$sampler, GPRV4F32:$addy), + !strconcat(IL_OP_SAMPLE.Text, + "_resource($ptr)_sampler($sampler)_coordtype(normalized) $dst, $addy"), + [(set GPRV4I32:$dst, + (int_AMDIL_image2d_array_read_norm ADDR64:$ptr, GPRI32:$sampler, GPRV4F32:$addy))]>; +def IMAGE3D64_READ : ILFormat<IL_OP_SAMPLE, (outs GPRV4I32:$dst), + (ins MEMI64:$ptr, GPRI32:$sampler, GPRV4F32:$addy), + !strconcat(IL_OP_SAMPLE.Text, + "_resource($ptr)_sampler($sampler)_coordtype(normalized) $dst, $addy"), + [(set GPRV4I32:$dst, + (int_AMDIL_image3d_read_norm ADDR64:$ptr, GPRI32:$sampler, GPRV4F32:$addy))]>; +def IMAGE1D64_READ_UNNORM : ILFormat<IL_OP_SAMPLE, (outs GPRV4I32:$dst), + (ins MEMI64:$ptr, GPRI32:$sampler, GPRV4F32:$addy), + !strconcat(IL_OP_SAMPLE.Text, + "_resource($ptr)_sampler($sampler)_coordtype(unnormalized) $dst, $addy"), + [(set GPRV4I32:$dst, + (int_AMDIL_image1d_read_unnorm ADDR64:$ptr, GPRI32:$sampler, GPRV4F32:$addy))]>; +def IMAGE1DA64_READ_UNNORM : ILFormat<IL_OP_SAMPLE, (outs GPRV4I32:$dst), + (ins MEMI64:$ptr, GPRI32:$sampler, GPRV4F32:$addy), + !strconcat(IL_OP_SAMPLE.Text, + "_resource($ptr)_sampler($sampler)_coordtype(unnormalized) $dst, $addy"), + [(set GPRV4I32:$dst, + (int_AMDIL_image1d_array_read_unnorm ADDR64:$ptr, GPRI32:$sampler, GPRV4F32:$addy))]>; +def IMAGE2D64_READ_UNNORM : ILFormat<IL_OP_SAMPLE, (outs GPRV4I32:$dst), + (ins MEMI64:$ptr, GPRI32:$sampler, GPRV4F32:$addy), + !strconcat(IL_OP_SAMPLE.Text, + "_resource($ptr)_sampler($sampler)_coordtype(unnormalized) $dst, $addy"), + [(set GPRV4I32:$dst, + (int_AMDIL_image2d_read_unnorm ADDR64:$ptr, GPRI32:$sampler, GPRV4F32:$addy))]>; +def IMAGE2DA64_READ_UNNORM : ILFormat<IL_OP_SAMPLE, (outs GPRV4I32:$dst), + (ins MEMI64:$ptr, GPRI32:$sampler, GPRV4F32:$addy), + !strconcat(IL_OP_SAMPLE.Text, + "_resource($ptr)_sampler($sampler)_coordtype(unnormalized) $dst, $addy"), + [(set GPRV4I32:$dst, + (int_AMDIL_image2d_array_read_unnorm ADDR64:$ptr, GPRI32:$sampler, GPRV4F32:$addy))]>; +def IMAGE3D64_READ_UNNORM : ILFormat<IL_OP_SAMPLE, (outs GPRV4I32:$dst), + (ins MEMI64:$ptr, GPRI32:$sampler, GPRV4F32:$addy), + !strconcat(IL_OP_SAMPLE.Text, + "_resource($ptr)_sampler($sampler)_coordtype(unnormalized) $dst, $addy"), + [(set GPRV4I32:$dst, + (int_AMDIL_image3d_read_unnorm ADDR64:$ptr, GPRI32:$sampler, GPRV4F32:$addy))]>; +def IMAGE1D64_INFO0 : ILFormat<IL_OP_MOV, (outs GPRV4I32:$dst), + (ins MEMI64:$ptr), + !strconcat(IL_OP_MOV.Text, " $dst, $ptr"), + [(set GPRV4I32:$dst, (int_AMDIL_image1d_info0 ADDR64:$ptr))]>; +def IMAGE1D64_INFO1 : ILFormat<IL_OP_MOV, (outs GPRV4I32:$dst), + (ins MEMI64:$ptr), + !strconcat(IL_OP_MOV.Text, " $dst, $ptr"), + [(set GPRV4I32:$dst, (int_AMDIL_image1d_info1 ADDR64:$ptr))]>; +def IMAGE1DA64_INFO0 : ILFormat<IL_OP_MOV, (outs GPRV4I32:$dst), + (ins MEMI64:$ptr), + !strconcat(IL_OP_MOV.Text, " $dst, $ptr"), + [(set GPRV4I32:$dst, (int_AMDIL_image1d_array_info0 ADDR64:$ptr))]>; +def IMAGE1DA64_INFO1 : ILFormat<IL_OP_MOV, (outs GPRV4I32:$dst), + (ins MEMI64:$ptr), + !strconcat(IL_OP_MOV.Text, " $dst, $ptr"), + [(set GPRV4I32:$dst, (int_AMDIL_image1d_array_info1 ADDR64:$ptr))]>; +def IMAGE2DA64_INFO0 : ILFormat<IL_OP_MOV, (outs GPRV4I32:$dst), + (ins MEMI64:$ptr), + !strconcat(IL_OP_MOV.Text, " $dst, $ptr"), + [(set GPRV4I32:$dst, (int_AMDIL_image2d_array_info0 ADDR64:$ptr))]>; +def IMAGE2DA64_INFO1 : ILFormat<IL_OP_MOV, (outs GPRV4I32:$dst), + (ins MEMI64:$ptr), + !strconcat(IL_OP_MOV.Text, " $dst, $ptr"), + [(set GPRV4I32:$dst, (int_AMDIL_image2d_array_info1 ADDR64:$ptr))]>; +def IMAGE2D64_INFO0 : ILFormat<IL_OP_MOV, (outs GPRV4I32:$dst), + (ins MEMI64:$ptr), + !strconcat(IL_OP_MOV.Text, " $dst, $ptr"), + [(set GPRV4I32:$dst, (int_AMDIL_image2d_info0 ADDR64:$ptr))]>; +def IMAGE2D64_INFO1 : ILFormat<IL_OP_MOV, (outs GPRV4I32:$dst), + (ins MEMI64:$ptr), + !strconcat(IL_OP_MOV.Text, " $dst, $ptr"), + [(set GPRV4I32:$dst, (int_AMDIL_image2d_info1 ADDR64:$ptr))]>; +def IMAGE3D64_INFO0 : ILFormat<IL_OP_MOV, (outs GPRV4I32:$dst), + (ins MEMI64:$ptr), + !strconcat(IL_OP_MOV.Text, " $dst, $ptr"), + [(set GPRV4I32:$dst, (int_AMDIL_image3d_info0 ADDR64:$ptr))]>; +def IMAGE3D64_INFO1 : ILFormat<IL_OP_MOV, (outs GPRV4I32:$dst), + (ins MEMI64:$ptr), + !strconcat(IL_OP_MOV.Text, " $dst, $ptr"), + [(set GPRV4I32:$dst, (int_AMDIL_image3d_info1 ADDR64:$ptr))]>; +def IMAGE1D64_WRITE : ILFormat<IL_OP_UAV_STORE, (outs), + (ins MEMI64:$ptr, GPRV2I32:$addy, GPRV4I32:$data), + !strconcat(IL_OP_UAV_STORE.Text, + "_id($ptr) $addy, $data"), + [(int_AMDIL_image1d_write ADDR64:$ptr, GPRV2I32:$addy, GPRV4I32:$data)]>; +def IMAGE1DA64_WRITE : ILFormat<IL_OP_UAV_STORE, (outs), + (ins MEMI64:$ptr, GPRV2I32:$addy, GPRV4I32:$data), + !strconcat(IL_OP_UAV_STORE.Text, + "_id($ptr) $addy, $data"), + [(int_AMDIL_image1d_array_write ADDR64:$ptr, GPRV2I32:$addy, GPRV4I32:$data)]>; +def IMAGE2D64_WRITE : ILFormat<IL_OP_UAV_STORE, (outs), + (ins MEMI64:$ptr, GPRV2I32:$addy, GPRV4I32:$data), + !strconcat(IL_OP_UAV_STORE.Text, + "_id($ptr) $addy, $data"), + [(int_AMDIL_image2d_write ADDR64:$ptr, GPRV2I32:$addy, GPRV4I32:$data)]>; +def IMAGE2DA64_WRITE : ILFormat<IL_OP_UAV_STORE, (outs), + (ins MEMI64:$ptr, GPRV2I32:$addy, GPRV4I32:$data), + !strconcat(IL_OP_UAV_STORE.Text, + "_id($ptr) $addy, $data"), + [(int_AMDIL_image2d_array_write ADDR64:$ptr, GPRV2I32:$addy, GPRV4I32:$data)]>; +def IMAGE3D64_WRITE : ILFormat<IL_OP_UAV_STORE, (outs), + (ins MEMI64:$ptr, GPRV4I32:$addy, GPRV4I32:$data), + !strconcat(IL_OP_UAV_STORE.Text, + "_id($ptr) $addy, $data"), + [(int_AMDIL_image3d_write ADDR64:$ptr, GPRV4I32:$addy, GPRV4I32:$data)]>; +let hasSideEffects= 1 in { + // All of the noret atomic functions + def ATOM_G64_ADD_NORET : BinAtomNoRet64<IL_OP_UAV_ADD, + "_id($id)", atom_g_add_noret>; + def ATOM_G64_AND_NORET : BinAtomNoRet64<IL_OP_UAV_AND, + "_id($id)", atom_g_and_noret>; + def ATOM_G64_MAX_NORET : BinAtomNoRet64<IL_OP_UAV_MAX, + "_id($id)", atom_g_max_noret>; + def ATOM_G64_MIN_NORET : BinAtomNoRet64<IL_OP_UAV_MIN, + "_id($id)", atom_g_min_noret>; + def ATOM_G64_UMAX_NORET : BinAtomNoRet64<IL_OP_UAV_UMAX, + "_id($id)", atom_g_umax_noret>; + def ATOM_G64_UMIN_NORET : BinAtomNoRet64<IL_OP_UAV_UMIN, + "_id($id)", atom_g_umin_noret>; + def ATOM_G64_OR_NORET : BinAtomNoRet64<IL_OP_UAV_OR, + "_id($id)", atom_g_or_noret>; + def ATOM_G64_RSUB_NORET : BinAtomNoRet64<IL_OP_UAV_RSUB, + "_id($id)", atom_g_rsub_noret>; + def ATOM_G64_SUB_NORET : BinAtomNoRet64<IL_OP_UAV_SUB, + "_id($id)", atom_g_sub_noret>; + def ATOM_G64_XOR_NORET : BinAtomNoRet64<IL_OP_UAV_XOR, + "_id($id)", atom_g_xor_noret>; + def ATOM_G64_INC_NORET : BinAtomNoRet64<IL_OP_UAV_INC, + "_id($id)", atom_g_inc_noret>; + def ATOM_G64_DEC_NORET : BinAtomNoRet64<IL_OP_UAV_DEC, + "_id($id)", atom_g_dec_noret>; + def ATOM_G64_CMPXCHG_NORET : CmpXChgNoRet64<IL_OP_UAV_CMP, + "_id($id)", atom_g_cmpxchg_noret>; + def ATOM_A64_ADD_NORET : BinAtomNoRet64<IL_OP_UAV_ADD, + "_id($id)_arena", atom_g_add_noret>; + def ATOM_A64_AND_NORET : BinAtomNoRet64<IL_OP_UAV_AND, + "_id($id)_arena", atom_g_and_noret>; + def ATOM_A64_MAX_NORET : BinAtomNoRet64<IL_OP_UAV_MAX, + "_id($id)_arena", atom_g_max_noret>; + def ATOM_A64_MIN_NORET : BinAtomNoRet64<IL_OP_UAV_MIN, + "_id($id)_arena", atom_g_min_noret>; + def ATOM_A64_UMAX_NORET : BinAtomNoRet64<IL_OP_UAV_UMAX, + "_id($id)_arena", atom_g_umax_noret>; + def ATOM_A64_UMIN_NORET : BinAtomNoRet64<IL_OP_UAV_UMIN, + "_id($id)_arena", atom_g_umin_noret>; + def ATOM_A64_OR_NORET : BinAtomNoRet64<IL_OP_UAV_OR, + "_id($id)_arena", atom_g_or_noret>; + def ATOM_A64_RSUB_NORET : BinAtomNoRet64<IL_OP_UAV_RSUB, + "_id($id)_arena", atom_g_rsub_noret>; + def ATOM_A64_SUB_NORET : BinAtomNoRet64<IL_OP_UAV_SUB, + "_id($id)_arena", atom_g_sub_noret>; + def ATOM_A64_XOR_NORET : BinAtomNoRet64<IL_OP_UAV_XOR, + "_id($id)_arena", atom_g_xor_noret>; + def ATOM_A64_INC_NORET : BinAtomNoRet64<IL_OP_UAV_INC, + "_id($id)_arena", atom_g_inc_noret>; + def ATOM_A64_DEC_NORET : BinAtomNoRet64<IL_OP_UAV_DEC, + "_id($id)_arena", atom_g_dec_noret>; + def ATOM_A64_CMPXCHG_NORET : CmpXChgNoRet64<IL_OP_UAV_CMP, + "_id($id)_arena", atom_g_cmpxchg_noret>; + def ATOM_L64_ADD_NORET : BinAtomNoRet64<IL_OP_LDS_ADD, + "_resource($id)", atom_l_add_noret>; + def ATOM_L64_AND_NORET : BinAtomNoRet64<IL_OP_LDS_AND, + "_resource($id)", atom_l_and_noret>; + def ATOM_L64_MAX_NORET : BinAtomNoRet64<IL_OP_LDS_MAX, + "_resource($id)", atom_l_max_noret>; + def ATOM_L64_MIN_NORET : BinAtomNoRet64<IL_OP_LDS_MIN, + "_resource($id)", atom_l_min_noret>; + def ATOM_L64_UMAX_NORET : BinAtomNoRet64<IL_OP_LDS_UMAX, + "_resource($id)", atom_l_umax_noret>; + def ATOM_L64_UMIN_NORET : BinAtomNoRet64<IL_OP_LDS_UMIN, + "_resource($id)", atom_l_umin_noret>; + def ATOM_L64_MSKOR_NORET : TriAtomNoRet64<IL_OP_LDS_MSKOR, + "_resource($id)", atom_l_mskor_noret>; + def ATOM_L64_OR_NORET : BinAtomNoRet64<IL_OP_LDS_OR, + "_resource($id)", atom_l_or_noret>; + def ATOM_L64_RSUB_NORET : BinAtomNoRet64<IL_OP_LDS_RSUB, + "_resource($id)", atom_l_rsub_noret>; + def ATOM_L64_SUB_NORET : BinAtomNoRet64<IL_OP_LDS_SUB, + "_resource($id)", atom_l_sub_noret>; + def ATOM_L64_XOR_NORET : BinAtomNoRet64<IL_OP_LDS_XOR, + "_resource($id)", atom_l_xor_noret>; + def ATOM_L64_INC_NORET : BinAtomNoRet64<IL_OP_LDS_INC, + "_resource($id)", atom_l_inc_noret>; + def ATOM_L64_DEC_NORET : BinAtomNoRet64<IL_OP_LDS_DEC, + "_resource($id)", atom_l_dec_noret>; + def ATOM_L64_CMPXCHG_NORET : TriAtomNoRet64<IL_OP_LDS_CMP, + "_resource($id)", atom_l_cmpxchg_noret>; + def ATOM_R64_ADD_NORET : BinAtomNoRet64<IL_OP_GDS_ADD, + "_resource($id)", atom_r_add_noret>; + def ATOM_R64_AND_NORET : BinAtomNoRet64<IL_OP_GDS_AND, + "_resource($id)", atom_r_and_noret>; + def ATOM_R64_MAX_NORET : BinAtomNoRet64<IL_OP_GDS_MAX, + "_resource($id)", atom_r_max_noret>; + def ATOM_R64_MIN_NORET : BinAtomNoRet64<IL_OP_GDS_MIN, + "_resource($id)", atom_r_min_noret>; + def ATOM_R64_UMAX_NORET : BinAtomNoRet64<IL_OP_GDS_UMAX, + "_resource($id)", atom_r_umax_noret>; + def ATOM_R64_UMIN_NORET : BinAtomNoRet64<IL_OP_GDS_UMIN, + "_resource($id)", atom_r_umin_noret>; + def ATOM_R64_MSKOR_NORET : TriAtomNoRet64<IL_OP_GDS_MSKOR, + "_resource($id)", atom_r_mskor_noret>; + def ATOM_R64_OR_NORET : BinAtomNoRet64<IL_OP_GDS_OR, + "_resource($id)", atom_r_or_noret>; + def ATOM_R64_RSUB_NORET : BinAtomNoRet64<IL_OP_GDS_RSUB, + "_resource($id)", atom_r_rsub_noret>; + def ATOM_R64_SUB_NORET : BinAtomNoRet64<IL_OP_GDS_SUB, + "_resource($id)", atom_r_sub_noret>; + def ATOM_R64_XOR_NORET : BinAtomNoRet64<IL_OP_GDS_XOR, + "_resource($id)", atom_r_xor_noret>; + def ATOM_R64_INC_NORET : BinAtomNoRet64<IL_OP_GDS_INC, + "_resource($id)", atom_r_inc_noret>; + def ATOM_R64_DEC_NORET : BinAtomNoRet64<IL_OP_GDS_DEC, + "_resource($id)", atom_r_dec_noret>; + def ATOM_R64_CMPXCHG_NORET : CmpXChgNoRet64<IL_OP_GDS_CMP, + "_resource($id)", atom_r_cmpxchg_noret>; + def APPEND_ALLOC64_NORET : AppendNoRet64<IL_OP_APPEND_BUF_ALLOC, + "_id($id)", append_alloc_noret>; + def APPEND_CONSUME64_NORET : AppendNoRet64<IL_OP_APPEND_BUF_CONSUME, + "_id($id)", append_consume_noret>; + // All of the atomic functions that return + def ATOM_G64_ADD : BinAtom64<IL_OP_UAV_READ_ADD, + "_id($id)", atom_g_add>; + def ATOM_G64_AND : BinAtom64<IL_OP_UAV_READ_AND, + "_id($id)", atom_g_and>; + def ATOM_G64_MAX : BinAtom64<IL_OP_UAV_READ_MAX, + "_id($id)", atom_g_max>; + def ATOM_G64_MIN : BinAtom64<IL_OP_UAV_READ_MIN, + "_id($id)", atom_g_min>; + def ATOM_G64_UMAX : BinAtom64<IL_OP_UAV_READ_UMAX, + "_id($id)", atom_g_umax>; + def ATOM_G64_UMIN : BinAtom64<IL_OP_UAV_READ_UMIN, + "_id($id)", atom_g_umin>; + def ATOM_G64_OR : BinAtom64<IL_OP_UAV_READ_OR, + "_id($id)", atom_g_or>; + def ATOM_G64_RSUB : BinAtom64<IL_OP_UAV_READ_RSUB, + "_id($id)", atom_g_rsub>; + def ATOM_G64_SUB : BinAtom64<IL_OP_UAV_READ_SUB, + "_id($id)", atom_g_sub>; + def ATOM_G64_XOR : BinAtom64<IL_OP_UAV_READ_XOR, + "_id($id)", atom_g_xor>; + def ATOM_G64_INC : BinAtom64<IL_OP_UAV_READ_INC, + "_id($id)", atom_g_inc>; + def ATOM_G64_DEC : BinAtom64<IL_OP_UAV_READ_DEC, + "_id($id)", atom_g_dec>; + def ATOM_G64_XCHG : BinAtom64<IL_OP_UAV_READ_XCHG, + "_id($id)", atom_g_xchg>; + def ATOM_G64_CMPXCHG : CmpXChg64<IL_OP_UAV_READ_CMPXCHG, + "_id($id)", atom_g_cmpxchg>; + // Arena atomic accesses + def ATOM_A64_ADD : BinAtom64<IL_OP_UAV_READ_ADD, + "_id($id)_arena", atom_g_add>; + def ATOM_A64_AND : BinAtom64<IL_OP_UAV_READ_AND, + "_id($id)_arena", atom_g_and>; + def ATOM_A64_MAX : BinAtom64<IL_OP_UAV_READ_MAX, + "_id($id)_arena", atom_g_max>; + def ATOM_A64_MIN : BinAtom64<IL_OP_UAV_READ_MIN, + "_id($id)_arena", atom_g_min>; + def ATOM_A64_UMAX : BinAtom64<IL_OP_UAV_READ_UMAX, + "_id($id)_arena", atom_g_umax>; + def ATOM_A64_UMIN : BinAtom64<IL_OP_UAV_READ_UMIN, + "_id($id)_arena", atom_g_umin>; + def ATOM_A64_OR : BinAtom64<IL_OP_UAV_READ_OR, + "_id($id)_arena", atom_g_or>; + def ATOM_A64_RSUB : BinAtom64<IL_OP_UAV_READ_RSUB, + "_id($id)_arena", atom_g_rsub>; + def ATOM_A64_SUB : BinAtom64<IL_OP_UAV_READ_SUB, + "_id($id)_arena", atom_g_sub>; + def ATOM_A64_XOR : BinAtom64<IL_OP_UAV_READ_XOR, + "_id($id)_arena", atom_g_xor>; + def ATOM_A64_INC : BinAtom64<IL_OP_UAV_READ_INC, + "_id($id)_arena", atom_g_inc>; + def ATOM_A64_DEC : BinAtom64<IL_OP_UAV_READ_DEC, + "_id($id)_arena", atom_g_dec>; + def ATOM_A64_XCHG : BinAtom64<IL_OP_UAV_READ_XCHG, + "_id($id)_arena", atom_g_xchg>; + def ATOM_A64_CMPXCHG : CmpXChg64<IL_OP_UAV_READ_CMPXCHG, + "_id($id)_arena", atom_g_cmpxchg>; + def ATOM_L64_ADD : BinAtom64<IL_OP_LDS_READ_ADD, + "_resource($id)", atom_l_add>; + def ATOM_L64_AND : BinAtom64<IL_OP_LDS_READ_AND, + "_resource($id)", atom_l_and>; + def ATOM_L64_MAX : BinAtom64<IL_OP_LDS_READ_MAX, + "_resource($id)", atom_l_max>; + def ATOM_L64_MIN : BinAtom64<IL_OP_LDS_READ_MIN, + "_resource($id)", atom_l_min>; + def ATOM_L64_UMAX : BinAtom64<IL_OP_LDS_READ_UMAX, + "_resource($id)", atom_l_umax>; + def ATOM_L64_UMIN : BinAtom64<IL_OP_LDS_READ_UMIN, + "_resource($id)", atom_l_umin>; + def ATOM_L64_OR : BinAtom64<IL_OP_LDS_READ_OR, + "_resource($id)", atom_l_or>; + def ATOM_L64_MSKOR : TriAtom64<IL_OP_LDS_READ_MSKOR, + "_resource($id)", atom_l_mskor>; + def ATOM_L64_RSUB : BinAtom64<IL_OP_LDS_READ_RSUB, + "_resource($id)", atom_l_rsub>; + def ATOM_L64_SUB : BinAtom64<IL_OP_LDS_READ_SUB, + "_resource($id)", atom_l_sub>; + def ATOM_L64_XOR : BinAtom64<IL_OP_LDS_READ_XOR, + "_resource($id)", atom_l_xor>; + def ATOM_L64_INC : BinAtom64<IL_OP_LDS_READ_INC, + "_resource($id)", atom_l_inc>; + def ATOM_L64_DEC : BinAtom64<IL_OP_LDS_READ_DEC, + "_resource($id)", atom_l_dec>; + def ATOM_L64_XCHG : BinAtom64<IL_OP_LDS_READ_XCHG, + "_resource($id)", atom_l_xchg>; + def ATOM_L64_CMPXCHG : TriAtom64<IL_OP_LDS_READ_CMPXCHG, + "_resource($id)", atom_l_cmpxchg>; + def ATOM_R64_ADD : BinAtom64<IL_OP_GDS_READ_ADD, + "_resource($id)", atom_r_add>; + def ATOM_R64_AND : BinAtom64<IL_OP_GDS_READ_AND, + "_resource($id)", atom_r_and>; + def ATOM_R64_MAX : BinAtom64<IL_OP_GDS_READ_MAX, + "_resource($id)", atom_r_max>; + def ATOM_R64_MIN : BinAtom64<IL_OP_GDS_READ_MIN, + "_resource($id)", atom_r_min>; + def ATOM_R64_UMAX : BinAtom64<IL_OP_GDS_READ_UMAX, + "_resource($id)", atom_r_umax>; + def ATOM_R64_UMIN : BinAtom64<IL_OP_GDS_READ_UMIN, + "_resource($id)", atom_r_umin>; + def ATOM_R64_OR : BinAtom64<IL_OP_GDS_READ_OR, + "_resource($id)", atom_r_or>; + def ATOM_R64_MSKOR : TriAtom64<IL_OP_GDS_READ_MSKOR, + "_resource($id)", atom_r_mskor>; + def ATOM_R64_RSUB : BinAtom64<IL_OP_GDS_READ_RSUB, + "_resource($id)", atom_r_rsub>; + def ATOM_R64_SUB : BinAtom64<IL_OP_GDS_READ_SUB, + "_resource($id)", atom_r_sub>; + def ATOM_R64_XOR : BinAtom64<IL_OP_GDS_READ_XOR, + "_resource($id)", atom_r_xor>; + def ATOM_R64_INC : BinAtom64<IL_OP_GDS_READ_INC, + "_resource($id)", atom_r_inc>; + def ATOM_R64_DEC : BinAtom64<IL_OP_GDS_READ_DEC, + "_resource($id)", atom_r_dec>; + def ATOM_R64_XCHG : BinAtom64<IL_OP_GDS_READ_XCHG, + "_resource($id)", atom_r_xchg>; + def ATOM_R64_CMPXCHG : CmpXChg64<IL_OP_GDS_READ_CMPXCHG, + "_resource($id)", atom_r_cmpxchg>; + def APPEND_ALLOC64 : Append64<IL_OP_APPEND_BUF_ALLOC, + "_id($id)", append_alloc>; + def APPEND_CONSUME64 : Append64<IL_OP_APPEND_BUF_CONSUME, + "_id($id)", append_consume>; +} +} +/* +def SEMAPHORE_INIT : BinaryOpNoRet<IL_OP_SEMAPHORE_INIT, (outs), + (ins MEMI32:$ptr, i32imm:$val), + !strconcat(IL_OP_SEMAPHORE_INIT.Text, "_id($ptr)_value($val)"), + [(int_AMDIL_semaphore_init ADDR:$ptr, timm:$val)]>; + +def SEMAPHORE_WAIT : UnaryOpNoRet<IL_OP_SEMAPHORE_WAIT, (outs), + (ins MEMI32:$ptr), + !strconcat(IL_OP_SEMAPHORE_WAIT.Text, "_id($ptr)"), + [(int_AMDIL_semaphore_wait ADDR:$ptr)]>; + +def SEMAPHORE_SIGNAL : UnaryOpNoRet<IL_OP_SEMAPHORE_SIGNAL, (outs), + (ins MEMI32:$ptr), + !strconcat(IL_OP_SEMAPHORE_SIGNAL.Text, "_id($ptr)"), + [(int_AMDIL_semaphore_signal ADDR:$ptr)]>; +*/ |