Cross Reference: /gem5/configs/common/cores/arm/HPI.py

Deleted Added

sdiff udiff text old ( 13432:6ce67b7e6e44 ) new ( 13731:67cd980cb20f )

full compact

HPI.py (13432:6ce67b7e6e44)	HPI.py (13731:67cd980cb20f)
1# Copyright (c) 2014-2017 ARM Limited 2# All rights reserved. 3# 4# The license below extends only to copyright in the software and shall 5# not be construed as granting a license to any other intellectual 6# property including but not limited to intellectual property relating 7# to a hardware implementation of the functionality of the software 8# licensed hereunder. You may use the software subject to the license 9# terms below provided that you ensure that this notice is replicated 10# unmodified and in its entirety in all distributions of the software, 11# modified or unmodified, in source code or in binary form. 12# 13# Redistribution and use in source and binary forms, with or without 14# modification, are permitted provided that the following conditions are 15# met: redistributions of source code must retain the above copyright 16# notice, this list of conditions and the following disclaimer; 17# redistributions in binary form must reproduce the above copyright 18# notice, this list of conditions and the following disclaimer in the 19# documentation and/or other materials provided with the distribution; 20# neither the name of the copyright holders nor the names of its 21# contributors may be used to endorse or promote products derived from 22# this software without specific prior written permission. 23# 24# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 25# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 26# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR 27# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT 28# OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 29# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT 30# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 31# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 32# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 33# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE 34# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 35# 36# Authors: Andrew Bardsley 37# 38 39"""The High-Performance In-order (HPI) CPU timing model is tuned to be 40representative of a modern in-order ARMv8-A implementation. The HPI 41core and its supporting simulation scripts, namely starter_se.py and 42starter_fs.py (under /configs/example/arm/) are part of the ARM 43Research Starter Kit on System Modeling. More information can be found 44at: http://www.arm.com/ResearchEnablement/SystemModeling 45 46""" 47 48from __future__ import print_function 49 50from m5.objects import * 51 52# Simple function to allow a string of [01x_] to be converted into a 53# mask and value for use with MinorFUTiming 54def make_implicant(implicant_string): 55 ret_mask = 0 56 ret_match = 0 57 58 shift = False 59 for char in implicant_string: 60 char = char.lower() 61 if shift: 62 ret_mask <<= 1 63 ret_match <<= 1 64 65 shift = True 66 if char == '_': 67 shift = False 68 elif char == '0': 69 ret_mask \|= 1 70 elif char == '1': 71 ret_mask \|= 1 72 ret_match \|= 1 73 elif char == 'x': 74 pass 75 else: 76 print("Can't parse implicant character", char) 77 78 return (ret_mask, ret_match) 79 80# ,----- 36 thumb 81# \| ,--- 35 bigThumb 82# \| \|,-- 34 aarch64 83a64_inst = make_implicant('0_01xx__xxxx_xxxx_xxxx_xxxx__xxxx_xxxx_xxxx_xxxx') 84a32_inst = make_implicant('0_00xx__xxxx_xxxx_xxxx_xxxx__xxxx_xxxx_xxxx_xxxx') 85t32_inst = make_implicant('1_10xx__xxxx_xxxx_xxxx_xxxx__xxxx_xxxx_xxxx_xxxx') 86t16_inst = make_implicant('1_00xx__xxxx_xxxx_xxxx_xxxx__xxxx_xxxx_xxxx_xxxx') 87any_inst = make_implicant('x_xxxx__xxxx_xxxx_xxxx_xxxx__xxxx_xxxx_xxxx_xxxx') 88# \| \|\| 89any_a64_inst = \ 90 make_implicant('x_x1xx__xxxx_xxxx_xxxx_xxxx__xxxx_xxxx_xxxx_xxxx') 91any_non_a64_inst = \ 92 make_implicant('x_x0xx__xxxx_xxxx_xxxx_xxxx__xxxx_xxxx_xxxx_xxxx') 93 94def encode_opcode(pattern): 95 def encode(opcode_string): 96 a64_mask, a64_match = pattern 97 mask, match = make_implicant(opcode_string) 98 return (a64_mask \| mask), (a64_match \| match) 99 return encode 100 101a64_opcode = encode_opcode(a64_inst) 102a32_opcode = encode_opcode(a32_inst) 103t32_opcode = encode_opcode(t32_inst) 104t16_opcode = encode_opcode(t16_inst) 105 106# These definitions (in some form) should probably be part of TimingExpr 107 108def literal(value): 109 def body(env): 110 ret = TimingExprLiteral() 111 ret.value = value 112 return ret 113 return body 114 115def bin(op, left, right): 116 def body(env): 117 ret = TimingExprBin() 118 ret.op = 'timingExpr' + op 119 ret.left = left(env) 120 ret.right = right(env) 121 return ret 122 return body 123 124def un(op, arg): 125 def body(env): 126 ret = TimingExprUn() 127 ret.op = 'timingExpr' + op 128 ret.arg = arg(env) 129 return ret 130 return body 131 132def ref(name): 133 def body(env): 134 if name in env: 135 ret = TimingExprRef() 136 ret.index = env[name] 137 else: 138 print("Invalid expression name", name) 139 ret = TimingExprNull() 140 return ret 141 return body 142 143def if_expr(cond, true_expr, false_expr): 144 def body(env): 145 ret = TimingExprIf() 146 ret.cond = cond(env) 147 ret.trueExpr = true_expr(env) 148 ret.falseExpr = false_expr(env) 149 return ret 150 return body 151 152def src(index): 153 def body(env): 154 ret = TimingExprSrcReg() 155 ret.index = index 156 return ret 157 return body 158 159def int_reg(reg): 160 def body(env): 161 ret = TimingExprReadIntReg() 162 ret.reg = reg(env) 163 return ret 164 return body 165 166def let(bindings, expr): 167 def body(env): 168 ret = TimingExprLet() 169 let_bindings = [] 170 new_env = {} 171 i = 0 172 173 # Make the sub-expression as null to start with 174 for name, binding in bindings: 175 new_env[name] = i 176 i += 1 177 178 defns = [] 179 # Then apply them to the produced new env	1# Copyright (c) 2014-2017 ARM Limited 2# All rights reserved. 3# 4# The license below extends only to copyright in the software and shall 5# not be construed as granting a license to any other intellectual 6# property including but not limited to intellectual property relating 7# to a hardware implementation of the functionality of the software 8# licensed hereunder. You may use the software subject to the license 9# terms below provided that you ensure that this notice is replicated 10# unmodified and in its entirety in all distributions of the software, 11# modified or unmodified, in source code or in binary form. 12# 13# Redistribution and use in source and binary forms, with or without 14# modification, are permitted provided that the following conditions are 15# met: redistributions of source code must retain the above copyright 16# notice, this list of conditions and the following disclaimer; 17# redistributions in binary form must reproduce the above copyright 18# notice, this list of conditions and the following disclaimer in the 19# documentation and/or other materials provided with the distribution; 20# neither the name of the copyright holders nor the names of its 21# contributors may be used to endorse or promote products derived from 22# this software without specific prior written permission. 23# 24# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 25# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 26# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR 27# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT 28# OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 29# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT 30# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 31# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 32# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 33# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE 34# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 35# 36# Authors: Andrew Bardsley 37# 38 39"""The High-Performance In-order (HPI) CPU timing model is tuned to be 40representative of a modern in-order ARMv8-A implementation. The HPI 41core and its supporting simulation scripts, namely starter_se.py and 42starter_fs.py (under /configs/example/arm/) are part of the ARM 43Research Starter Kit on System Modeling. More information can be found 44at: http://www.arm.com/ResearchEnablement/SystemModeling 45 46""" 47 48from __future__ import print_function 49 50from m5.objects import * 51 52# Simple function to allow a string of [01x_] to be converted into a 53# mask and value for use with MinorFUTiming 54def make_implicant(implicant_string): 55 ret_mask = 0 56 ret_match = 0 57 58 shift = False 59 for char in implicant_string: 60 char = char.lower() 61 if shift: 62 ret_mask <<= 1 63 ret_match <<= 1 64 65 shift = True 66 if char == '_': 67 shift = False 68 elif char == '0': 69 ret_mask \|= 1 70 elif char == '1': 71 ret_mask \|= 1 72 ret_match \|= 1 73 elif char == 'x': 74 pass 75 else: 76 print("Can't parse implicant character", char) 77 78 return (ret_mask, ret_match) 79 80# ,----- 36 thumb 81# \| ,--- 35 bigThumb 82# \| \|,-- 34 aarch64 83a64_inst = make_implicant('0_01xx__xxxx_xxxx_xxxx_xxxx__xxxx_xxxx_xxxx_xxxx') 84a32_inst = make_implicant('0_00xx__xxxx_xxxx_xxxx_xxxx__xxxx_xxxx_xxxx_xxxx') 85t32_inst = make_implicant('1_10xx__xxxx_xxxx_xxxx_xxxx__xxxx_xxxx_xxxx_xxxx') 86t16_inst = make_implicant('1_00xx__xxxx_xxxx_xxxx_xxxx__xxxx_xxxx_xxxx_xxxx') 87any_inst = make_implicant('x_xxxx__xxxx_xxxx_xxxx_xxxx__xxxx_xxxx_xxxx_xxxx') 88# \| \|\| 89any_a64_inst = \ 90 make_implicant('x_x1xx__xxxx_xxxx_xxxx_xxxx__xxxx_xxxx_xxxx_xxxx') 91any_non_a64_inst = \ 92 make_implicant('x_x0xx__xxxx_xxxx_xxxx_xxxx__xxxx_xxxx_xxxx_xxxx') 93 94def encode_opcode(pattern): 95 def encode(opcode_string): 96 a64_mask, a64_match = pattern 97 mask, match = make_implicant(opcode_string) 98 return (a64_mask \| mask), (a64_match \| match) 99 return encode 100 101a64_opcode = encode_opcode(a64_inst) 102a32_opcode = encode_opcode(a32_inst) 103t32_opcode = encode_opcode(t32_inst) 104t16_opcode = encode_opcode(t16_inst) 105 106# These definitions (in some form) should probably be part of TimingExpr 107 108def literal(value): 109 def body(env): 110 ret = TimingExprLiteral() 111 ret.value = value 112 return ret 113 return body 114 115def bin(op, left, right): 116 def body(env): 117 ret = TimingExprBin() 118 ret.op = 'timingExpr' + op 119 ret.left = left(env) 120 ret.right = right(env) 121 return ret 122 return body 123 124def un(op, arg): 125 def body(env): 126 ret = TimingExprUn() 127 ret.op = 'timingExpr' + op 128 ret.arg = arg(env) 129 return ret 130 return body 131 132def ref(name): 133 def body(env): 134 if name in env: 135 ret = TimingExprRef() 136 ret.index = env[name] 137 else: 138 print("Invalid expression name", name) 139 ret = TimingExprNull() 140 return ret 141 return body 142 143def if_expr(cond, true_expr, false_expr): 144 def body(env): 145 ret = TimingExprIf() 146 ret.cond = cond(env) 147 ret.trueExpr = true_expr(env) 148 ret.falseExpr = false_expr(env) 149 return ret 150 return body 151 152def src(index): 153 def body(env): 154 ret = TimingExprSrcReg() 155 ret.index = index 156 return ret 157 return body 158 159def int_reg(reg): 160 def body(env): 161 ret = TimingExprReadIntReg() 162 ret.reg = reg(env) 163 return ret 164 return body 165 166def let(bindings, expr): 167 def body(env): 168 ret = TimingExprLet() 169 let_bindings = [] 170 new_env = {} 171 i = 0 172 173 # Make the sub-expression as null to start with 174 for name, binding in bindings: 175 new_env[name] = i 176 i += 1 177 178 defns = [] 179 # Then apply them to the produced new env
180 for i in xrange(0, len(bindings)):	180 for i in range(0, len(bindings)):
181 name, binding_expr = bindings[i] 182 defns.append(binding_expr(new_env)) 183 184 ret.defns = defns 185 ret.expr = expr(new_env) 186 187 return ret 188 return body 189 190def expr_top(expr): 191 return expr([]) 192 193class HPI_DefaultInt(MinorFUTiming): 194 description = 'HPI_DefaultInt' 195 mask, match = any_non_a64_inst 196 srcRegsRelativeLats = [3, 3, 2, 2, 2, 1, 0] 197 198class HPI_DefaultA64Int(MinorFUTiming): 199 description = 'HPI_DefaultA64Int' 200 mask, match = any_a64_inst 201 # r, l, (c) 202 srcRegsRelativeLats = [2, 2, 2, 0] 203 204class HPI_DefaultMul(MinorFUTiming): 205 description = 'HPI_DefaultMul' 206 mask, match = any_non_a64_inst 207 # f, f, f, r, l, a? 208 srcRegsRelativeLats = [0, 0, 0, 0, 0, 0, 0] 209 210class HPI_DefaultA64Mul(MinorFUTiming): 211 description = 'HPI_DefaultA64Mul' 212 mask, match = any_a64_inst 213 # a (zr for mul), l, r 214 srcRegsRelativeLats = [0, 0, 0, 0] 215 # extraCommitLat = 1 216 217class HPI_DefaultVfp(MinorFUTiming): 218 description = 'HPI_DefaultVfp' 219 mask, match = any_non_a64_inst 220 # cpsr, z, z, z, cpacr, fpexc, l_lo, r_lo, l_hi, r_hi (from vadd2h) 221 srcRegsRelativeLats = [5, 5, 5, 5, 5, 5, 2, 2, 2, 2, 2, 2, 2, 2, 0] 222 223class HPI_DefaultA64Vfp(MinorFUTiming): 224 description = 'HPI_DefaultA64Vfp' 225 mask, match = any_a64_inst 226 # cpsr, cpacr_el1, fpscr_exc, ... 227 srcRegsRelativeLats = [5, 5, 5, 2] 228 229class HPI_FMADD_A64(MinorFUTiming): 230 description = 'HPI_FMADD_A64' 231 mask, match = a64_opcode('0001_1111_0x0x_xxxx__0xxx_xxxx_xxxx_xxxx') 232 # t 233 # cpsr, cpacr_el1, fpscr_exc, 1, 1, 2, 2, 3, 3, fpscr_exc, d, d, d, d 234 srcRegsRelativeLats = [5, 5, 5, 0, 0, 0, 0, 1, 1, 0, 0, 0, 0, 0] 235 236class HPI_FMSUB_D_A64(MinorFUTiming): 237 description = 'HPI_FMSUB_D_A64' 238 mask, match = a64_opcode('0001_1111_0x0x_xxxx__1xxx_xxxx_xxxx_xxxx') 239 # t 240 # cpsr, cpacr_el1, fpscr_exc, 1, 1, 2, 2, 3, 3, fpscr_exc, d, d, d, d 241 srcRegsRelativeLats = [5, 5, 5, 0, 0, 0, 0, 1, 1, 0, 0, 0, 0, 0] 242 243class HPI_FMOV_A64(MinorFUTiming): 244 description = 'HPI_FMOV_A64' 245 mask, match = a64_opcode('0001_1110_0x10_0000__0100_00xx_xxxx_xxxx') 246 # cpsr, cpacr_el1, fpscr_exc, 1, 1, 2, 2, 3, 3, fpscr_exc, d, d, d, d 247 srcRegsRelativeLats = [5, 5, 5, 0] 248 249class HPI_ADD_SUB_vector_scalar_A64(MinorFUTiming): 250 description = 'HPI_ADD_SUB_vector_scalar_A64' 251 mask, match = a64_opcode('01x1_1110_xx1x_xxxx__1000_01xx_xxxx_xxxx') 252 # cpsr, z, z, z, cpacr, fpexc, l0, r0, l1, r1, l2, r2, l3, r3 (for vadd2h) 253 srcRegsRelativeLats = [5, 5, 5, 4] 254 255 256class HPI_ADD_SUB_vector_vector_A64(MinorFUTiming): 257 description = 'HPI_ADD_SUB_vector_vector_A64' 258 mask, match = a64_opcode('0xx0_1110_xx1x_xxxx__1000_01xx_xxxx_xxxx') 259 # cpsr, z, z, z, cpacr, fpexc, l0, r0, l1, r1, l2, r2, l3, r3 (for vadd2h) 260 srcRegsRelativeLats = [5, 5, 5, 4] 261 262class HPI_FDIV_scalar_32_A64(MinorFUTiming): 263 description = 'HPI_FDIV_scalar_32_A64' 264 mask, match = a64_opcode('0001_1110_001x_xxxx__0001_10xx_xxxx_xxxx') 265 extraCommitLat = 6 266 srcRegsRelativeLats = [0, 0, 0, 20, 4] 267 268class HPI_FDIV_scalar_64_A64(MinorFUTiming): 269 description = 'HPI_FDIV_scalar_64_A64' 270 mask, match = a64_opcode('0001_1110_011x_xxxx__0001_10xx_xxxx_xxxx') 271 extraCommitLat = 15 272 srcRegsRelativeLats = [0, 0, 0, 20, 4] 273 274# CINC CINV CSEL CSET CSETM CSINC CSINC CSINV CSINV CSNEG 275class HPI_Cxxx_A64(MinorFUTiming): 276 description = 'HPI_Cxxx_A64' 277 mask, match = a64_opcode('xx01_1010_100x_xxxx_xxxx__0xxx_xxxx_xxxx') 278 srcRegsRelativeLats = [3, 3, 3, 2, 2] 279 280class HPI_DefaultMem(MinorFUTiming): 281 description = 'HPI_DefaultMem' 282 mask, match = any_non_a64_inst 283 srcRegsRelativeLats = [1, 1, 1, 1, 1, 2] 284 # Assume that LDR/STR take 2 cycles for resolving dependencies 285 # (1 + 1 of the FU) 286 extraAssumedLat = 2 287 288class HPI_DefaultMem64(MinorFUTiming): 289 description = 'HPI_DefaultMem64' 290 mask, match = any_a64_inst 291 srcRegsRelativeLats = [2] 292 # Assume that LDR/STR take 2 cycles for resolving dependencies 293 # (1 + 1 of the FU) 294 extraAssumedLat = 3 295 296class HPI_DataProcessingMovShiftr(MinorFUTiming): 297 description = 'HPI_DataProcessingMovShiftr' 298 mask, match = a32_opcode('xxxx_0001_101x_xxxx__xxxx_xxxx_xxx1_xxxx') 299 srcRegsRelativeLats = [3, 3, 2, 2, 2, 1, 0] 300 301class HPI_DataProcessingMayShift(MinorFUTiming): 302 description = 'HPI_DataProcessingMayShift' 303 mask, match = a32_opcode('xxxx_000x_xxxx_xxxx__xxxx_xxxx_xxxx_xxxx') 304 srcRegsRelativeLats = [3, 3, 2, 2, 1, 1, 0] 305 306class HPI_DataProcessingNoShift(MinorFUTiming): 307 description = 'HPI_DataProcessingNoShift' 308 mask, match = a32_opcode('xxxx_000x_xxxx_xxxx__xxxx_0000_0xx0_xxxx') 309 srcRegsRelativeLats = [3, 3, 2, 2, 2, 1, 0] 310 311class HPI_DataProcessingAllowShifti(MinorFUTiming): 312 description = 'HPI_DataProcessingAllowShifti' 313 mask, match = a32_opcode('xxxx_000x_xxxx_xxxx__xxxx_xxxx_xxx0_xxxx') 314 srcRegsRelativeLats = [3, 3, 2, 2, 1, 1, 0] 315 316class HPI_DataProcessingSuppressShift(MinorFUTiming): 317 description = 'HPI_DataProcessingSuppressShift' 318 mask, match = a32_opcode('xxxx_000x_xxxx_xxxx__xxxx_xxxx_xxxx_xxxx') 319 srcRegsRelativeLats = [] 320 suppress = True 321 322class HPI_DataProcessingSuppressBranch(MinorFUTiming): 323 description = 'HPI_DataProcessingSuppressBranch' 324 mask, match = a32_opcode('xxxx_1010_xxxx_xxxx__xxxx_xxxx_xxxx_xxxx') 325 srcRegsRelativeLats = [] 326 suppress = True 327 328class HPI_BFI_T1(MinorFUTiming): 329 description = 'HPI_BFI_T1' 330 mask, match = t32_opcode('1111_0x11_0110_xxxx__0xxx_xxxx_xxxx_xxxx') 331 srcRegsRelativeLats = [0, 0, 0, 1, 1, 0] 332 333class HPI_BFI_A1(MinorFUTiming): 334 description = 'HPI_BFI_A1' 335 mask, match = a32_opcode('xxxx_0111_110x_xxxx__xxxx_xxxx_x001_xxxx') 336 # f, f, f, dest, src 337 srcRegsRelativeLats = [0, 0, 0, 1, 1, 0] 338 339class HPI_CLZ_T1(MinorFUTiming): 340 description = 'HPI_CLZ_T1' 341 mask, match = t32_opcode('1111_1010_1011_xxxx__1111_xxxx_1000_xxxx') 342 srcRegsRelativeLats = [3, 3, 2, 2, 2, 1, 0] 343 344class HPI_CLZ_A1(MinorFUTiming): 345 description = 'HPI_CLZ_A1' 346 mask, match = a32_opcode('xxxx_0001_0110_xxxx__xxxx_xxxx_0001_xxxx') 347 srcRegsRelativeLats = [3, 3, 2, 2, 2, 1, 0] 348 349class HPI_CMN_immediate_A1(MinorFUTiming): 350 description = 'HPI_CMN_immediate_A1' 351 mask, match = a32_opcode('xxxx_0011_0111_xxxx__xxxx_xxxx_xxxx_xxxx') 352 srcRegsRelativeLats = [3, 3, 3, 2, 2, 3, 3, 3, 0] 353 354class HPI_CMN_register_A1(MinorFUTiming): 355 description = 'HPI_CMN_register_A1' 356 mask, match = a32_opcode('xxxx_0001_0111_xxxx__xxxx_xxxx_xxx0_xxxx') 357 srcRegsRelativeLats = [3, 3, 3, 2, 2, 3, 3, 3, 0] 358 359class HPI_CMP_immediate_A1(MinorFUTiming): 360 description = 'HPI_CMP_immediate_A1' 361 mask, match = a32_opcode('xxxx_0011_0101_xxxx__xxxx_xxxx_xxxx_xxxx') 362 srcRegsRelativeLats = [3, 3, 3, 2, 2, 3, 3, 3, 0] 363 364class HPI_CMP_register_A1(MinorFUTiming): 365 description = 'HPI_CMP_register_A1' 366 mask, match = a32_opcode('xxxx_0001_0101_xxxx__xxxx_xxxx_xxx0_xxxx') 367 srcRegsRelativeLats = [3, 3, 3, 2, 2, 3, 3, 3, 0] 368 369class HPI_MLA_T1(MinorFUTiming): 370 description = 'HPI_MLA_T1' 371 mask, match = t32_opcode('1111_1011_0000_xxxx__xxxx_xxxx_0000_xxxx') 372 # z, z, z, a, l?, r? 373 srcRegsRelativeLats = [0, 0, 0, 0, 0, 2, 0] 374 375class HPI_MLA_A1(MinorFUTiming): 376 description = 'HPI_MLA_A1' 377 mask, match = a32_opcode('xxxx_0000_001x_xxxx__xxxx_xxxx_1001_xxxx') 378 # z, z, z, a, l?, r? 379 srcRegsRelativeLats = [0, 0, 0, 0, 0, 2, 0] 380 381class HPI_MADD_A64(MinorFUTiming): 382 description = 'HPI_MADD_A64' 383 mask, match = a64_opcode('x001_1011_000x_xxxx__0xxx_xxxx_xxxx_xxxx') 384 # a, l?, r? 385 srcRegsRelativeLats = [1, 1, 1, 0] 386 extraCommitLat = 1 387 388class HPI_MLS_T1(MinorFUTiming): 389 description = 'HPI_MLS_T1' 390 mask, match = t32_opcode('1111_1011_0000_xxxx__xxxx_xxxx_0001_xxxx') 391 # z, z, z, l?, a, r? 392 srcRegsRelativeLats = [0, 0, 0, 2, 0, 0, 0] 393 394class HPI_MLS_A1(MinorFUTiming): 395 description = 'HPI_MLS_A1' 396 mask, match = a32_opcode('xxxx_0000_0110_xxxx__xxxx_xxxx_1001_xxxx') 397 # z, z, z, l?, a, r? 398 srcRegsRelativeLats = [0, 0, 0, 2, 0, 0, 0] 399 400class HPI_MOVT_A1(MinorFUTiming): 401 description = 'HPI_MOVT_A1' 402 mask, match = t32_opcode('xxxx_0010_0100_xxxx__xxxx_xxxx_xxxx_xxxx') 403 404class HPI_MUL_T1(MinorFUTiming): 405 description = 'HPI_MUL_T1' 406 mask, match = t16_opcode('0100_0011_01xx_xxxx') 407class HPI_MUL_T2(MinorFUTiming): 408 description = 'HPI_MUL_T2' 409 mask, match = t32_opcode('1111_1011_0000_xxxx_1111_xxxx_0000_xxxx') 410 411class HPI_PKH_T1(MinorFUTiming): 412 description = 'HPI_PKH_T1' 413 mask, match = t32_opcode('1110_1010_110x_xxxx__xxxx_xxxx_xxxx_xxxx') 414 srcRegsRelativeLats = [0, 0, 0, 2, 1, 0] 415 416class HPI_PKH_A1(MinorFUTiming): 417 description = 'HPI_PKH_A1' 418 mask, match = a32_opcode('xxxx_0110_1000_xxxx__xxxx_xxxx_xx01_xxxx') 419 srcRegsRelativeLats = [0, 0, 0, 2, 1, 0] 420 421class HPI_QADD_QSUB_T1(MinorFUTiming): 422 description = 'HPI_QADD_QSUB_T1' 423 mask, match = t32_opcode('1111_1010_1000_xxxx__1111_xxxx_10x0_xxxx') 424 srcRegsRelativeLats = [0, 0, 0, 1, 1, 0] 425 426class HPI_QADD_QSUB_A1(MinorFUTiming): 427 description = 'HPI_QADD_QSUB_A1' 428 mask, match = a32_opcode('xxxx_0001_00x0_xxxx__xxxx_xxxx_0101_xxxx') 429 srcRegsRelativeLats = [0, 0, 0, 1, 1, 0] 430 431# T1 QADD16 QADD8 QSUB16 QSUB8 UQADD16 UQADD8 UQSUB16 UQSUB8 432class HPI_QADD_ETC_T1(MinorFUTiming): 433 description = 'HPI_QADD_ETC_T1' 434 mask, match = t32_opcode('1111_1010_1x0x_xxxx__1111_xxxx_0x01_xxxx') 435 srcRegsRelativeLats = [0, 0, 0, 1, 1, 0] 436 437# A1 QADD16 QADD8 QSAX QSUB16 QSUB8 UQADD16 UQADD8 UQASX UQSAX UQSUB16 UQSUB8 438class HPI_QADD_ETC_A1(MinorFUTiming): 439 description = 'HPI_QADD_ETC_A1' 440 mask, match = a32_opcode('xxxx_0110_0x10_xxxx__xxxx_xxxx_xxx1_xxxx') 441 srcRegsRelativeLats = [0, 0, 0, 1, 1, 0] 442 443class HPI_QASX_QSAX_UQASX_UQSAX_T1(MinorFUTiming): 444 description = 'HPI_QASX_QSAX_UQASX_UQSAX_T1' 445 mask, match = t32_opcode('1111_1010_1x10_xxxx__1111_xxxx_0x01_xxxx') 446 srcRegsRelativeLats = [0, 0, 0, 1, 1, 0] 447 448class HPI_QDADD_QDSUB_T1(MinorFUTiming): 449 description = 'HPI_QDADD_QDSUB_T1' 450 mask, match = t32_opcode('1111_1010_1000_xxxx__1111_xxxx_10x1_xxxx') 451 srcRegsRelativeLats = [0, 0, 0, 0, 1, 0] 452 453class HPI_QDADD_QDSUB_A1(MinorFUTiming): 454 description = 'HPI_QDADD_QSUB_A1' 455 mask, match = a32_opcode('xxxx_0001_01x0_xxxx__xxxx_xxxx_0101_xxxx') 456 srcRegsRelativeLats = [0, 0, 0, 0, 1, 0] 457 458class HPI_RBIT_A1(MinorFUTiming): 459 description = 'HPI_RBIT_A1' 460 mask, match = a32_opcode('xxxx_0110_1111_xxxx__xxxx_xxxx_0011_xxxx') 461 srcRegsRelativeLats = [0, 0, 0, 1, 0] 462 463class HPI_REV_REV16_A1(MinorFUTiming): 464 description = 'HPI_REV_REV16_A1' 465 mask, match = a32_opcode('xxxx_0110_1011_xxxx__xxxx_xxxx_x011_xxxx') 466 srcRegsRelativeLats = [0, 0, 0, 1, 0] 467 468class HPI_REVSH_A1(MinorFUTiming): 469 description = 'HPI_REVSH_A1' 470 mask, match = a32_opcode('xxxx_0110_1111_xxxx__xxxx_xxxx_1011_xxxx') 471 srcRegsRelativeLats = [0, 0, 0, 1, 0] 472 473class HPI_ADD_ETC_A1(MinorFUTiming): 474 description = 'HPI_ADD_ETC_A1' 475 mask, match = a32_opcode('xxxx_0110_0xx1_xxxx__xxxx_xxxx_x001_xxxx') 476 srcRegsRelativeLats = [0, 0, 0, 2, 2, 0] 477 478class HPI_ADD_ETC_T1(MinorFUTiming): 479 description = 'HPI_ADD_ETC_A1' 480 mask, match = t32_opcode('1111_1010_100x_xxxx__1111_xxxx_0xx0_xxxx') 481 srcRegsRelativeLats = [0, 0, 0, 2, 2, 0] 482 483class HPI_SASX_SHASX_UASX_UHASX_A1(MinorFUTiming): 484 description = 'HPI_SASX_SHASX_UASX_UHASX_A1' 485 mask, match = a32_opcode('xxxx_0110_0xx1_xxxx__xxxx_xxxx_0011_xxxx') 486 srcRegsRelativeLats = [3, 3, 2, 2, 2, 1, 0] 487 488class HPI_SBFX_UBFX_A1(MinorFUTiming): 489 description = 'HPI_SBFX_UBFX_A1' 490 mask, match = a32_opcode('xxxx_0111_1x1x_xxxx__xxxx_xxxx_x101_xxxx') 491 srcRegsRelativeLats = [0, 0, 0, 1, 0] 492 493### SDIV 494 495sdiv_lat_expr = expr_top(let([ 496 ('left', un('SignExtend32To64', int_reg(src(4)))), 497 ('right', un('SignExtend32To64', int_reg(src(3)))), 498 ('either_signed', bin('Or', 499 bin('SLessThan', ref('left'), literal(0)), 500 bin('SLessThan', ref('right'), literal(0)))), 501 ('left_size', un('SizeInBits', un('Abs', ref('left')))), 502 ('signed_adjust', if_expr(ref('either_signed'), literal(1), literal(0))), 503 ('right_size', un('SizeInBits', 504 bin('UDiv', un('Abs', ref('right')), 505 if_expr(ref('either_signed'), literal(4), literal(2))))), 506 ('left_minus_right', if_expr( 507 bin('SLessThan', ref('left_size'), ref('right_size')), 508 literal(0), 509 bin('Sub', ref('left_size'), ref('right_size')))) 510 ], 511 bin('Add', 512 ref('signed_adjust'), 513 if_expr(bin('Equal', ref('right'), literal(0)), 514 literal(0), 515 bin('UDiv', ref('left_minus_right'), literal(4)))) 516 )) 517 518sdiv_lat_expr64 = expr_top(let([ 519 ('left', un('SignExtend32To64', int_reg(src(0)))), 520 ('right', un('SignExtend32To64', int_reg(src(1)))), 521 ('either_signed', bin('Or', 522 bin('SLessThan', ref('left'), literal(0)), 523 bin('SLessThan', ref('right'), literal(0)))), 524 ('left_size', un('SizeInBits', un('Abs', ref('left')))), 525 ('signed_adjust', if_expr(ref('either_signed'), literal(1), literal(0))), 526 ('right_size', un('SizeInBits', 527 bin('UDiv', un('Abs', ref('right')), 528 if_expr(ref('either_signed'), literal(4), literal(2))))), 529 ('left_minus_right', if_expr( 530 bin('SLessThan', ref('left_size'), ref('right_size')), 531 literal(0), 532 bin('Sub', ref('left_size'), ref('right_size')))) 533 ], 534 bin('Add', 535 ref('signed_adjust'), 536 if_expr(bin('Equal', ref('right'), literal(0)), 537 literal(0), 538 bin('UDiv', ref('left_minus_right'), literal(4)))) 539 )) 540 541class HPI_SDIV_A1(MinorFUTiming): 542 description = 'HPI_SDIV_A1' 543 mask, match = a32_opcode('xxxx_0111_0001_xxxx__xxxx_xxxx_0001_xxxx') 544 extraCommitLat = 0 545 srcRegsRelativeLats = [] 546 extraCommitLatExpr = sdiv_lat_expr 547 548class HPI_SDIV_A64(MinorFUTiming): 549 description = 'HPI_SDIV_A64' 550 mask, match = a64_opcode('x001_1010_110x_xxxx__0000_11xx_xxxx_xxxx') 551 extraCommitLat = 0 552 srcRegsRelativeLats = [] 553 extraCommitLatExpr = sdiv_lat_expr64 554 555### SEL 556 557class HPI_SEL_A1(MinorFUTiming): 558 description = 'HPI_SEL_A1' 559 mask, match = a32_opcode('xxxx_0110_1000_xxxx__xxxx_xxxx_1011_xxxx') 560 srcRegsRelativeLats = [0, 0, 0, 0, 2, 2, 0] 561 562class HPI_SEL_A1_Suppress(MinorFUTiming): 563 description = 'HPI_SEL_A1_Suppress' 564 mask, match = a32_opcode('xxxx_0110_1000_xxxx__xxxx_xxxx_1011_xxxx') 565 srcRegsRelativeLats = [] 566 suppress = True 567 568class HPI_SHSAX_SSAX_UHSAX_USAX_A1(MinorFUTiming): 569 description = 'HPI_SHSAX_SSAX_UHSAX_USAX_A1' 570 mask, match = a32_opcode('xxxx_0110_0xx1_xxxx__xxxx_xxxx_0101_xxxx') 571 # As Default 572 srcRegsRelativeLats = [3, 3, 2, 2, 2, 1, 0] 573 574class HPI_USUB_ETC_A1(MinorFUTiming): 575 description = 'HPI_USUB_ETC_A1' 576 mask, match = a32_opcode('xxxx_0110_0xx1_xxxx__xxxx_xxxx_x111_xxxx') 577 srcRegsRelativeLats = [0, 0, 0, 2, 2, 0] 578 579class HPI_SMLABB_T1(MinorFUTiming): 580 description = 'HPI_SMLABB_T1' 581 mask, match = t32_opcode('1111_1011_0001_xxxx__xxxx_xxxx_00xx_xxxx') 582 srcRegsRelativeLats = [0, 0, 0, 0, 0, 2, 0] 583 584class HPI_SMLABB_A1(MinorFUTiming): 585 description = 'HPI_SMLABB_A1' 586 mask, match = a32_opcode('xxxx_0001_0000_xxxx__xxxx_xxxx_1xx0_xxxx') 587 srcRegsRelativeLats = [0, 0, 0, 0, 0, 2, 0] 588 589class HPI_SMLAD_T1(MinorFUTiming): 590 description = 'HPI_SMLAD_T1' 591 mask, match = t32_opcode('1111_1011_0010_xxxx__xxxx_xxxx_000x_xxxx') 592 srcRegsRelativeLats = [0, 0, 0, 0, 0, 2, 0] 593 594class HPI_SMLAD_A1(MinorFUTiming): 595 description = 'HPI_SMLAD_A1' 596 mask, match = a32_opcode('xxxx_0111_0000_xxxx__xxxx_xxxx_00x1_xxxx') 597 # z, z, z, l, r, a 598 srcRegsRelativeLats = [0, 0, 0, 0, 0, 2, 0] 599 600class HPI_SMLAL_T1(MinorFUTiming): 601 description = 'HPI_SMLAL_T1' 602 mask, match = t32_opcode('1111_1011_1100_xxxx__xxxx_xxxx_0000_xxxx') 603class HPI_SMLAL_A1(MinorFUTiming): 604 description = 'HPI_SMLAL_A1' 605 mask, match = a32_opcode('xxxx_0000_111x_xxxx__xxxx_xxxx_1001_xxxx') 606 607class HPI_SMLALBB_T1(MinorFUTiming): 608 description = 'HPI_SMLALBB_T1' 609 mask, match = t32_opcode('1111_1011_1100_xxxx__xxxx_xxxx_10xx_xxxx') 610class HPI_SMLALBB_A1(MinorFUTiming): 611 description = 'HPI_SMLALBB_A1' 612 mask, match = a32_opcode('xxxx_0001_0100_xxxx__xxxx_xxxx_1xx0_xxxx') 613 614class HPI_SMLALD_T1(MinorFUTiming): 615 description = 'HPI_SMLALD_T1' 616 mask, match = t32_opcode('1111_1011_1100_xxxx__xxxx_xxxx_110x_xxxx') 617class HPI_SMLALD_A1(MinorFUTiming): 618 description = 'HPI_SMLALD_A1' 619 mask, match = a32_opcode('xxxx_0111_0100_xxxx__xxxx_xxxx_00x1_xxxx') 620 621class HPI_SMLAWB_T1(MinorFUTiming): 622 description = 'HPI_SMLAWB_T1' 623 mask, match = t32_opcode('1111_1011_0011_xxxx__xxxx_xxxx_000x_xxxx') 624 srcRegsRelativeLats = [0, 0, 0, 0, 0, 2, 0] 625 626class HPI_SMLAWB_A1(MinorFUTiming): 627 description = 'HPI_SMLAWB_A1' 628 mask, match = a32_opcode('xxxx_0001_0010_xxxx__xxxx_xxxx_1x00_xxxx') 629 srcRegsRelativeLats = [0, 0, 0, 0, 0, 2, 0] 630 631class HPI_SMLSD_A1(MinorFUTiming): 632 description = 'HPI_SMLSD_A1' 633 mask, match = a32_opcode('xxxx_0111_0000_xxxx__xxxx_xxxx_01x1_xxxx') 634 635class HPI_SMLSLD_T1(MinorFUTiming): 636 description = 'HPI_SMLSLD_T1' 637 mask, match = t32_opcode('1111_1011_1101_xxxx__xxxx_xxxx_110x_xxxx') 638class HPI_SMLSLD_A1(MinorFUTiming): 639 description = 'HPI_SMLSLD_A1' 640 mask, match = a32_opcode('xxxx_0111_0100_xxxx__xxxx_xxxx_01x1_xxxx') 641 642class HPI_SMMLA_T1(MinorFUTiming): 643 description = 'HPI_SMMLA_T1' 644 mask, match = t32_opcode('1111_1011_0101_xxxx__xxxx_xxxx_000x_xxxx') 645 # ^^^^ != 1111 646 srcRegsRelativeLats = [0, 0, 0, 2, 0, 0, 0] 647 648class HPI_SMMLA_A1(MinorFUTiming): 649 description = 'HPI_SMMLA_A1' 650 # Note that this must be after the encoding for SMMUL 651 mask, match = a32_opcode('xxxx_0111_0101_xxxx__xxxx_xxxx_00x1_xxxx') 652 # ^^^^ != 1111 653 srcRegsRelativeLats = [0, 0, 0, 2, 0, 0, 0] 654 655class HPI_SMMLS_T1(MinorFUTiming): 656 description = 'HPI_SMMLS_T1' 657 mask, match = t32_opcode('1111_1011_0110_xxxx__xxxx_xxxx_000x_xxxx') 658 srcRegsRelativeLats = [0, 0, 0, 2, 0, 0, 0] 659 660class HPI_SMMLS_A1(MinorFUTiming): 661 description = 'HPI_SMMLS_A1' 662 mask, match = a32_opcode('xxxx_0111_0101_xxxx__xxxx_xxxx_11x1_xxxx') 663 srcRegsRelativeLats = [0, 0, 0, 2, 0, 0, 0] 664 665class HPI_SMMUL_T1(MinorFUTiming): 666 description = 'HPI_SMMUL_T1' 667 mask, match = t32_opcode('1111_1011_0101_xxxx__1111_xxxx_000x_xxxx') 668 srcRegsRelativeLats = [0, 0, 0, 0, 0, 0] 669 670class HPI_SMMUL_A1(MinorFUTiming): 671 description = 'HPI_SMMUL_A1' 672 mask, match = a32_opcode('xxxx_0111_0101_xxxx__1111_xxxx_00x1_xxxx') 673 srcRegsRelativeLats = [0, 0, 0, 0, 0, 0] 674 675class HPI_SMUAD_T1(MinorFUTiming): 676 description = 'HPI_SMUAD_T1' 677 mask, match = t32_opcode('1111_1011_0010_xxxx__1111_xxxx_000x_xxxx') 678class HPI_SMUAD_A1(MinorFUTiming): 679 description = 'HPI_SMUAD_A1' 680 mask, match = a32_opcode('xxxx_0111_0000_xxxx__1111_xxxx_00x1_xxxx') 681 682class HPI_SMULBB_T1(MinorFUTiming): 683 description = 'HPI_SMULBB_T1' 684 mask, match = t32_opcode('1111_1011_0001_xxxx__1111_xxxx_00xx_xxxx') 685class HPI_SMULBB_A1(MinorFUTiming): 686 description = 'HPI_SMULBB_A1' 687 mask, match = a32_opcode('xxxx_0001_0110_xxxx__xxxx_xxxx_1xx0_xxxx') 688 689class HPI_SMULL_T1(MinorFUTiming): 690 description = 'HPI_SMULL_T1' 691 mask, match = t32_opcode('1111_1011_1000_xxxx__xxxx_xxxx_0000_xxxx') 692class HPI_SMULL_A1(MinorFUTiming): 693 description = 'HPI_SMULL_A1' 694 mask, match = a32_opcode('xxxx_0000_110x_xxxx__xxxx_xxxx_1001_xxxx') 695 696class HPI_SMULWB_T1(MinorFUTiming): 697 description = 'HPI_SMULWB_T1' 698 mask, match = t32_opcode('1111_1011_0011_xxxx__1111_xxxx_000x_xxxx') 699class HPI_SMULWB_A1(MinorFUTiming): 700 description = 'HPI_SMULWB_A1' 701 mask, match = a32_opcode('xxxx_0001_0010_xxxx__xxxx_xxxx_1x10_xxxx') 702 703class HPI_SMUSD_T1(MinorFUTiming): 704 description = 'HPI_SMUSD_T1' 705 mask, match = t32_opcode('1111_1011_0100_xxxx__1111_xxxx_000x_xxxx') 706class HPI_SMUSD_A1(MinorFUTiming): 707 description = 'HPI_SMUSD_A1' 708 mask, match = a32_opcode('xxxx_0111_0000_xxxx__1111_xxxx_01x1_xxxx') 709 710class HPI_SSAT_USAT_no_shift_A1(MinorFUTiming): 711 description = 'HPI_SSAT_USAT_no_shift_A1' 712 # Order before shift 713 mask, match = a32_opcode('xxxx_0110_1x1x_xxxx__xxxx_0000_0001_xxxx') 714 srcRegsRelativeLats = [0, 0, 0, 2, 0] 715 716class HPI_SSAT_USAT_shift_A1(MinorFUTiming): 717 description = 'HPI_SSAT_USAT_shift_A1' 718 # Order after shift 719 mask, match = a32_opcode('xxxx_0110_1x1x_xxxx__xxxx_xxxx_xx01_xxxx') 720 srcRegsRelativeLats = [0, 0, 0, 1, 0] 721 722class HPI_SSAT16_USAT16_A1(MinorFUTiming): 723 description = 'HPI_SSAT16_USAT16_A1' 724 mask, match = a32_opcode('xxxx_0110_1x10_xxxx__xxxx_xxxx_0011_xxxx') 725 srcRegsRelativeLats = [0, 0, 0, 2, 0] 726 727class HPI_SXTAB_T1(MinorFUTiming): 728 description = 'HPI_SXTAB_T1' 729 mask, match = t32_opcode('1111_1010_0100_xxxx__1111_xxxx_1xxx_xxxx') 730 srcRegsRelativeLats = [0, 0, 0, 1, 2, 0] 731 732class HPI_SXTAB_SXTAB16_SXTAH_UXTAB_UXTAB16_UXTAH_A1(MinorFUTiming): 733 description = 'HPI_SXTAB_SXTAB16_SXTAH_UXTAB_UXTAB16_UXTAH_A1' 734 # Place AFTER HPI_SXTB_SXTB16_SXTH_UXTB_UXTB16_UXTH_A1 735 # e6[9d][^f]0070 are undefined 736 mask, match = a32_opcode('xxxx_0110_1xxx_xxxx__xxxx_xxxx_0111_xxxx') 737 srcRegsRelativeLats = [0, 0, 0, 1, 2, 0] 738 739class HPI_SXTAB16_T1(MinorFUTiming): 740 description = 'HPI_SXTAB16_T1' 741 mask, match = t32_opcode('1111_1010_0010_xxxx__1111_xxxx_1xxx_xxxx') 742 srcRegsRelativeLats = [0, 0, 0, 1, 2, 0] 743 744class HPI_SXTAH_T1(MinorFUTiming): 745 description = 'HPI_SXTAH_T1' 746 mask, match = t32_opcode('1111_1010_0000_xxxx__1111_xxxx_1xxx_xxxx') 747 srcRegsRelativeLats = [0, 0, 0, 1, 2, 0] 748 749class HPI_SXTB_T1(MinorFUTiming): 750 description = 'HPI_SXTB_T1' 751 mask, match = t16_opcode('1011_0010_01xx_xxxx') 752class HPI_SXTB_T2(MinorFUTiming): 753 description = 'HPI_SXTB_T2' 754 mask, match = t32_opcode('1111_1010_0100_1111__1111_xxxx_1xxx_xxxx') 755 srcRegsRelativeLats = [0, 0, 0, 1, 2, 0] 756 757class HPI_SXTB_SXTB16_SXTH_UXTB_UXTB16_UXTH_A1(MinorFUTiming): 758 description = 'HPI_SXTB_SXTB16_SXTH_UXTB_UXTB16_UXTH_A1' 759 # e6[9d]f0070 are undefined 760 mask, match = a32_opcode('xxxx_0110_1xxx_1111__xxxx_xxxx_0111_xxxx') 761 srcRegsRelativeLats = [0, 0, 0, 2, 0] 762 763class HPI_SXTB16_T1(MinorFUTiming): 764 description = 'HPI_SXTB16_T1' 765 mask, match = t32_opcode('1111_1010_0010_1111__1111_xxxx_1xxx_xxxx') 766 srcRegsRelativeLats = [0, 0, 0, 1, 2, 0] 767 768class HPI_SXTH_T1(MinorFUTiming): 769 description = 'HPI_SXTH_T1' 770 mask, match = t16_opcode('1011_0010_00xx_xxxx') 771class HPI_SXTH_T2(MinorFUTiming): 772 description = 'HPI_SXTH_T2' 773 mask, match = t32_opcode('1111_1010_0000_1111__1111_xxxx_1xxx_xxxx') 774 srcRegsRelativeLats = [0, 0, 0, 1, 2, 0] 775 776class HPI_UDIV_T1(MinorFUTiming): 777 description = 'HPI_UDIV_T1' 778 mask, match = t32_opcode('1111_1011_1011_xxxx__xxxx_xxxx_1111_xxxx') 779 780udiv_lat_expr = expr_top(let([ 781 ('left', int_reg(src(4))), 782 ('right', int_reg(src(3))), 783 ('left_size', un('SizeInBits', ref('left'))), 784 ('right_size', un('SizeInBits', 785 bin('UDiv', ref('right'), literal(2)))), 786 ('left_minus_right', if_expr( 787 bin('SLessThan', ref('left_size'), ref('right_size')), 788 literal(0), 789 bin('Sub', ref('left_size'), ref('right_size')))) 790 ], 791 if_expr(bin('Equal', ref('right'), literal(0)), 792 literal(0), 793 bin('UDiv', ref('left_minus_right'), literal(4))) 794 )) 795 796class HPI_UDIV_A1(MinorFUTiming): 797 description = 'HPI_UDIV_A1' 798 mask, match = a32_opcode('xxxx_0111_0011_xxxx__xxxx_xxxx_0001_xxxx') 799 extraCommitLat = 0 800 srcRegsRelativeLats = [] 801 extraCommitLatExpr = udiv_lat_expr 802 803class HPI_UMAAL_T1(MinorFUTiming): 804 description = 'HPI_UMAAL_T1' 805 mask, match = t32_opcode('1111_1011_1110_xxxx__xxxx_xxxx_0110_xxxx') 806 # z, z, z, dlo, dhi, l, r 807 extraCommitLat = 1 808 srcRegsRelativeLats = [0, 0, 0, 0, 0, 0, 0, 0] 809 810class HPI_UMAAL_A1(MinorFUTiming): 811 description = 'HPI_UMAAL_A1' 812 mask, match = a32_opcode('xxxx_0000_0100_xxxx__xxxx_xxxx_1001_xxxx') 813 # z, z, z, dlo, dhi, l, r 814 extraCommitLat = 1 815 srcRegsRelativeLats = [0, 0, 0, 0, 0, 0, 0, 0] 816 817class HPI_UMLAL_T1(MinorFUTiming): 818 description = 'HPI_UMLAL_T1' 819 mask, match = t32_opcode('1111_1011_1110_xxxx__xxxx_xxxx_0000_xxxx') 820 821class HPI_UMLAL_A1(MinorFUTiming): 822 description = 'HPI_UMLAL_A1' 823 mask, match = t32_opcode('xxxx_0000_101x_xxxx__xxxx_xxxx_1001_xxxx') 824 825class HPI_UMULL_T1(MinorFUTiming): 826 description = 'HPI_UMULL_T1' 827 mask, match = t32_opcode('1111_1011_1010_xxxx__xxxx_xxxx_0000_xxxx') 828 829class HPI_UMULL_A1(MinorFUTiming): 830 description = 'HPI_UMULL_A1' 831 mask, match = a32_opcode('xxxx_0000_100x_xxxx__xxxx_xxxx_1001_xxxx') 832 833class HPI_USAD8_USADA8_A1(MinorFUTiming): 834 description = 'HPI_USAD8_USADA8_A1' 835 mask, match = a32_opcode('xxxx_0111_1000_xxxx__xxxx_xxxx_0001_xxxx') 836 srcRegsRelativeLats = [0, 0, 0, 0, 0, 2, 0] 837 838class HPI_USAD8_USADA8_A1_Suppress(MinorFUTiming): 839 description = 'HPI_USAD8_USADA8_A1_Suppress' 840 mask, match = a32_opcode('xxxx_0111_1000_xxxx__xxxx_xxxx_0001_xxxx') 841 srcRegsRelativeLats = [] 842 suppress = True 843 844class HPI_VMOV_immediate_A1(MinorFUTiming): 845 description = 'HPI_VMOV_register_A1' 846 mask, match = a32_opcode('1111_0010_0x10_xxxx_xxxx_0001_xxx1_xxxx') 847 # cpsr, z, z, z, hcptr, nsacr, cpacr, fpexc, scr 848 srcRegsRelativeLats = [5, 5, 5, 5, 5, 5, 5, 5, 5, 0] 849 850class HPI_VMRS_A1(MinorFUTiming): 851 description = 'HPI_VMRS_A1' 852 mask, match = a32_opcode('xxxx_1110_1111_0001_xxxx_1010_xxx1_xxxx') 853 # cpsr,z,z,z,hcptr,nsacr,cpacr,scr,r42 854 srcRegsRelativeLats = [5, 5, 5, 5, 5, 5, 5, 5, 5, 0] 855 856class HPI_VMOV_register_A2(MinorFUTiming): 857 description = 'HPI_VMOV_register_A2' 858 mask, match = a32_opcode('xxxx_1110_1x11_0000_xxxx_101x_01x0_xxxx') 859 # cpsr, z, r39, z, hcptr, nsacr, cpacr, fpexc, scr, f4, f5, f0, f1 860 srcRegsRelativeLats = \ 861 [5, 5, 5, 5, 5, 5, 5, 5, 5, 4, 4, 4, 4, 4, 4, 4, 4, 0] 862 863# VADD.I16 D/VADD.F32 D/VADD.I8 D/VADD.I32 D 864class HPI_VADD2H_A32(MinorFUTiming): 865 description = 'Vadd2hALU' 866 mask, match = a32_opcode('1111_0010_0xxx_xxxx__xxxx_1000_xxx0_xxxx') 867 # cpsr, z, z, z, cpacr, fpexc, l0, r0, l1, r1, l2, r2, l3, r3 (for vadd2h) 868 srcRegsRelativeLats = [0, 0, 0, 0, 0, 0, 4, 4, 4, 4, 4, 4, 4, 4, 0] 869 870# VAQQHN.I16 Q/VAQQHN.I32 Q/VAQQHN.I64 Q 871class HPI_VADDHN_A32(MinorFUTiming): 872 description = 'VaddhnALU' 873 mask, match = a32_opcode('1111_0010_1xxx_xxxx__xxxx_0100_x0x0_xxxx') 874 # cpsr, z, z, z, cpacr, fpexc, l0, l1, l2, l3, r0, r1, r2, r3 875 srcRegsRelativeLats = [0, 0, 0, 0, 0, 0, 3, 3, 3, 3, 3, 3, 3, 3, 0] 876 877class HPI_VADDL_A32(MinorFUTiming): 878 description = 'VaddlALU' 879 mask, match = a32_opcode('1111_001x_1xxx_xxxx__xxxx_0000_x0x0_xxxx') 880 srcRegsRelativeLats = [0, 0, 0, 0, 0, 0, 3, 3, 3, 3, 3, 3, 3, 3, 0] 881 882class HPI_VADDW_A32(MinorFUTiming): 883 description = 'HPI_VADDW_A32' 884 mask, match = a32_opcode('1111_001x_1xxx_xxxx__xxxx_0001_x0x0_xxxx') 885 # cpsr, z, z, z, cpacr, fpexc, l0, l1, l2, l3, r0, r1 886 srcRegsRelativeLats = [0, 0, 0, 0, 0, 0, 4, 4, 4, 4, 3, 3, 0] 887 888# VHADD/VHSUB S8,S16,S32,U8,U16,U32 Q and D 889class HPI_VHADD_A32(MinorFUTiming): 890 description = 'HPI_VHADD_A32' 891 mask, match = a32_opcode('1111_001x_0xxx_xxxx__xxxx_00x0_xxx0_xxxx') 892 srcRegsRelativeLats = [0, 0, 0, 0, 0, 0, 4, 4, 4, 4, 4, 4, 4, 4, 0] 893 894class HPI_VPADAL_A32(MinorFUTiming): 895 description = 'VpadalALU' 896 mask, match = a32_opcode('1111_0011_1x11_xx00__xxxx_0110_xxx0_xxxx') 897 srcRegsRelativeLats = [0, 0, 0, 0, 0, 0, 2, 2, 2, 2, 2, 2, 2, 2, 0] 898 899# VPADDH.I16 900class HPI_VPADDH_A32(MinorFUTiming): 901 description = 'VpaddhALU' 902 mask, match = a32_opcode('1111_0010_0xxx_xxxx__xxxx_1011_xxx1_xxxx') 903 srcRegsRelativeLats = [0, 0, 0, 0, 0, 0, 3, 3, 3, 3, 3, 3, 3, 3, 0] 904 905# VPADDH.F32 906class HPI_VPADDS_A32(MinorFUTiming): 907 description = 'VpaddsALU' 908 mask, match = a32_opcode('1111_0011_0x0x_xxxx__xxxx_1101_xxx0_xxxx') 909 srcRegsRelativeLats = [0, 0, 0, 0, 0, 0, 2, 2, 2, 2, 2, 2, 2, 2, 0] 910 911# VPADDL.S16 912class HPI_VPADDL_A32(MinorFUTiming): 913 description = 'VpaddlALU' 914 mask, match = a32_opcode('1111_0011_1x11_xx00__xxxx_0010_xxx0_xxxx') 915 srcRegsRelativeLats = [0, 0, 0, 0, 0, 0, 3, 3, 3, 3, 3, 3, 3, 3, 0] 916 917# VRADDHN.I16 918class HPI_VRADDHN_A32(MinorFUTiming): 919 description = 'HPI_VRADDHN_A32' 920 mask, match = a32_opcode('1111_0011_1xxx_xxxx__xxxx_0100_x0x0_xxxx') 921 # cpsr, z, z, z, cpacr, fpexc, l0, l1, l2, l3, r0, r1, r2, r3 922 srcRegsRelativeLats = [0, 0, 0, 0, 0, 0, 4, 4, 4, 4, 4, 4, 4, 4, 0] 923 924class HPI_VRHADD_A32(MinorFUTiming): 925 description = 'VrhaddALU' 926 mask, match = a32_opcode('1111_001x_0xxx_xxxx__xxxx_0001_xxx0_xxxx') 927 srcRegsRelativeLats = [0, 0, 0, 0, 0, 0, 4, 4, 4, 4, 4, 4, 4, 4, 0] 928 929class HPI_VQADD_A32(MinorFUTiming): 930 description = 'VqaddALU' 931 mask, match = a32_opcode('1111_001x_0xxx_xxxx__xxxx_0000_xxx1_xxxx') 932 srcRegsRelativeLats = [0, 0, 0, 0, 0, 0, 3, 3, 3, 3, 3, 3, 3, 3, 0] 933 934class HPI_VANDQ_A32(MinorFUTiming): 935 description = 'VandqALU' 936 mask, match = a32_opcode('1111_0010_0x00_xxxx__xxxx_0001_xxx1_xxxx') 937 srcRegsRelativeLats = [0, 0, 0, 0, 0, 0, 5, 5, 5, 5, 5, 5, 5, 5, 0] 938 939# VMUL (integer) 940class HPI_VMULI_A32(MinorFUTiming): 941 description = 'VmuliALU' 942 mask, match = a32_opcode('1111_001x_0xxx_xxxx__xxxx_1001_xxx1_xxxx') 943 srcRegsRelativeLats = [0, 0, 0, 0, 0, 0, 2, 2, 2, 2, 2, 2, 2, 2, 0] 944 945# VBIC (reg) 946class HPI_VBIC_A32(MinorFUTiming): 947 description = 'VbicALU' 948 mask, match = a32_opcode('1111_0010_0x01_xxxx__xxxx_0001_xxx1_xxxx') 949 srcRegsRelativeLats = [0, 0, 0, 0, 0, 0, 5, 5, 5, 5, 5, 5, 5, 5, 0] 950 951# VBIF VBIT VBSL 952class HPI_VBIF_ETC_A32(MinorFUTiming): 953 description = 'VbifALU' 954 mask, match = a32_opcode('1111_0011_0xxx_xxxx__xxxx_0001_xxx1_xxxx') 955 srcRegsRelativeLats = \ 956 [0, 0, 0, 0, 0, 0, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 0] 957 958class HPI_VACGE_A32(MinorFUTiming): 959 description = 'VacgeALU' 960 mask, match = a32_opcode('1111_0011_0xxx_xxxx__xxxx_1110_xxx1_xxxx') 961 srcRegsRelativeLats = [0, 0, 0, 0, 0, 0, 4, 4, 4, 4, 4, 4, 4, 4, 4, 0] 962 963# VCEQ.F32 964class HPI_VCEQ_A32(MinorFUTiming): 965 description = 'VceqALU' 966 mask, match = a32_opcode('1111_0010_0x0x_xxxx__xxxx_1110_xxx0_xxxx') 967 srcRegsRelativeLats = [0, 0, 0, 0, 0, 0, 4, 4, 4, 4, 4, 4, 4, 4, 4, 0] 968 969# VCEQ.[IS]... register 970class HPI_VCEQI_A32(MinorFUTiming): 971 description = 'VceqiALU' 972 mask, match = a32_opcode('1111_0011_0xxx_xxxx__xxxx_1000_xxx1_xxxx') 973 srcRegsRelativeLats = [0, 0, 0, 0, 0, 0, 4, 4, 4, 4, 4, 4, 4, 4, 4, 0] 974 975# VCEQ.[IS]... immediate 976class HPI_VCEQII_A32(MinorFUTiming): 977 description = 'HPI_VCEQII_A32' 978 mask, match = a32_opcode('1111_0011_1x11_xx01__xxxx_0x01_0xx0_xxxx') 979 srcRegsRelativeLats = [0, 0, 0, 0, 0, 0, 4, 4, 4, 4, 4, 4, 4, 4, 4, 0] 980 981class HPI_VTST_A32(MinorFUTiming): 982 description = 'HPI_VTST_A32' 983 mask, match = a32_opcode('1111_0010_0xxx_xxxx__xxxx_1000_xxx1_xxxx') 984 srcRegsRelativeLats = [0, 0, 0, 0, 0, 0, 3, 3, 3, 3, 3, 3, 3, 3, 3, 0] 985 986class HPI_VCLZ_A32(MinorFUTiming): 987 description = 'HPI_VCLZ_A32' 988 mask, match = a32_opcode('1111_0011_1x11_xx00__xxxx_0100_1xx0_xxxx') 989 srcRegsRelativeLats = [0, 0, 0, 0, 0, 0, 4, 4, 4, 4, 4, 4, 4, 4, 4, 0] 990 991class HPI_VCNT_A32(MinorFUTiming): 992 description = 'HPI_VCNT_A32' 993 mask, match = a32_opcode('1111_0011_1x11_xx00__xxxx_0101_0xx0_xxxx') 994 srcRegsRelativeLats = [0, 0, 0, 0, 0, 0, 4, 4, 4, 4, 4, 4, 4, 4, 4, 0] 995 996class HPI_VEXT_A32(MinorFUTiming): 997 description = 'HPI_VCNT_A32' 998 mask, match = a32_opcode('1111_0010_1x11_xxxx__xxxx_xxxx_xxx0_xxxx') 999 srcRegsRelativeLats = [0, 0, 0, 0, 0, 0, 4, 4, 4, 4, 4, 4, 4, 4, 4, 0] 1000 1001# VMAX VMIN integer 1002class HPI_VMAXI_A32(MinorFUTiming): 1003 description = 'HPI_VMAXI_A32' 1004 mask, match = a32_opcode('1111_001x_0xxx_xxxx__xxxx_0110_xxxx_xxxx') 1005 srcRegsRelativeLats = [0, 0, 0, 0, 0, 0, 4, 4, 4, 4, 4, 4, 4, 4, 4, 0] 1006 1007# VMAX VMIN float 1008class HPI_VMAXS_A32(MinorFUTiming): 1009 description = 'HPI_VMAXS_A32' 1010 mask, match = a32_opcode('1111_0010_0xxx_xxxx__xxxx_1111_xxx0_xxxx') 1011 srcRegsRelativeLats = [0, 0, 0, 0, 0, 0, 2, 2, 2, 2, 2, 2, 2, 2, 2, 0] 1012 1013# VNEG integer 1014class HPI_VNEGI_A32(MinorFUTiming): 1015 description = 'HPI_VNEGI_A32' 1016 mask, match = a32_opcode('1111_0011_1x11_xx01__xxxx_0x11_1xx0_xxxx') 1017 srcRegsRelativeLats = [0, 0, 0, 0, 0, 0, 4, 4, 4, 4, 4, 4, 4, 4, 4, 0] 1018 1019# VNEG float 1020class HPI_VNEGF_A32(MinorFUTiming): 1021 description = 'HPI_VNEGF_A32' 1022 mask, match = a32_opcode('xxxx_1110_1x11_0001__xxxx_101x_01x0_xxxx') 1023 srcRegsRelativeLats = [0, 0, 0, 0, 0, 0, 2, 2, 2, 2, 2, 2, 2, 2, 2, 0] 1024 1025# VREV16 VREV32 VREV64 1026class HPI_VREVN_A32(MinorFUTiming): 1027 description = 'HPI_VREVN_A32' 1028 mask, match = a32_opcode('1111_0011_1x11_xx00__xxxx_000x_xxx0_xxxx') 1029 srcRegsRelativeLats = [0, 0, 0, 0, 0, 0, 4, 4, 4, 4, 4, 4, 4, 4, 4, 0] 1030 1031class HPI_VQNEG_A32(MinorFUTiming): 1032 description = 'HPI_VQNEG_A32' 1033 mask, match = a32_opcode('1111_0011_1x11_xx00__xxxx_0111_1xx0_xxxx') 1034 srcRegsRelativeLats = [0, 0, 0, 0, 0, 0, 3, 3, 3, 3, 3, 3, 3, 3, 3, 0] 1035 1036class HPI_VSWP_A32(MinorFUTiming): 1037 description = 'HPI_VSWP_A32' 1038 mask, match = a32_opcode('1111_0011_1x11_xx10__xxxx_0000_0xx0_xxxx') 1039 srcRegsRelativeLats = [0, 0, 0, 0, 0, 0, 4, 4, 4, 4, 4, 4, 4, 4, 0] 1040 1041class HPI_VTRN_A32(MinorFUTiming): 1042 description = 'HPI_VTRN_A32' 1043 mask, match = a32_opcode('1111_0011_1x11_xx10__xxxx_0000_1xx0_xxxx') 1044 # cpsr, z, z, z, cpact, fpexc, o0, d0, o1, d1, o2, d2, o3, d3 1045 srcRegsRelativeLats = [0, 0, 0, 0, 0, 0, 2, 2, 2, 2, 2, 2, 2, 2, 0] 1046 1047# VQMOVN VQMOVUN 1048class HPI_VQMOVN_A32(MinorFUTiming): 1049 description = 'HPI_VQMOVN_A32' 1050 mask, match = a32_opcode('1111_0011_1x11_xx10__xxxx_0010_xxx0_xxxx') 1051 # cpsr, z, z, z, cpact, fpexc, o[0], o[1], o[2], o[3], fpscr 1052 srcRegsRelativeLats = [0, 0, 0, 0, 0, 0, 2, 2, 2, 2, 2, 0] 1053 1054# VUZP double word 1055class HPI_VUZP_A32(MinorFUTiming): 1056 description = 'HPI_VUZP_A32' 1057 mask, match = a32_opcode('1111_0011_1x11_xx10__xxxx_0001_00x0_xxxx') 1058 srcRegsRelativeLats = [0, 0, 0, 0, 0, 0, 3, 3, 3, 3, 3, 3, 3, 3, 0] 1059 1060# VDIV.F32 1061class HPI_VDIV32_A32(MinorFUTiming): 1062 description = 'HPI_VDIV32_A32' 1063 mask, match = a32_opcode('xxxx_1110_1x00_xxxx__xxxx_1010_x0x0_xxxx') 1064 # cpsr, z, z, z, cpact, fpexc, fpscr_exc, l, r 1065 extraCommitLat = 9 1066 srcRegsRelativeLats = [0, 0, 0, 0, 0, 0, 20, 4, 4, 0] 1067 1068# VDIV.F64 1069class HPI_VDIV64_A32(MinorFUTiming): 1070 description = 'HPI_VDIV64_A32' 1071 mask, match = a32_opcode('xxxx_1110_1x00_xxxx__xxxx_1011_x0x0_xxxx') 1072 # cpsr, z, z, z, cpact, fpexc, fpscr_exc, l, r 1073 extraCommitLat = 18 1074 srcRegsRelativeLats = [0, 0, 0, 0, 0, 0, 20, 4, 4, 0] 1075 1076class HPI_VZIP_A32(MinorFUTiming): 1077 description = 'HPI_VZIP_A32' 1078 mask, match = a32_opcode('1111_0011_1x11_xx10__xxxx_0001_1xx0_xxxx') 1079 srcRegsRelativeLats = [0, 0, 0, 0, 0, 0, 4, 4, 4, 4, 4, 4, 4, 4, 0] 1080 1081# VPMAX integer 1082class HPI_VPMAX_A32(MinorFUTiming): 1083 description = 'HPI_VPMAX_A32' 1084 mask, match = a32_opcode('1111_001x_0xxx_xxxx__xxxx_1010_xxxx_xxxx') 1085 # cpsr, z, z, z, cpact, fpexc, l0, r0, l1, r1, fpscr 1086 srcRegsRelativeLats = [0, 0, 0, 0, 0, 0, 4, 4, 4, 4, 4, 0] 1087 1088# VPMAX float 1089class HPI_VPMAXF_A32(MinorFUTiming): 1090 description = 'HPI_VPMAXF_A32' 1091 mask, match = a32_opcode('1111_0011_0xxx_xxxx__xxxx_1111_xxx0_xxxx') 1092 srcRegsRelativeLats = [0, 0, 0, 0, 0, 0, 2, 2, 2, 2, 0] 1093 1094class HPI_VMOVN_A32(MinorFUTiming): 1095 description = 'HPI_VMOVN_A32' 1096 mask, match = a32_opcode('1111_0011_1x11_xx10__xxxx_0010_00x0_xxxx') 1097 srcRegsRelativeLats = [0, 0, 0, 0, 0, 0, 4, 4, 4, 4, 0] 1098 1099class HPI_VMOVL_A32(MinorFUTiming): 1100 description = 'HPI_VMOVL_A32' 1101 mask, match = a32_opcode('1111_001x_1xxx_x000__xxxx_1010_00x1_xxxx') 1102 srcRegsRelativeLats = [0, 0, 0, 0, 0, 0, 4, 4, 4, 4, 0] 1103 1104# VSQRT.F64 1105class HPI_VSQRT64_A32(MinorFUTiming): 1106 description = 'HPI_VSQRT64_A32' 1107 mask, match = a32_opcode('xxxx_1110_1x11_0001__xxxx_1011_11x0_xxxx') 1108 extraCommitLat = 18 1109 srcRegsRelativeLats = [] 1110 1111# VSQRT.F32 1112class HPI_VSQRT32_A32(MinorFUTiming): 1113 description = 'HPI_VSQRT32_A32' 1114 mask, match = a32_opcode('xxxx_1110_1x11_0001__xxxx_1010_11x0_xxxx') 1115 extraCommitLat = 9 1116 srcRegsRelativeLats = [] 1117 1118class HPI_FloatSimdFU(MinorFU): 1119 opClasses = minorMakeOpClassSet([ 1120 'FloatAdd', 'FloatCmp', 'FloatCvt', 'FloatMult', 'FloatDiv', 1121 'FloatSqrt', 'FloatMisc', 'FloatMultAcc', 1122 'SimdAdd', 'SimdAddAcc', 'SimdAlu', 'SimdCmp', 'SimdCvt', 1123 'SimdMisc', 'SimdMult', 'SimdMultAcc', 'SimdShift', 'SimdShiftAcc', 1124 'SimdSqrt', 'SimdFloatAdd', 'SimdFloatAlu', 'SimdFloatCmp', 1125 'SimdFloatCvt', 'SimdFloatDiv', 'SimdFloatMisc', 'SimdFloatMult', 1126 'SimdFloatMultAcc', 'SimdFloatSqrt']) 1127 1128 timings = [ 1129 # VUZP and VZIP must be before VADDW/L 1130 HPI_VUZP_A32(), HPI_VZIP_A32(), 1131 HPI_VADD2H_A32(), HPI_VADDHN_A32(), 1132 HPI_VADDL_A32(), HPI_VADDW_A32(), 1133 HPI_VHADD_A32(), HPI_VPADAL_A32(), 1134 HPI_VPADDH_A32(), HPI_VPADDS_A32(), 1135 HPI_VPADDL_A32(), HPI_VRADDHN_A32(), 1136 HPI_VRHADD_A32(), HPI_VQADD_A32(), 1137 HPI_VANDQ_A32(), HPI_VBIC_A32(), 1138 HPI_VBIF_ETC_A32(), HPI_VACGE_A32(), 1139 HPI_VCEQ_A32(), HPI_VCEQI_A32(), 1140 HPI_VCEQII_A32(), HPI_VTST_A32(), 1141 HPI_VCLZ_A32(), HPI_VCNT_A32(), 1142 HPI_VEXT_A32(), HPI_VMAXI_A32(), 1143 HPI_VMAXS_A32(), HPI_VNEGI_A32(), 1144 HPI_VNEGF_A32(), HPI_VREVN_A32(), 1145 HPI_VQNEG_A32(), HPI_VSWP_A32(), 1146 HPI_VTRN_A32(), HPI_VPMAX_A32(), 1147 HPI_VPMAXF_A32(), HPI_VMOVN_A32(), 1148 HPI_VMRS_A1(), 1149 HPI_VMOV_immediate_A1(), 1150 HPI_VMOV_register_A2(), 1151 HPI_VQMOVN_A32(), HPI_VMOVL_A32(), 1152 HPI_VDIV32_A32(), HPI_VDIV64_A32(), 1153 HPI_VSQRT32_A32(), HPI_VSQRT64_A32(), 1154 HPI_VMULI_A32(), 1155 # Add before here 1156 HPI_FMADD_A64(), 1157 HPI_FMSUB_D_A64(), 1158 HPI_FMOV_A64(), 1159 HPI_ADD_SUB_vector_scalar_A64(), 1160 HPI_ADD_SUB_vector_vector_A64(), 1161 HPI_FDIV_scalar_32_A64(), HPI_FDIV_scalar_64_A64(), 1162 HPI_DefaultA64Vfp(), 1163 HPI_DefaultVfp()] 1164 1165 opLat = 6 1166 1167class HPI_IntFU(MinorFU): 1168 opClasses = minorMakeOpClassSet(['IntAlu']) 1169 # IMPORTANT! Keep the order below, add new entries at the head 1170 timings = [ 1171 HPI_SSAT_USAT_no_shift_A1(), 1172 HPI_SSAT_USAT_shift_A1(), 1173 HPI_SSAT16_USAT16_A1(), 1174 HPI_QADD_QSUB_A1(), 1175 HPI_QADD_QSUB_T1(), 1176 HPI_QADD_ETC_A1(), 1177 HPI_QASX_QSAX_UQASX_UQSAX_T1(), 1178 HPI_QADD_ETC_T1(), 1179 HPI_USUB_ETC_A1(), 1180 HPI_ADD_ETC_A1(), 1181 HPI_ADD_ETC_T1(), 1182 HPI_QDADD_QDSUB_A1(), 1183 HPI_QDADD_QDSUB_T1(), 1184 HPI_SASX_SHASX_UASX_UHASX_A1(), 1185 HPI_SHSAX_SSAX_UHSAX_USAX_A1(), 1186 HPI_SXTB_SXTB16_SXTH_UXTB_UXTB16_UXTH_A1(), 1187 1188 # Must be after HPI_SXTB_SXTB16_SXTH_UXTB_UXTB16_UXTH_A1 1189 HPI_SXTAB_SXTAB16_SXTAH_UXTAB_UXTAB16_UXTAH_A1(), 1190 1191 HPI_SXTAB_T1(), 1192 HPI_SXTAB16_T1(), 1193 HPI_SXTAH_T1(), 1194 HPI_SXTB_T2(), 1195 HPI_SXTB16_T1(), 1196 HPI_SXTH_T2(), 1197 1198 HPI_PKH_A1(), 1199 HPI_PKH_T1(), 1200 HPI_SBFX_UBFX_A1(), 1201 HPI_SEL_A1(), 1202 HPI_RBIT_A1(), 1203 HPI_REV_REV16_A1(), 1204 HPI_REVSH_A1(), 1205 HPI_USAD8_USADA8_A1(), 1206 HPI_BFI_A1(), 1207 HPI_BFI_T1(), 1208 1209 HPI_CMN_register_A1(), 1210 HPI_CMN_immediate_A1(), 1211 HPI_CMP_register_A1(), 1212 HPI_CMP_immediate_A1(), 1213 1214 HPI_DataProcessingNoShift(), 1215 HPI_DataProcessingMovShiftr(), 1216 HPI_DataProcessingMayShift(), 1217 1218 HPI_Cxxx_A64(), 1219 1220 HPI_DefaultA64Int(), 1221 HPI_DefaultInt()] 1222 opLat = 3 1223 1224class HPI_Int2FU(MinorFU): 1225 opClasses = minorMakeOpClassSet(['IntAlu']) 1226 # IMPORTANT! Keep the order below, add new entries at the head 1227 timings = [ 1228 HPI_SSAT_USAT_no_shift_A1(), 1229 HPI_SSAT_USAT_shift_A1(), 1230 HPI_SSAT16_USAT16_A1(), 1231 HPI_QADD_QSUB_A1(), 1232 HPI_QADD_QSUB_T1(), 1233 HPI_QADD_ETC_A1(), 1234 HPI_QASX_QSAX_UQASX_UQSAX_T1(), 1235 HPI_QADD_ETC_T1(), 1236 HPI_USUB_ETC_A1(), 1237 HPI_ADD_ETC_A1(), 1238 HPI_ADD_ETC_T1(), 1239 HPI_QDADD_QDSUB_A1(), 1240 HPI_QDADD_QDSUB_T1(), 1241 HPI_SASX_SHASX_UASX_UHASX_A1(), 1242 HPI_SHSAX_SSAX_UHSAX_USAX_A1(), 1243 HPI_SXTB_SXTB16_SXTH_UXTB_UXTB16_UXTH_A1(), 1244 1245 # Must be after HPI_SXTB_SXTB16_SXTH_UXTB_UXTB16_UXTH_A1 1246 HPI_SXTAB_SXTAB16_SXTAH_UXTAB_UXTAB16_UXTAH_A1(), 1247 1248 HPI_SXTAB_T1(), 1249 HPI_SXTAB16_T1(), 1250 HPI_SXTAH_T1(), 1251 HPI_SXTB_T2(), 1252 HPI_SXTB16_T1(), 1253 HPI_SXTH_T2(), 1254 1255 HPI_PKH_A1(), 1256 HPI_PKH_T1(), 1257 HPI_SBFX_UBFX_A1(), 1258 HPI_SEL_A1_Suppress(), 1259 HPI_RBIT_A1(), 1260 HPI_REV_REV16_A1(), 1261 HPI_REVSH_A1(), 1262 HPI_USAD8_USADA8_A1_Suppress(), 1263 HPI_BFI_A1(), 1264 HPI_BFI_T1(), 1265 1266 HPI_CMN_register_A1(), # Need to check for shift 1267 HPI_CMN_immediate_A1(), 1268 HPI_CMP_register_A1(), # Need to check for shift 1269 HPI_CMP_immediate_A1(), 1270 1271 HPI_DataProcessingNoShift(), 1272 HPI_DataProcessingAllowShifti(), 1273 # HPI_DataProcessingAllowMovShiftr(), 1274 1275 # Data processing ops that match SuppressShift but are not 1276 # to be suppressed here 1277 HPI_CLZ_A1(), 1278 HPI_CLZ_T1(), 1279 HPI_DataProcessingSuppressShift(), 1280 # Can you dual issue a branch? 1281 # HPI_DataProcessingSuppressBranch(), 1282 HPI_Cxxx_A64(), 1283 1284 HPI_DefaultA64Int(), 1285 HPI_DefaultInt()] 1286 opLat = 3 1287 1288class HPI_IntMulFU(MinorFU): 1289 opClasses = minorMakeOpClassSet(['IntMult']) 1290 timings = [ 1291 HPI_MLA_A1(), HPI_MLA_T1(), 1292 HPI_MLS_A1(), HPI_MLS_T1(), 1293 HPI_SMLABB_A1(), HPI_SMLABB_T1(), 1294 HPI_SMLAWB_A1(), HPI_SMLAWB_T1(), 1295 HPI_SMLAD_A1(), HPI_SMLAD_T1(), 1296 HPI_SMMUL_A1(), HPI_SMMUL_T1(), 1297 # SMMUL_A1 must be before SMMLA_A1 1298 HPI_SMMLA_A1(), HPI_SMMLA_T1(), 1299 HPI_SMMLS_A1(), HPI_SMMLS_T1(), 1300 HPI_UMAAL_A1(), HPI_UMAAL_T1(), 1301 1302 HPI_MADD_A64(), 1303 HPI_DefaultA64Mul(), 1304 HPI_DefaultMul()] 1305 opLat = 3 1306 cantForwardFromFUIndices = [0, 1, 5] # Int1, Int2, Mem 1307 1308class HPI_IntDivFU(MinorFU): 1309 opClasses = minorMakeOpClassSet(['IntDiv']) 1310 timings = [HPI_SDIV_A1(), HPI_UDIV_A1(), 1311 HPI_SDIV_A64()] 1312 issueLat = 3 1313 opLat = 3 1314 1315class HPI_MemFU(MinorFU): 1316 opClasses = minorMakeOpClassSet(['MemRead', 'MemWrite', 'FloatMemRead', 1317 'FloatMemWrite']) 1318 timings = [HPI_DefaultMem(), HPI_DefaultMem64()] 1319 opLat = 1 1320 cantForwardFromFUIndices = [5] # Mem (this FU) 1321 1322class HPI_MiscFU(MinorFU): 1323 opClasses = minorMakeOpClassSet(['IprAccess', 'InstPrefetch']) 1324 opLat = 1 1325 1326class HPI_FUPool(MinorFUPool): 1327 funcUnits = [HPI_IntFU(), # 0 1328 HPI_Int2FU(), # 1 1329 HPI_IntMulFU(), # 2 1330 HPI_IntDivFU(), # 3 1331 HPI_FloatSimdFU(), # 4 1332 HPI_MemFU(), # 5 1333 HPI_MiscFU() # 6 1334 ] 1335 1336class HPI_DTB(ArmTLB): 1337 size = 256 1338 1339class HPI_ITB(ArmTLB): 1340 size = 256 1341 1342class HPI_WalkCache(Cache): 1343 data_latency = 4 1344 tag_latency = 4 1345 response_latency = 4 1346 mshrs = 6 1347 tgts_per_mshr = 8 1348 size = '1kB' 1349 assoc = 8 1350 write_buffers = 16 1351 1352class HPI_BP(TournamentBP): 1353 localPredictorSize = 64 1354 localCtrBits = 2 1355 localHistoryTableSize = 64 1356 globalPredictorSize = 1024 1357 globalCtrBits = 2 1358 choicePredictorSize = 1024 1359 choiceCtrBits = 2 1360 BTBEntries = 128 1361 BTBTagSize = 18 1362 RASSize = 8 1363 instShiftAmt = 2 1364 1365class HPI_ICache(Cache): 1366 data_latency = 1 1367 tag_latency = 1 1368 response_latency = 1 1369 mshrs = 2 1370 tgts_per_mshr = 8 1371 size = '32kB' 1372 assoc = 2 1373 # No prefetcher, this is handled by the core 1374 1375class HPI_DCache(Cache): 1376 data_latency = 1 1377 tag_latency = 1 1378 response_latency = 1 1379 mshrs = 4 1380 tgts_per_mshr = 8 1381 size = '32kB' 1382 assoc = 4 1383 write_buffers = 4 1384 prefetcher = StridePrefetcher( 1385 queue_size=4, 1386 degree=4) 1387 1388class HPI_L2(Cache): 1389 data_latency = 13 1390 tag_latency = 13 1391 response_latency = 5 1392 mshrs = 4 1393 tgts_per_mshr = 8 1394 size = '1024kB' 1395 assoc = 16 1396 write_buffers = 16 1397 # prefetcher FIXME 1398 1399class HPI(MinorCPU): 1400 # Inherit the doc string from the module to avoid repeating it 1401 # here. 1402 __doc__ = __doc__ 1403 1404 fetch1LineSnapWidth = 0 1405 fetch1LineWidth = 0 1406 fetch1FetchLimit = 1 1407 fetch1ToFetch2ForwardDelay = 1 1408 fetch1ToFetch2BackwardDelay = 1 1409 1410 fetch2InputBufferSize = 2 1411 fetch2ToDecodeForwardDelay = 1 1412 fetch2CycleInput = True 1413 1414 decodeInputBufferSize = 3 1415 decodeToExecuteForwardDelay = 1 1416 decodeInputWidth = 2 1417 decodeCycleInput = True 1418 1419 executeInputWidth = 2 1420 executeCycleInput = True 1421 executeIssueLimit = 2 1422 1423 # Only allow one ld/st to be issued but the second ld/st FU allows 1424 # back-to-back loads 1425 executeMemoryIssueLimit = 1 1426 1427 executeCommitLimit = 2 1428 executeMemoryCommitLimit = 1 1429 executeInputBufferSize = 7 1430 1431 executeMaxAccessesInMemory = 2 1432 1433 executeLSQMaxStoreBufferStoresPerCycle = 2 1434 executeLSQRequestsQueueSize = 1 1435 executeLSQTransfersQueueSize = 2 1436 executeLSQStoreBufferSize = 5 1437 executeBranchDelay = 1 1438 executeFuncUnits = HPI_FUPool() 1439 executeSetTraceTimeOnCommit = True 1440 executeSetTraceTimeOnIssue = False 1441 1442 executeAllowEarlyMemoryIssue = True 1443 1444 enableIdling = True 1445 1446 branchPred = HPI_BP() 1447 1448 itb = HPI_ITB() 1449 dtb = HPI_DTB() 1450 1451__all__ = [ 1452 "HPI_BP", 1453 "HPI_ITB", "HPI_DTB", "HPI_WalkCache", 1454 "HPI_ICache", "HPI_DCache", "HPI_L2", 1455 "HPI", 1456]	181 name, binding_expr = bindings[i] 182 defns.append(binding_expr(new_env)) 183 184 ret.defns = defns 185 ret.expr = expr(new_env) 186 187 return ret 188 return body 189 190def expr_top(expr): 191 return expr([]) 192 193class HPI_DefaultInt(MinorFUTiming): 194 description = 'HPI_DefaultInt' 195 mask, match = any_non_a64_inst 196 srcRegsRelativeLats = [3, 3, 2, 2, 2, 1, 0] 197 198class HPI_DefaultA64Int(MinorFUTiming): 199 description = 'HPI_DefaultA64Int' 200 mask, match = any_a64_inst 201 # r, l, (c) 202 srcRegsRelativeLats = [2, 2, 2, 0] 203 204class HPI_DefaultMul(MinorFUTiming): 205 description = 'HPI_DefaultMul' 206 mask, match = any_non_a64_inst 207 # f, f, f, r, l, a? 208 srcRegsRelativeLats = [0, 0, 0, 0, 0, 0, 0] 209 210class HPI_DefaultA64Mul(MinorFUTiming): 211 description = 'HPI_DefaultA64Mul' 212 mask, match = any_a64_inst 213 # a (zr for mul), l, r 214 srcRegsRelativeLats = [0, 0, 0, 0] 215 # extraCommitLat = 1 216 217class HPI_DefaultVfp(MinorFUTiming): 218 description = 'HPI_DefaultVfp' 219 mask, match = any_non_a64_inst 220 # cpsr, z, z, z, cpacr, fpexc, l_lo, r_lo, l_hi, r_hi (from vadd2h) 221 srcRegsRelativeLats = [5, 5, 5, 5, 5, 5, 2, 2, 2, 2, 2, 2, 2, 2, 0] 222 223class HPI_DefaultA64Vfp(MinorFUTiming): 224 description = 'HPI_DefaultA64Vfp' 225 mask, match = any_a64_inst 226 # cpsr, cpacr_el1, fpscr_exc, ... 227 srcRegsRelativeLats = [5, 5, 5, 2] 228 229class HPI_FMADD_A64(MinorFUTiming): 230 description = 'HPI_FMADD_A64' 231 mask, match = a64_opcode('0001_1111_0x0x_xxxx__0xxx_xxxx_xxxx_xxxx') 232 # t 233 # cpsr, cpacr_el1, fpscr_exc, 1, 1, 2, 2, 3, 3, fpscr_exc, d, d, d, d 234 srcRegsRelativeLats = [5, 5, 5, 0, 0, 0, 0, 1, 1, 0, 0, 0, 0, 0] 235 236class HPI_FMSUB_D_A64(MinorFUTiming): 237 description = 'HPI_FMSUB_D_A64' 238 mask, match = a64_opcode('0001_1111_0x0x_xxxx__1xxx_xxxx_xxxx_xxxx') 239 # t 240 # cpsr, cpacr_el1, fpscr_exc, 1, 1, 2, 2, 3, 3, fpscr_exc, d, d, d, d 241 srcRegsRelativeLats = [5, 5, 5, 0, 0, 0, 0, 1, 1, 0, 0, 0, 0, 0] 242 243class HPI_FMOV_A64(MinorFUTiming): 244 description = 'HPI_FMOV_A64' 245 mask, match = a64_opcode('0001_1110_0x10_0000__0100_00xx_xxxx_xxxx') 246 # cpsr, cpacr_el1, fpscr_exc, 1, 1, 2, 2, 3, 3, fpscr_exc, d, d, d, d 247 srcRegsRelativeLats = [5, 5, 5, 0] 248 249class HPI_ADD_SUB_vector_scalar_A64(MinorFUTiming): 250 description = 'HPI_ADD_SUB_vector_scalar_A64' 251 mask, match = a64_opcode('01x1_1110_xx1x_xxxx__1000_01xx_xxxx_xxxx') 252 # cpsr, z, z, z, cpacr, fpexc, l0, r0, l1, r1, l2, r2, l3, r3 (for vadd2h) 253 srcRegsRelativeLats = [5, 5, 5, 4] 254 255 256class HPI_ADD_SUB_vector_vector_A64(MinorFUTiming): 257 description = 'HPI_ADD_SUB_vector_vector_A64' 258 mask, match = a64_opcode('0xx0_1110_xx1x_xxxx__1000_01xx_xxxx_xxxx') 259 # cpsr, z, z, z, cpacr, fpexc, l0, r0, l1, r1, l2, r2, l3, r3 (for vadd2h) 260 srcRegsRelativeLats = [5, 5, 5, 4] 261 262class HPI_FDIV_scalar_32_A64(MinorFUTiming): 263 description = 'HPI_FDIV_scalar_32_A64' 264 mask, match = a64_opcode('0001_1110_001x_xxxx__0001_10xx_xxxx_xxxx') 265 extraCommitLat = 6 266 srcRegsRelativeLats = [0, 0, 0, 20, 4] 267 268class HPI_FDIV_scalar_64_A64(MinorFUTiming): 269 description = 'HPI_FDIV_scalar_64_A64' 270 mask, match = a64_opcode('0001_1110_011x_xxxx__0001_10xx_xxxx_xxxx') 271 extraCommitLat = 15 272 srcRegsRelativeLats = [0, 0, 0, 20, 4] 273 274# CINC CINV CSEL CSET CSETM CSINC CSINC CSINV CSINV CSNEG 275class HPI_Cxxx_A64(MinorFUTiming): 276 description = 'HPI_Cxxx_A64' 277 mask, match = a64_opcode('xx01_1010_100x_xxxx_xxxx__0xxx_xxxx_xxxx') 278 srcRegsRelativeLats = [3, 3, 3, 2, 2] 279 280class HPI_DefaultMem(MinorFUTiming): 281 description = 'HPI_DefaultMem' 282 mask, match = any_non_a64_inst 283 srcRegsRelativeLats = [1, 1, 1, 1, 1, 2] 284 # Assume that LDR/STR take 2 cycles for resolving dependencies 285 # (1 + 1 of the FU) 286 extraAssumedLat = 2 287 288class HPI_DefaultMem64(MinorFUTiming): 289 description = 'HPI_DefaultMem64' 290 mask, match = any_a64_inst 291 srcRegsRelativeLats = [2] 292 # Assume that LDR/STR take 2 cycles for resolving dependencies 293 # (1 + 1 of the FU) 294 extraAssumedLat = 3 295 296class HPI_DataProcessingMovShiftr(MinorFUTiming): 297 description = 'HPI_DataProcessingMovShiftr' 298 mask, match = a32_opcode('xxxx_0001_101x_xxxx__xxxx_xxxx_xxx1_xxxx') 299 srcRegsRelativeLats = [3, 3, 2, 2, 2, 1, 0] 300 301class HPI_DataProcessingMayShift(MinorFUTiming): 302 description = 'HPI_DataProcessingMayShift' 303 mask, match = a32_opcode('xxxx_000x_xxxx_xxxx__xxxx_xxxx_xxxx_xxxx') 304 srcRegsRelativeLats = [3, 3, 2, 2, 1, 1, 0] 305 306class HPI_DataProcessingNoShift(MinorFUTiming): 307 description = 'HPI_DataProcessingNoShift' 308 mask, match = a32_opcode('xxxx_000x_xxxx_xxxx__xxxx_0000_0xx0_xxxx') 309 srcRegsRelativeLats = [3, 3, 2, 2, 2, 1, 0] 310 311class HPI_DataProcessingAllowShifti(MinorFUTiming): 312 description = 'HPI_DataProcessingAllowShifti' 313 mask, match = a32_opcode('xxxx_000x_xxxx_xxxx__xxxx_xxxx_xxx0_xxxx') 314 srcRegsRelativeLats = [3, 3, 2, 2, 1, 1, 0] 315 316class HPI_DataProcessingSuppressShift(MinorFUTiming): 317 description = 'HPI_DataProcessingSuppressShift' 318 mask, match = a32_opcode('xxxx_000x_xxxx_xxxx__xxxx_xxxx_xxxx_xxxx') 319 srcRegsRelativeLats = [] 320 suppress = True 321 322class HPI_DataProcessingSuppressBranch(MinorFUTiming): 323 description = 'HPI_DataProcessingSuppressBranch' 324 mask, match = a32_opcode('xxxx_1010_xxxx_xxxx__xxxx_xxxx_xxxx_xxxx') 325 srcRegsRelativeLats = [] 326 suppress = True 327 328class HPI_BFI_T1(MinorFUTiming): 329 description = 'HPI_BFI_T1' 330 mask, match = t32_opcode('1111_0x11_0110_xxxx__0xxx_xxxx_xxxx_xxxx') 331 srcRegsRelativeLats = [0, 0, 0, 1, 1, 0] 332 333class HPI_BFI_A1(MinorFUTiming): 334 description = 'HPI_BFI_A1' 335 mask, match = a32_opcode('xxxx_0111_110x_xxxx__xxxx_xxxx_x001_xxxx') 336 # f, f, f, dest, src 337 srcRegsRelativeLats = [0, 0, 0, 1, 1, 0] 338 339class HPI_CLZ_T1(MinorFUTiming): 340 description = 'HPI_CLZ_T1' 341 mask, match = t32_opcode('1111_1010_1011_xxxx__1111_xxxx_1000_xxxx') 342 srcRegsRelativeLats = [3, 3, 2, 2, 2, 1, 0] 343 344class HPI_CLZ_A1(MinorFUTiming): 345 description = 'HPI_CLZ_A1' 346 mask, match = a32_opcode('xxxx_0001_0110_xxxx__xxxx_xxxx_0001_xxxx') 347 srcRegsRelativeLats = [3, 3, 2, 2, 2, 1, 0] 348 349class HPI_CMN_immediate_A1(MinorFUTiming): 350 description = 'HPI_CMN_immediate_A1' 351 mask, match = a32_opcode('xxxx_0011_0111_xxxx__xxxx_xxxx_xxxx_xxxx') 352 srcRegsRelativeLats = [3, 3, 3, 2, 2, 3, 3, 3, 0] 353 354class HPI_CMN_register_A1(MinorFUTiming): 355 description = 'HPI_CMN_register_A1' 356 mask, match = a32_opcode('xxxx_0001_0111_xxxx__xxxx_xxxx_xxx0_xxxx') 357 srcRegsRelativeLats = [3, 3, 3, 2, 2, 3, 3, 3, 0] 358 359class HPI_CMP_immediate_A1(MinorFUTiming): 360 description = 'HPI_CMP_immediate_A1' 361 mask, match = a32_opcode('xxxx_0011_0101_xxxx__xxxx_xxxx_xxxx_xxxx') 362 srcRegsRelativeLats = [3, 3, 3, 2, 2, 3, 3, 3, 0] 363 364class HPI_CMP_register_A1(MinorFUTiming): 365 description = 'HPI_CMP_register_A1' 366 mask, match = a32_opcode('xxxx_0001_0101_xxxx__xxxx_xxxx_xxx0_xxxx') 367 srcRegsRelativeLats = [3, 3, 3, 2, 2, 3, 3, 3, 0] 368 369class HPI_MLA_T1(MinorFUTiming): 370 description = 'HPI_MLA_T1' 371 mask, match = t32_opcode('1111_1011_0000_xxxx__xxxx_xxxx_0000_xxxx') 372 # z, z, z, a, l?, r? 373 srcRegsRelativeLats = [0, 0, 0, 0, 0, 2, 0] 374 375class HPI_MLA_A1(MinorFUTiming): 376 description = 'HPI_MLA_A1' 377 mask, match = a32_opcode('xxxx_0000_001x_xxxx__xxxx_xxxx_1001_xxxx') 378 # z, z, z, a, l?, r? 379 srcRegsRelativeLats = [0, 0, 0, 0, 0, 2, 0] 380 381class HPI_MADD_A64(MinorFUTiming): 382 description = 'HPI_MADD_A64' 383 mask, match = a64_opcode('x001_1011_000x_xxxx__0xxx_xxxx_xxxx_xxxx') 384 # a, l?, r? 385 srcRegsRelativeLats = [1, 1, 1, 0] 386 extraCommitLat = 1 387 388class HPI_MLS_T1(MinorFUTiming): 389 description = 'HPI_MLS_T1' 390 mask, match = t32_opcode('1111_1011_0000_xxxx__xxxx_xxxx_0001_xxxx') 391 # z, z, z, l?, a, r? 392 srcRegsRelativeLats = [0, 0, 0, 2, 0, 0, 0] 393 394class HPI_MLS_A1(MinorFUTiming): 395 description = 'HPI_MLS_A1' 396 mask, match = a32_opcode('xxxx_0000_0110_xxxx__xxxx_xxxx_1001_xxxx') 397 # z, z, z, l?, a, r? 398 srcRegsRelativeLats = [0, 0, 0, 2, 0, 0, 0] 399 400class HPI_MOVT_A1(MinorFUTiming): 401 description = 'HPI_MOVT_A1' 402 mask, match = t32_opcode('xxxx_0010_0100_xxxx__xxxx_xxxx_xxxx_xxxx') 403 404class HPI_MUL_T1(MinorFUTiming): 405 description = 'HPI_MUL_T1' 406 mask, match = t16_opcode('0100_0011_01xx_xxxx') 407class HPI_MUL_T2(MinorFUTiming): 408 description = 'HPI_MUL_T2' 409 mask, match = t32_opcode('1111_1011_0000_xxxx_1111_xxxx_0000_xxxx') 410 411class HPI_PKH_T1(MinorFUTiming): 412 description = 'HPI_PKH_T1' 413 mask, match = t32_opcode('1110_1010_110x_xxxx__xxxx_xxxx_xxxx_xxxx') 414 srcRegsRelativeLats = [0, 0, 0, 2, 1, 0] 415 416class HPI_PKH_A1(MinorFUTiming): 417 description = 'HPI_PKH_A1' 418 mask, match = a32_opcode('xxxx_0110_1000_xxxx__xxxx_xxxx_xx01_xxxx') 419 srcRegsRelativeLats = [0, 0, 0, 2, 1, 0] 420 421class HPI_QADD_QSUB_T1(MinorFUTiming): 422 description = 'HPI_QADD_QSUB_T1' 423 mask, match = t32_opcode('1111_1010_1000_xxxx__1111_xxxx_10x0_xxxx') 424 srcRegsRelativeLats = [0, 0, 0, 1, 1, 0] 425 426class HPI_QADD_QSUB_A1(MinorFUTiming): 427 description = 'HPI_QADD_QSUB_A1' 428 mask, match = a32_opcode('xxxx_0001_00x0_xxxx__xxxx_xxxx_0101_xxxx') 429 srcRegsRelativeLats = [0, 0, 0, 1, 1, 0] 430 431# T1 QADD16 QADD8 QSUB16 QSUB8 UQADD16 UQADD8 UQSUB16 UQSUB8 432class HPI_QADD_ETC_T1(MinorFUTiming): 433 description = 'HPI_QADD_ETC_T1' 434 mask, match = t32_opcode('1111_1010_1x0x_xxxx__1111_xxxx_0x01_xxxx') 435 srcRegsRelativeLats = [0, 0, 0, 1, 1, 0] 436 437# A1 QADD16 QADD8 QSAX QSUB16 QSUB8 UQADD16 UQADD8 UQASX UQSAX UQSUB16 UQSUB8 438class HPI_QADD_ETC_A1(MinorFUTiming): 439 description = 'HPI_QADD_ETC_A1' 440 mask, match = a32_opcode('xxxx_0110_0x10_xxxx__xxxx_xxxx_xxx1_xxxx') 441 srcRegsRelativeLats = [0, 0, 0, 1, 1, 0] 442 443class HPI_QASX_QSAX_UQASX_UQSAX_T1(MinorFUTiming): 444 description = 'HPI_QASX_QSAX_UQASX_UQSAX_T1' 445 mask, match = t32_opcode('1111_1010_1x10_xxxx__1111_xxxx_0x01_xxxx') 446 srcRegsRelativeLats = [0, 0, 0, 1, 1, 0] 447 448class HPI_QDADD_QDSUB_T1(MinorFUTiming): 449 description = 'HPI_QDADD_QDSUB_T1' 450 mask, match = t32_opcode('1111_1010_1000_xxxx__1111_xxxx_10x1_xxxx') 451 srcRegsRelativeLats = [0, 0, 0, 0, 1, 0] 452 453class HPI_QDADD_QDSUB_A1(MinorFUTiming): 454 description = 'HPI_QDADD_QSUB_A1' 455 mask, match = a32_opcode('xxxx_0001_01x0_xxxx__xxxx_xxxx_0101_xxxx') 456 srcRegsRelativeLats = [0, 0, 0, 0, 1, 0] 457 458class HPI_RBIT_A1(MinorFUTiming): 459 description = 'HPI_RBIT_A1' 460 mask, match = a32_opcode('xxxx_0110_1111_xxxx__xxxx_xxxx_0011_xxxx') 461 srcRegsRelativeLats = [0, 0, 0, 1, 0] 462 463class HPI_REV_REV16_A1(MinorFUTiming): 464 description = 'HPI_REV_REV16_A1' 465 mask, match = a32_opcode('xxxx_0110_1011_xxxx__xxxx_xxxx_x011_xxxx') 466 srcRegsRelativeLats = [0, 0, 0, 1, 0] 467 468class HPI_REVSH_A1(MinorFUTiming): 469 description = 'HPI_REVSH_A1' 470 mask, match = a32_opcode('xxxx_0110_1111_xxxx__xxxx_xxxx_1011_xxxx') 471 srcRegsRelativeLats = [0, 0, 0, 1, 0] 472 473class HPI_ADD_ETC_A1(MinorFUTiming): 474 description = 'HPI_ADD_ETC_A1' 475 mask, match = a32_opcode('xxxx_0110_0xx1_xxxx__xxxx_xxxx_x001_xxxx') 476 srcRegsRelativeLats = [0, 0, 0, 2, 2, 0] 477 478class HPI_ADD_ETC_T1(MinorFUTiming): 479 description = 'HPI_ADD_ETC_A1' 480 mask, match = t32_opcode('1111_1010_100x_xxxx__1111_xxxx_0xx0_xxxx') 481 srcRegsRelativeLats = [0, 0, 0, 2, 2, 0] 482 483class HPI_SASX_SHASX_UASX_UHASX_A1(MinorFUTiming): 484 description = 'HPI_SASX_SHASX_UASX_UHASX_A1' 485 mask, match = a32_opcode('xxxx_0110_0xx1_xxxx__xxxx_xxxx_0011_xxxx') 486 srcRegsRelativeLats = [3, 3, 2, 2, 2, 1, 0] 487 488class HPI_SBFX_UBFX_A1(MinorFUTiming): 489 description = 'HPI_SBFX_UBFX_A1' 490 mask, match = a32_opcode('xxxx_0111_1x1x_xxxx__xxxx_xxxx_x101_xxxx') 491 srcRegsRelativeLats = [0, 0, 0, 1, 0] 492 493### SDIV 494 495sdiv_lat_expr = expr_top(let([ 496 ('left', un('SignExtend32To64', int_reg(src(4)))), 497 ('right', un('SignExtend32To64', int_reg(src(3)))), 498 ('either_signed', bin('Or', 499 bin('SLessThan', ref('left'), literal(0)), 500 bin('SLessThan', ref('right'), literal(0)))), 501 ('left_size', un('SizeInBits', un('Abs', ref('left')))), 502 ('signed_adjust', if_expr(ref('either_signed'), literal(1), literal(0))), 503 ('right_size', un('SizeInBits', 504 bin('UDiv', un('Abs', ref('right')), 505 if_expr(ref('either_signed'), literal(4), literal(2))))), 506 ('left_minus_right', if_expr( 507 bin('SLessThan', ref('left_size'), ref('right_size')), 508 literal(0), 509 bin('Sub', ref('left_size'), ref('right_size')))) 510 ], 511 bin('Add', 512 ref('signed_adjust'), 513 if_expr(bin('Equal', ref('right'), literal(0)), 514 literal(0), 515 bin('UDiv', ref('left_minus_right'), literal(4)))) 516 )) 517 518sdiv_lat_expr64 = expr_top(let([ 519 ('left', un('SignExtend32To64', int_reg(src(0)))), 520 ('right', un('SignExtend32To64', int_reg(src(1)))), 521 ('either_signed', bin('Or', 522 bin('SLessThan', ref('left'), literal(0)), 523 bin('SLessThan', ref('right'), literal(0)))), 524 ('left_size', un('SizeInBits', un('Abs', ref('left')))), 525 ('signed_adjust', if_expr(ref('either_signed'), literal(1), literal(0))), 526 ('right_size', un('SizeInBits', 527 bin('UDiv', un('Abs', ref('right')), 528 if_expr(ref('either_signed'), literal(4), literal(2))))), 529 ('left_minus_right', if_expr( 530 bin('SLessThan', ref('left_size'), ref('right_size')), 531 literal(0), 532 bin('Sub', ref('left_size'), ref('right_size')))) 533 ], 534 bin('Add', 535 ref('signed_adjust'), 536 if_expr(bin('Equal', ref('right'), literal(0)), 537 literal(0), 538 bin('UDiv', ref('left_minus_right'), literal(4)))) 539 )) 540 541class HPI_SDIV_A1(MinorFUTiming): 542 description = 'HPI_SDIV_A1' 543 mask, match = a32_opcode('xxxx_0111_0001_xxxx__xxxx_xxxx_0001_xxxx') 544 extraCommitLat = 0 545 srcRegsRelativeLats = [] 546 extraCommitLatExpr = sdiv_lat_expr 547 548class HPI_SDIV_A64(MinorFUTiming): 549 description = 'HPI_SDIV_A64' 550 mask, match = a64_opcode('x001_1010_110x_xxxx__0000_11xx_xxxx_xxxx') 551 extraCommitLat = 0 552 srcRegsRelativeLats = [] 553 extraCommitLatExpr = sdiv_lat_expr64 554 555### SEL 556 557class HPI_SEL_A1(MinorFUTiming): 558 description = 'HPI_SEL_A1' 559 mask, match = a32_opcode('xxxx_0110_1000_xxxx__xxxx_xxxx_1011_xxxx') 560 srcRegsRelativeLats = [0, 0, 0, 0, 2, 2, 0] 561 562class HPI_SEL_A1_Suppress(MinorFUTiming): 563 description = 'HPI_SEL_A1_Suppress' 564 mask, match = a32_opcode('xxxx_0110_1000_xxxx__xxxx_xxxx_1011_xxxx') 565 srcRegsRelativeLats = [] 566 suppress = True 567 568class HPI_SHSAX_SSAX_UHSAX_USAX_A1(MinorFUTiming): 569 description = 'HPI_SHSAX_SSAX_UHSAX_USAX_A1' 570 mask, match = a32_opcode('xxxx_0110_0xx1_xxxx__xxxx_xxxx_0101_xxxx') 571 # As Default 572 srcRegsRelativeLats = [3, 3, 2, 2, 2, 1, 0] 573 574class HPI_USUB_ETC_A1(MinorFUTiming): 575 description = 'HPI_USUB_ETC_A1' 576 mask, match = a32_opcode('xxxx_0110_0xx1_xxxx__xxxx_xxxx_x111_xxxx') 577 srcRegsRelativeLats = [0, 0, 0, 2, 2, 0] 578 579class HPI_SMLABB_T1(MinorFUTiming): 580 description = 'HPI_SMLABB_T1' 581 mask, match = t32_opcode('1111_1011_0001_xxxx__xxxx_xxxx_00xx_xxxx') 582 srcRegsRelativeLats = [0, 0, 0, 0, 0, 2, 0] 583 584class HPI_SMLABB_A1(MinorFUTiming): 585 description = 'HPI_SMLABB_A1' 586 mask, match = a32_opcode('xxxx_0001_0000_xxxx__xxxx_xxxx_1xx0_xxxx') 587 srcRegsRelativeLats = [0, 0, 0, 0, 0, 2, 0] 588 589class HPI_SMLAD_T1(MinorFUTiming): 590 description = 'HPI_SMLAD_T1' 591 mask, match = t32_opcode('1111_1011_0010_xxxx__xxxx_xxxx_000x_xxxx') 592 srcRegsRelativeLats = [0, 0, 0, 0, 0, 2, 0] 593 594class HPI_SMLAD_A1(MinorFUTiming): 595 description = 'HPI_SMLAD_A1' 596 mask, match = a32_opcode('xxxx_0111_0000_xxxx__xxxx_xxxx_00x1_xxxx') 597 # z, z, z, l, r, a 598 srcRegsRelativeLats = [0, 0, 0, 0, 0, 2, 0] 599 600class HPI_SMLAL_T1(MinorFUTiming): 601 description = 'HPI_SMLAL_T1' 602 mask, match = t32_opcode('1111_1011_1100_xxxx__xxxx_xxxx_0000_xxxx') 603class HPI_SMLAL_A1(MinorFUTiming): 604 description = 'HPI_SMLAL_A1' 605 mask, match = a32_opcode('xxxx_0000_111x_xxxx__xxxx_xxxx_1001_xxxx') 606 607class HPI_SMLALBB_T1(MinorFUTiming): 608 description = 'HPI_SMLALBB_T1' 609 mask, match = t32_opcode('1111_1011_1100_xxxx__xxxx_xxxx_10xx_xxxx') 610class HPI_SMLALBB_A1(MinorFUTiming): 611 description = 'HPI_SMLALBB_A1' 612 mask, match = a32_opcode('xxxx_0001_0100_xxxx__xxxx_xxxx_1xx0_xxxx') 613 614class HPI_SMLALD_T1(MinorFUTiming): 615 description = 'HPI_SMLALD_T1' 616 mask, match = t32_opcode('1111_1011_1100_xxxx__xxxx_xxxx_110x_xxxx') 617class HPI_SMLALD_A1(MinorFUTiming): 618 description = 'HPI_SMLALD_A1' 619 mask, match = a32_opcode('xxxx_0111_0100_xxxx__xxxx_xxxx_00x1_xxxx') 620 621class HPI_SMLAWB_T1(MinorFUTiming): 622 description = 'HPI_SMLAWB_T1' 623 mask, match = t32_opcode('1111_1011_0011_xxxx__xxxx_xxxx_000x_xxxx') 624 srcRegsRelativeLats = [0, 0, 0, 0, 0, 2, 0] 625 626class HPI_SMLAWB_A1(MinorFUTiming): 627 description = 'HPI_SMLAWB_A1' 628 mask, match = a32_opcode('xxxx_0001_0010_xxxx__xxxx_xxxx_1x00_xxxx') 629 srcRegsRelativeLats = [0, 0, 0, 0, 0, 2, 0] 630 631class HPI_SMLSD_A1(MinorFUTiming): 632 description = 'HPI_SMLSD_A1' 633 mask, match = a32_opcode('xxxx_0111_0000_xxxx__xxxx_xxxx_01x1_xxxx') 634 635class HPI_SMLSLD_T1(MinorFUTiming): 636 description = 'HPI_SMLSLD_T1' 637 mask, match = t32_opcode('1111_1011_1101_xxxx__xxxx_xxxx_110x_xxxx') 638class HPI_SMLSLD_A1(MinorFUTiming): 639 description = 'HPI_SMLSLD_A1' 640 mask, match = a32_opcode('xxxx_0111_0100_xxxx__xxxx_xxxx_01x1_xxxx') 641 642class HPI_SMMLA_T1(MinorFUTiming): 643 description = 'HPI_SMMLA_T1' 644 mask, match = t32_opcode('1111_1011_0101_xxxx__xxxx_xxxx_000x_xxxx') 645 # ^^^^ != 1111 646 srcRegsRelativeLats = [0, 0, 0, 2, 0, 0, 0] 647 648class HPI_SMMLA_A1(MinorFUTiming): 649 description = 'HPI_SMMLA_A1' 650 # Note that this must be after the encoding for SMMUL 651 mask, match = a32_opcode('xxxx_0111_0101_xxxx__xxxx_xxxx_00x1_xxxx') 652 # ^^^^ != 1111 653 srcRegsRelativeLats = [0, 0, 0, 2, 0, 0, 0] 654 655class HPI_SMMLS_T1(MinorFUTiming): 656 description = 'HPI_SMMLS_T1' 657 mask, match = t32_opcode('1111_1011_0110_xxxx__xxxx_xxxx_000x_xxxx') 658 srcRegsRelativeLats = [0, 0, 0, 2, 0, 0, 0] 659 660class HPI_SMMLS_A1(MinorFUTiming): 661 description = 'HPI_SMMLS_A1' 662 mask, match = a32_opcode('xxxx_0111_0101_xxxx__xxxx_xxxx_11x1_xxxx') 663 srcRegsRelativeLats = [0, 0, 0, 2, 0, 0, 0] 664 665class HPI_SMMUL_T1(MinorFUTiming): 666 description = 'HPI_SMMUL_T1' 667 mask, match = t32_opcode('1111_1011_0101_xxxx__1111_xxxx_000x_xxxx') 668 srcRegsRelativeLats = [0, 0, 0, 0, 0, 0] 669 670class HPI_SMMUL_A1(MinorFUTiming): 671 description = 'HPI_SMMUL_A1' 672 mask, match = a32_opcode('xxxx_0111_0101_xxxx__1111_xxxx_00x1_xxxx') 673 srcRegsRelativeLats = [0, 0, 0, 0, 0, 0] 674 675class HPI_SMUAD_T1(MinorFUTiming): 676 description = 'HPI_SMUAD_T1' 677 mask, match = t32_opcode('1111_1011_0010_xxxx__1111_xxxx_000x_xxxx') 678class HPI_SMUAD_A1(MinorFUTiming): 679 description = 'HPI_SMUAD_A1' 680 mask, match = a32_opcode('xxxx_0111_0000_xxxx__1111_xxxx_00x1_xxxx') 681 682class HPI_SMULBB_T1(MinorFUTiming): 683 description = 'HPI_SMULBB_T1' 684 mask, match = t32_opcode('1111_1011_0001_xxxx__1111_xxxx_00xx_xxxx') 685class HPI_SMULBB_A1(MinorFUTiming): 686 description = 'HPI_SMULBB_A1' 687 mask, match = a32_opcode('xxxx_0001_0110_xxxx__xxxx_xxxx_1xx0_xxxx') 688 689class HPI_SMULL_T1(MinorFUTiming): 690 description = 'HPI_SMULL_T1' 691 mask, match = t32_opcode('1111_1011_1000_xxxx__xxxx_xxxx_0000_xxxx') 692class HPI_SMULL_A1(MinorFUTiming): 693 description = 'HPI_SMULL_A1' 694 mask, match = a32_opcode('xxxx_0000_110x_xxxx__xxxx_xxxx_1001_xxxx') 695 696class HPI_SMULWB_T1(MinorFUTiming): 697 description = 'HPI_SMULWB_T1' 698 mask, match = t32_opcode('1111_1011_0011_xxxx__1111_xxxx_000x_xxxx') 699class HPI_SMULWB_A1(MinorFUTiming): 700 description = 'HPI_SMULWB_A1' 701 mask, match = a32_opcode('xxxx_0001_0010_xxxx__xxxx_xxxx_1x10_xxxx') 702 703class HPI_SMUSD_T1(MinorFUTiming): 704 description = 'HPI_SMUSD_T1' 705 mask, match = t32_opcode('1111_1011_0100_xxxx__1111_xxxx_000x_xxxx') 706class HPI_SMUSD_A1(MinorFUTiming): 707 description = 'HPI_SMUSD_A1' 708 mask, match = a32_opcode('xxxx_0111_0000_xxxx__1111_xxxx_01x1_xxxx') 709 710class HPI_SSAT_USAT_no_shift_A1(MinorFUTiming): 711 description = 'HPI_SSAT_USAT_no_shift_A1' 712 # Order before shift 713 mask, match = a32_opcode('xxxx_0110_1x1x_xxxx__xxxx_0000_0001_xxxx') 714 srcRegsRelativeLats = [0, 0, 0, 2, 0] 715 716class HPI_SSAT_USAT_shift_A1(MinorFUTiming): 717 description = 'HPI_SSAT_USAT_shift_A1' 718 # Order after shift 719 mask, match = a32_opcode('xxxx_0110_1x1x_xxxx__xxxx_xxxx_xx01_xxxx') 720 srcRegsRelativeLats = [0, 0, 0, 1, 0] 721 722class HPI_SSAT16_USAT16_A1(MinorFUTiming): 723 description = 'HPI_SSAT16_USAT16_A1' 724 mask, match = a32_opcode('xxxx_0110_1x10_xxxx__xxxx_xxxx_0011_xxxx') 725 srcRegsRelativeLats = [0, 0, 0, 2, 0] 726 727class HPI_SXTAB_T1(MinorFUTiming): 728 description = 'HPI_SXTAB_T1' 729 mask, match = t32_opcode('1111_1010_0100_xxxx__1111_xxxx_1xxx_xxxx') 730 srcRegsRelativeLats = [0, 0, 0, 1, 2, 0] 731 732class HPI_SXTAB_SXTAB16_SXTAH_UXTAB_UXTAB16_UXTAH_A1(MinorFUTiming): 733 description = 'HPI_SXTAB_SXTAB16_SXTAH_UXTAB_UXTAB16_UXTAH_A1' 734 # Place AFTER HPI_SXTB_SXTB16_SXTH_UXTB_UXTB16_UXTH_A1 735 # e6[9d][^f]0070 are undefined 736 mask, match = a32_opcode('xxxx_0110_1xxx_xxxx__xxxx_xxxx_0111_xxxx') 737 srcRegsRelativeLats = [0, 0, 0, 1, 2, 0] 738 739class HPI_SXTAB16_T1(MinorFUTiming): 740 description = 'HPI_SXTAB16_T1' 741 mask, match = t32_opcode('1111_1010_0010_xxxx__1111_xxxx_1xxx_xxxx') 742 srcRegsRelativeLats = [0, 0, 0, 1, 2, 0] 743 744class HPI_SXTAH_T1(MinorFUTiming): 745 description = 'HPI_SXTAH_T1' 746 mask, match = t32_opcode('1111_1010_0000_xxxx__1111_xxxx_1xxx_xxxx') 747 srcRegsRelativeLats = [0, 0, 0, 1, 2, 0] 748 749class HPI_SXTB_T1(MinorFUTiming): 750 description = 'HPI_SXTB_T1' 751 mask, match = t16_opcode('1011_0010_01xx_xxxx') 752class HPI_SXTB_T2(MinorFUTiming): 753 description = 'HPI_SXTB_T2' 754 mask, match = t32_opcode('1111_1010_0100_1111__1111_xxxx_1xxx_xxxx') 755 srcRegsRelativeLats = [0, 0, 0, 1, 2, 0] 756 757class HPI_SXTB_SXTB16_SXTH_UXTB_UXTB16_UXTH_A1(MinorFUTiming): 758 description = 'HPI_SXTB_SXTB16_SXTH_UXTB_UXTB16_UXTH_A1' 759 # e6[9d]f0070 are undefined 760 mask, match = a32_opcode('xxxx_0110_1xxx_1111__xxxx_xxxx_0111_xxxx') 761 srcRegsRelativeLats = [0, 0, 0, 2, 0] 762 763class HPI_SXTB16_T1(MinorFUTiming): 764 description = 'HPI_SXTB16_T1' 765 mask, match = t32_opcode('1111_1010_0010_1111__1111_xxxx_1xxx_xxxx') 766 srcRegsRelativeLats = [0, 0, 0, 1, 2, 0] 767 768class HPI_SXTH_T1(MinorFUTiming): 769 description = 'HPI_SXTH_T1' 770 mask, match = t16_opcode('1011_0010_00xx_xxxx') 771class HPI_SXTH_T2(MinorFUTiming): 772 description = 'HPI_SXTH_T2' 773 mask, match = t32_opcode('1111_1010_0000_1111__1111_xxxx_1xxx_xxxx') 774 srcRegsRelativeLats = [0, 0, 0, 1, 2, 0] 775 776class HPI_UDIV_T1(MinorFUTiming): 777 description = 'HPI_UDIV_T1' 778 mask, match = t32_opcode('1111_1011_1011_xxxx__xxxx_xxxx_1111_xxxx') 779 780udiv_lat_expr = expr_top(let([ 781 ('left', int_reg(src(4))), 782 ('right', int_reg(src(3))), 783 ('left_size', un('SizeInBits', ref('left'))), 784 ('right_size', un('SizeInBits', 785 bin('UDiv', ref('right'), literal(2)))), 786 ('left_minus_right', if_expr( 787 bin('SLessThan', ref('left_size'), ref('right_size')), 788 literal(0), 789 bin('Sub', ref('left_size'), ref('right_size')))) 790 ], 791 if_expr(bin('Equal', ref('right'), literal(0)), 792 literal(0), 793 bin('UDiv', ref('left_minus_right'), literal(4))) 794 )) 795 796class HPI_UDIV_A1(MinorFUTiming): 797 description = 'HPI_UDIV_A1' 798 mask, match = a32_opcode('xxxx_0111_0011_xxxx__xxxx_xxxx_0001_xxxx') 799 extraCommitLat = 0 800 srcRegsRelativeLats = [] 801 extraCommitLatExpr = udiv_lat_expr 802 803class HPI_UMAAL_T1(MinorFUTiming): 804 description = 'HPI_UMAAL_T1' 805 mask, match = t32_opcode('1111_1011_1110_xxxx__xxxx_xxxx_0110_xxxx') 806 # z, z, z, dlo, dhi, l, r 807 extraCommitLat = 1 808 srcRegsRelativeLats = [0, 0, 0, 0, 0, 0, 0, 0] 809 810class HPI_UMAAL_A1(MinorFUTiming): 811 description = 'HPI_UMAAL_A1' 812 mask, match = a32_opcode('xxxx_0000_0100_xxxx__xxxx_xxxx_1001_xxxx') 813 # z, z, z, dlo, dhi, l, r 814 extraCommitLat = 1 815 srcRegsRelativeLats = [0, 0, 0, 0, 0, 0, 0, 0] 816 817class HPI_UMLAL_T1(MinorFUTiming): 818 description = 'HPI_UMLAL_T1' 819 mask, match = t32_opcode('1111_1011_1110_xxxx__xxxx_xxxx_0000_xxxx') 820 821class HPI_UMLAL_A1(MinorFUTiming): 822 description = 'HPI_UMLAL_A1' 823 mask, match = t32_opcode('xxxx_0000_101x_xxxx__xxxx_xxxx_1001_xxxx') 824 825class HPI_UMULL_T1(MinorFUTiming): 826 description = 'HPI_UMULL_T1' 827 mask, match = t32_opcode('1111_1011_1010_xxxx__xxxx_xxxx_0000_xxxx') 828 829class HPI_UMULL_A1(MinorFUTiming): 830 description = 'HPI_UMULL_A1' 831 mask, match = a32_opcode('xxxx_0000_100x_xxxx__xxxx_xxxx_1001_xxxx') 832 833class HPI_USAD8_USADA8_A1(MinorFUTiming): 834 description = 'HPI_USAD8_USADA8_A1' 835 mask, match = a32_opcode('xxxx_0111_1000_xxxx__xxxx_xxxx_0001_xxxx') 836 srcRegsRelativeLats = [0, 0, 0, 0, 0, 2, 0] 837 838class HPI_USAD8_USADA8_A1_Suppress(MinorFUTiming): 839 description = 'HPI_USAD8_USADA8_A1_Suppress' 840 mask, match = a32_opcode('xxxx_0111_1000_xxxx__xxxx_xxxx_0001_xxxx') 841 srcRegsRelativeLats = [] 842 suppress = True 843 844class HPI_VMOV_immediate_A1(MinorFUTiming): 845 description = 'HPI_VMOV_register_A1' 846 mask, match = a32_opcode('1111_0010_0x10_xxxx_xxxx_0001_xxx1_xxxx') 847 # cpsr, z, z, z, hcptr, nsacr, cpacr, fpexc, scr 848 srcRegsRelativeLats = [5, 5, 5, 5, 5, 5, 5, 5, 5, 0] 849 850class HPI_VMRS_A1(MinorFUTiming): 851 description = 'HPI_VMRS_A1' 852 mask, match = a32_opcode('xxxx_1110_1111_0001_xxxx_1010_xxx1_xxxx') 853 # cpsr,z,z,z,hcptr,nsacr,cpacr,scr,r42 854 srcRegsRelativeLats = [5, 5, 5, 5, 5, 5, 5, 5, 5, 0] 855 856class HPI_VMOV_register_A2(MinorFUTiming): 857 description = 'HPI_VMOV_register_A2' 858 mask, match = a32_opcode('xxxx_1110_1x11_0000_xxxx_101x_01x0_xxxx') 859 # cpsr, z, r39, z, hcptr, nsacr, cpacr, fpexc, scr, f4, f5, f0, f1 860 srcRegsRelativeLats = \ 861 [5, 5, 5, 5, 5, 5, 5, 5, 5, 4, 4, 4, 4, 4, 4, 4, 4, 0] 862 863# VADD.I16 D/VADD.F32 D/VADD.I8 D/VADD.I32 D 864class HPI_VADD2H_A32(MinorFUTiming): 865 description = 'Vadd2hALU' 866 mask, match = a32_opcode('1111_0010_0xxx_xxxx__xxxx_1000_xxx0_xxxx') 867 # cpsr, z, z, z, cpacr, fpexc, l0, r0, l1, r1, l2, r2, l3, r3 (for vadd2h) 868 srcRegsRelativeLats = [0, 0, 0, 0, 0, 0, 4, 4, 4, 4, 4, 4, 4, 4, 0] 869 870# VAQQHN.I16 Q/VAQQHN.I32 Q/VAQQHN.I64 Q 871class HPI_VADDHN_A32(MinorFUTiming): 872 description = 'VaddhnALU' 873 mask, match = a32_opcode('1111_0010_1xxx_xxxx__xxxx_0100_x0x0_xxxx') 874 # cpsr, z, z, z, cpacr, fpexc, l0, l1, l2, l3, r0, r1, r2, r3 875 srcRegsRelativeLats = [0, 0, 0, 0, 0, 0, 3, 3, 3, 3, 3, 3, 3, 3, 0] 876 877class HPI_VADDL_A32(MinorFUTiming): 878 description = 'VaddlALU' 879 mask, match = a32_opcode('1111_001x_1xxx_xxxx__xxxx_0000_x0x0_xxxx') 880 srcRegsRelativeLats = [0, 0, 0, 0, 0, 0, 3, 3, 3, 3, 3, 3, 3, 3, 0] 881 882class HPI_VADDW_A32(MinorFUTiming): 883 description = 'HPI_VADDW_A32' 884 mask, match = a32_opcode('1111_001x_1xxx_xxxx__xxxx_0001_x0x0_xxxx') 885 # cpsr, z, z, z, cpacr, fpexc, l0, l1, l2, l3, r0, r1 886 srcRegsRelativeLats = [0, 0, 0, 0, 0, 0, 4, 4, 4, 4, 3, 3, 0] 887 888# VHADD/VHSUB S8,S16,S32,U8,U16,U32 Q and D 889class HPI_VHADD_A32(MinorFUTiming): 890 description = 'HPI_VHADD_A32' 891 mask, match = a32_opcode('1111_001x_0xxx_xxxx__xxxx_00x0_xxx0_xxxx') 892 srcRegsRelativeLats = [0, 0, 0, 0, 0, 0, 4, 4, 4, 4, 4, 4, 4, 4, 0] 893 894class HPI_VPADAL_A32(MinorFUTiming): 895 description = 'VpadalALU' 896 mask, match = a32_opcode('1111_0011_1x11_xx00__xxxx_0110_xxx0_xxxx') 897 srcRegsRelativeLats = [0, 0, 0, 0, 0, 0, 2, 2, 2, 2, 2, 2, 2, 2, 0] 898 899# VPADDH.I16 900class HPI_VPADDH_A32(MinorFUTiming): 901 description = 'VpaddhALU' 902 mask, match = a32_opcode('1111_0010_0xxx_xxxx__xxxx_1011_xxx1_xxxx') 903 srcRegsRelativeLats = [0, 0, 0, 0, 0, 0, 3, 3, 3, 3, 3, 3, 3, 3, 0] 904 905# VPADDH.F32 906class HPI_VPADDS_A32(MinorFUTiming): 907 description = 'VpaddsALU' 908 mask, match = a32_opcode('1111_0011_0x0x_xxxx__xxxx_1101_xxx0_xxxx') 909 srcRegsRelativeLats = [0, 0, 0, 0, 0, 0, 2, 2, 2, 2, 2, 2, 2, 2, 0] 910 911# VPADDL.S16 912class HPI_VPADDL_A32(MinorFUTiming): 913 description = 'VpaddlALU' 914 mask, match = a32_opcode('1111_0011_1x11_xx00__xxxx_0010_xxx0_xxxx') 915 srcRegsRelativeLats = [0, 0, 0, 0, 0, 0, 3, 3, 3, 3, 3, 3, 3, 3, 0] 916 917# VRADDHN.I16 918class HPI_VRADDHN_A32(MinorFUTiming): 919 description = 'HPI_VRADDHN_A32' 920 mask, match = a32_opcode('1111_0011_1xxx_xxxx__xxxx_0100_x0x0_xxxx') 921 # cpsr, z, z, z, cpacr, fpexc, l0, l1, l2, l3, r0, r1, r2, r3 922 srcRegsRelativeLats = [0, 0, 0, 0, 0, 0, 4, 4, 4, 4, 4, 4, 4, 4, 0] 923 924class HPI_VRHADD_A32(MinorFUTiming): 925 description = 'VrhaddALU' 926 mask, match = a32_opcode('1111_001x_0xxx_xxxx__xxxx_0001_xxx0_xxxx') 927 srcRegsRelativeLats = [0, 0, 0, 0, 0, 0, 4, 4, 4, 4, 4, 4, 4, 4, 0] 928 929class HPI_VQADD_A32(MinorFUTiming): 930 description = 'VqaddALU' 931 mask, match = a32_opcode('1111_001x_0xxx_xxxx__xxxx_0000_xxx1_xxxx') 932 srcRegsRelativeLats = [0, 0, 0, 0, 0, 0, 3, 3, 3, 3, 3, 3, 3, 3, 0] 933 934class HPI_VANDQ_A32(MinorFUTiming): 935 description = 'VandqALU' 936 mask, match = a32_opcode('1111_0010_0x00_xxxx__xxxx_0001_xxx1_xxxx') 937 srcRegsRelativeLats = [0, 0, 0, 0, 0, 0, 5, 5, 5, 5, 5, 5, 5, 5, 0] 938 939# VMUL (integer) 940class HPI_VMULI_A32(MinorFUTiming): 941 description = 'VmuliALU' 942 mask, match = a32_opcode('1111_001x_0xxx_xxxx__xxxx_1001_xxx1_xxxx') 943 srcRegsRelativeLats = [0, 0, 0, 0, 0, 0, 2, 2, 2, 2, 2, 2, 2, 2, 0] 944 945# VBIC (reg) 946class HPI_VBIC_A32(MinorFUTiming): 947 description = 'VbicALU' 948 mask, match = a32_opcode('1111_0010_0x01_xxxx__xxxx_0001_xxx1_xxxx') 949 srcRegsRelativeLats = [0, 0, 0, 0, 0, 0, 5, 5, 5, 5, 5, 5, 5, 5, 0] 950 951# VBIF VBIT VBSL 952class HPI_VBIF_ETC_A32(MinorFUTiming): 953 description = 'VbifALU' 954 mask, match = a32_opcode('1111_0011_0xxx_xxxx__xxxx_0001_xxx1_xxxx') 955 srcRegsRelativeLats = \ 956 [0, 0, 0, 0, 0, 0, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 0] 957 958class HPI_VACGE_A32(MinorFUTiming): 959 description = 'VacgeALU' 960 mask, match = a32_opcode('1111_0011_0xxx_xxxx__xxxx_1110_xxx1_xxxx') 961 srcRegsRelativeLats = [0, 0, 0, 0, 0, 0, 4, 4, 4, 4, 4, 4, 4, 4, 4, 0] 962 963# VCEQ.F32 964class HPI_VCEQ_A32(MinorFUTiming): 965 description = 'VceqALU' 966 mask, match = a32_opcode('1111_0010_0x0x_xxxx__xxxx_1110_xxx0_xxxx') 967 srcRegsRelativeLats = [0, 0, 0, 0, 0, 0, 4, 4, 4, 4, 4, 4, 4, 4, 4, 0] 968 969# VCEQ.[IS]... register 970class HPI_VCEQI_A32(MinorFUTiming): 971 description = 'VceqiALU' 972 mask, match = a32_opcode('1111_0011_0xxx_xxxx__xxxx_1000_xxx1_xxxx') 973 srcRegsRelativeLats = [0, 0, 0, 0, 0, 0, 4, 4, 4, 4, 4, 4, 4, 4, 4, 0] 974 975# VCEQ.[IS]... immediate 976class HPI_VCEQII_A32(MinorFUTiming): 977 description = 'HPI_VCEQII_A32' 978 mask, match = a32_opcode('1111_0011_1x11_xx01__xxxx_0x01_0xx0_xxxx') 979 srcRegsRelativeLats = [0, 0, 0, 0, 0, 0, 4, 4, 4, 4, 4, 4, 4, 4, 4, 0] 980 981class HPI_VTST_A32(MinorFUTiming): 982 description = 'HPI_VTST_A32' 983 mask, match = a32_opcode('1111_0010_0xxx_xxxx__xxxx_1000_xxx1_xxxx') 984 srcRegsRelativeLats = [0, 0, 0, 0, 0, 0, 3, 3, 3, 3, 3, 3, 3, 3, 3, 0] 985 986class HPI_VCLZ_A32(MinorFUTiming): 987 description = 'HPI_VCLZ_A32' 988 mask, match = a32_opcode('1111_0011_1x11_xx00__xxxx_0100_1xx0_xxxx') 989 srcRegsRelativeLats = [0, 0, 0, 0, 0, 0, 4, 4, 4, 4, 4, 4, 4, 4, 4, 0] 990 991class HPI_VCNT_A32(MinorFUTiming): 992 description = 'HPI_VCNT_A32' 993 mask, match = a32_opcode('1111_0011_1x11_xx00__xxxx_0101_0xx0_xxxx') 994 srcRegsRelativeLats = [0, 0, 0, 0, 0, 0, 4, 4, 4, 4, 4, 4, 4, 4, 4, 0] 995 996class HPI_VEXT_A32(MinorFUTiming): 997 description = 'HPI_VCNT_A32' 998 mask, match = a32_opcode('1111_0010_1x11_xxxx__xxxx_xxxx_xxx0_xxxx') 999 srcRegsRelativeLats = [0, 0, 0, 0, 0, 0, 4, 4, 4, 4, 4, 4, 4, 4, 4, 0] 1000 1001# VMAX VMIN integer 1002class HPI_VMAXI_A32(MinorFUTiming): 1003 description = 'HPI_VMAXI_A32' 1004 mask, match = a32_opcode('1111_001x_0xxx_xxxx__xxxx_0110_xxxx_xxxx') 1005 srcRegsRelativeLats = [0, 0, 0, 0, 0, 0, 4, 4, 4, 4, 4, 4, 4, 4, 4, 0] 1006 1007# VMAX VMIN float 1008class HPI_VMAXS_A32(MinorFUTiming): 1009 description = 'HPI_VMAXS_A32' 1010 mask, match = a32_opcode('1111_0010_0xxx_xxxx__xxxx_1111_xxx0_xxxx') 1011 srcRegsRelativeLats = [0, 0, 0, 0, 0, 0, 2, 2, 2, 2, 2, 2, 2, 2, 2, 0] 1012 1013# VNEG integer 1014class HPI_VNEGI_A32(MinorFUTiming): 1015 description = 'HPI_VNEGI_A32' 1016 mask, match = a32_opcode('1111_0011_1x11_xx01__xxxx_0x11_1xx0_xxxx') 1017 srcRegsRelativeLats = [0, 0, 0, 0, 0, 0, 4, 4, 4, 4, 4, 4, 4, 4, 4, 0] 1018 1019# VNEG float 1020class HPI_VNEGF_A32(MinorFUTiming): 1021 description = 'HPI_VNEGF_A32' 1022 mask, match = a32_opcode('xxxx_1110_1x11_0001__xxxx_101x_01x0_xxxx') 1023 srcRegsRelativeLats = [0, 0, 0, 0, 0, 0, 2, 2, 2, 2, 2, 2, 2, 2, 2, 0] 1024 1025# VREV16 VREV32 VREV64 1026class HPI_VREVN_A32(MinorFUTiming): 1027 description = 'HPI_VREVN_A32' 1028 mask, match = a32_opcode('1111_0011_1x11_xx00__xxxx_000x_xxx0_xxxx') 1029 srcRegsRelativeLats = [0, 0, 0, 0, 0, 0, 4, 4, 4, 4, 4, 4, 4, 4, 4, 0] 1030 1031class HPI_VQNEG_A32(MinorFUTiming): 1032 description = 'HPI_VQNEG_A32' 1033 mask, match = a32_opcode('1111_0011_1x11_xx00__xxxx_0111_1xx0_xxxx') 1034 srcRegsRelativeLats = [0, 0, 0, 0, 0, 0, 3, 3, 3, 3, 3, 3, 3, 3, 3, 0] 1035 1036class HPI_VSWP_A32(MinorFUTiming): 1037 description = 'HPI_VSWP_A32' 1038 mask, match = a32_opcode('1111_0011_1x11_xx10__xxxx_0000_0xx0_xxxx') 1039 srcRegsRelativeLats = [0, 0, 0, 0, 0, 0, 4, 4, 4, 4, 4, 4, 4, 4, 0] 1040 1041class HPI_VTRN_A32(MinorFUTiming): 1042 description = 'HPI_VTRN_A32' 1043 mask, match = a32_opcode('1111_0011_1x11_xx10__xxxx_0000_1xx0_xxxx') 1044 # cpsr, z, z, z, cpact, fpexc, o0, d0, o1, d1, o2, d2, o3, d3 1045 srcRegsRelativeLats = [0, 0, 0, 0, 0, 0, 2, 2, 2, 2, 2, 2, 2, 2, 0] 1046 1047# VQMOVN VQMOVUN 1048class HPI_VQMOVN_A32(MinorFUTiming): 1049 description = 'HPI_VQMOVN_A32' 1050 mask, match = a32_opcode('1111_0011_1x11_xx10__xxxx_0010_xxx0_xxxx') 1051 # cpsr, z, z, z, cpact, fpexc, o[0], o[1], o[2], o[3], fpscr 1052 srcRegsRelativeLats = [0, 0, 0, 0, 0, 0, 2, 2, 2, 2, 2, 0] 1053 1054# VUZP double word 1055class HPI_VUZP_A32(MinorFUTiming): 1056 description = 'HPI_VUZP_A32' 1057 mask, match = a32_opcode('1111_0011_1x11_xx10__xxxx_0001_00x0_xxxx') 1058 srcRegsRelativeLats = [0, 0, 0, 0, 0, 0, 3, 3, 3, 3, 3, 3, 3, 3, 0] 1059 1060# VDIV.F32 1061class HPI_VDIV32_A32(MinorFUTiming): 1062 description = 'HPI_VDIV32_A32' 1063 mask, match = a32_opcode('xxxx_1110_1x00_xxxx__xxxx_1010_x0x0_xxxx') 1064 # cpsr, z, z, z, cpact, fpexc, fpscr_exc, l, r 1065 extraCommitLat = 9 1066 srcRegsRelativeLats = [0, 0, 0, 0, 0, 0, 20, 4, 4, 0] 1067 1068# VDIV.F64 1069class HPI_VDIV64_A32(MinorFUTiming): 1070 description = 'HPI_VDIV64_A32' 1071 mask, match = a32_opcode('xxxx_1110_1x00_xxxx__xxxx_1011_x0x0_xxxx') 1072 # cpsr, z, z, z, cpact, fpexc, fpscr_exc, l, r 1073 extraCommitLat = 18 1074 srcRegsRelativeLats = [0, 0, 0, 0, 0, 0, 20, 4, 4, 0] 1075 1076class HPI_VZIP_A32(MinorFUTiming): 1077 description = 'HPI_VZIP_A32' 1078 mask, match = a32_opcode('1111_0011_1x11_xx10__xxxx_0001_1xx0_xxxx') 1079 srcRegsRelativeLats = [0, 0, 0, 0, 0, 0, 4, 4, 4, 4, 4, 4, 4, 4, 0] 1080 1081# VPMAX integer 1082class HPI_VPMAX_A32(MinorFUTiming): 1083 description = 'HPI_VPMAX_A32' 1084 mask, match = a32_opcode('1111_001x_0xxx_xxxx__xxxx_1010_xxxx_xxxx') 1085 # cpsr, z, z, z, cpact, fpexc, l0, r0, l1, r1, fpscr 1086 srcRegsRelativeLats = [0, 0, 0, 0, 0, 0, 4, 4, 4, 4, 4, 0] 1087 1088# VPMAX float 1089class HPI_VPMAXF_A32(MinorFUTiming): 1090 description = 'HPI_VPMAXF_A32' 1091 mask, match = a32_opcode('1111_0011_0xxx_xxxx__xxxx_1111_xxx0_xxxx') 1092 srcRegsRelativeLats = [0, 0, 0, 0, 0, 0, 2, 2, 2, 2, 0] 1093 1094class HPI_VMOVN_A32(MinorFUTiming): 1095 description = 'HPI_VMOVN_A32' 1096 mask, match = a32_opcode('1111_0011_1x11_xx10__xxxx_0010_00x0_xxxx') 1097 srcRegsRelativeLats = [0, 0, 0, 0, 0, 0, 4, 4, 4, 4, 0] 1098 1099class HPI_VMOVL_A32(MinorFUTiming): 1100 description = 'HPI_VMOVL_A32' 1101 mask, match = a32_opcode('1111_001x_1xxx_x000__xxxx_1010_00x1_xxxx') 1102 srcRegsRelativeLats = [0, 0, 0, 0, 0, 0, 4, 4, 4, 4, 0] 1103 1104# VSQRT.F64 1105class HPI_VSQRT64_A32(MinorFUTiming): 1106 description = 'HPI_VSQRT64_A32' 1107 mask, match = a32_opcode('xxxx_1110_1x11_0001__xxxx_1011_11x0_xxxx') 1108 extraCommitLat = 18 1109 srcRegsRelativeLats = [] 1110 1111# VSQRT.F32 1112class HPI_VSQRT32_A32(MinorFUTiming): 1113 description = 'HPI_VSQRT32_A32' 1114 mask, match = a32_opcode('xxxx_1110_1x11_0001__xxxx_1010_11x0_xxxx') 1115 extraCommitLat = 9 1116 srcRegsRelativeLats = [] 1117 1118class HPI_FloatSimdFU(MinorFU): 1119 opClasses = minorMakeOpClassSet([ 1120 'FloatAdd', 'FloatCmp', 'FloatCvt', 'FloatMult', 'FloatDiv', 1121 'FloatSqrt', 'FloatMisc', 'FloatMultAcc', 1122 'SimdAdd', 'SimdAddAcc', 'SimdAlu', 'SimdCmp', 'SimdCvt', 1123 'SimdMisc', 'SimdMult', 'SimdMultAcc', 'SimdShift', 'SimdShiftAcc', 1124 'SimdSqrt', 'SimdFloatAdd', 'SimdFloatAlu', 'SimdFloatCmp', 1125 'SimdFloatCvt', 'SimdFloatDiv', 'SimdFloatMisc', 'SimdFloatMult', 1126 'SimdFloatMultAcc', 'SimdFloatSqrt']) 1127 1128 timings = [ 1129 # VUZP and VZIP must be before VADDW/L 1130 HPI_VUZP_A32(), HPI_VZIP_A32(), 1131 HPI_VADD2H_A32(), HPI_VADDHN_A32(), 1132 HPI_VADDL_A32(), HPI_VADDW_A32(), 1133 HPI_VHADD_A32(), HPI_VPADAL_A32(), 1134 HPI_VPADDH_A32(), HPI_VPADDS_A32(), 1135 HPI_VPADDL_A32(), HPI_VRADDHN_A32(), 1136 HPI_VRHADD_A32(), HPI_VQADD_A32(), 1137 HPI_VANDQ_A32(), HPI_VBIC_A32(), 1138 HPI_VBIF_ETC_A32(), HPI_VACGE_A32(), 1139 HPI_VCEQ_A32(), HPI_VCEQI_A32(), 1140 HPI_VCEQII_A32(), HPI_VTST_A32(), 1141 HPI_VCLZ_A32(), HPI_VCNT_A32(), 1142 HPI_VEXT_A32(), HPI_VMAXI_A32(), 1143 HPI_VMAXS_A32(), HPI_VNEGI_A32(), 1144 HPI_VNEGF_A32(), HPI_VREVN_A32(), 1145 HPI_VQNEG_A32(), HPI_VSWP_A32(), 1146 HPI_VTRN_A32(), HPI_VPMAX_A32(), 1147 HPI_VPMAXF_A32(), HPI_VMOVN_A32(), 1148 HPI_VMRS_A1(), 1149 HPI_VMOV_immediate_A1(), 1150 HPI_VMOV_register_A2(), 1151 HPI_VQMOVN_A32(), HPI_VMOVL_A32(), 1152 HPI_VDIV32_A32(), HPI_VDIV64_A32(), 1153 HPI_VSQRT32_A32(), HPI_VSQRT64_A32(), 1154 HPI_VMULI_A32(), 1155 # Add before here 1156 HPI_FMADD_A64(), 1157 HPI_FMSUB_D_A64(), 1158 HPI_FMOV_A64(), 1159 HPI_ADD_SUB_vector_scalar_A64(), 1160 HPI_ADD_SUB_vector_vector_A64(), 1161 HPI_FDIV_scalar_32_A64(), HPI_FDIV_scalar_64_A64(), 1162 HPI_DefaultA64Vfp(), 1163 HPI_DefaultVfp()] 1164 1165 opLat = 6 1166 1167class HPI_IntFU(MinorFU): 1168 opClasses = minorMakeOpClassSet(['IntAlu']) 1169 # IMPORTANT! Keep the order below, add new entries at the head 1170 timings = [ 1171 HPI_SSAT_USAT_no_shift_A1(), 1172 HPI_SSAT_USAT_shift_A1(), 1173 HPI_SSAT16_USAT16_A1(), 1174 HPI_QADD_QSUB_A1(), 1175 HPI_QADD_QSUB_T1(), 1176 HPI_QADD_ETC_A1(), 1177 HPI_QASX_QSAX_UQASX_UQSAX_T1(), 1178 HPI_QADD_ETC_T1(), 1179 HPI_USUB_ETC_A1(), 1180 HPI_ADD_ETC_A1(), 1181 HPI_ADD_ETC_T1(), 1182 HPI_QDADD_QDSUB_A1(), 1183 HPI_QDADD_QDSUB_T1(), 1184 HPI_SASX_SHASX_UASX_UHASX_A1(), 1185 HPI_SHSAX_SSAX_UHSAX_USAX_A1(), 1186 HPI_SXTB_SXTB16_SXTH_UXTB_UXTB16_UXTH_A1(), 1187 1188 # Must be after HPI_SXTB_SXTB16_SXTH_UXTB_UXTB16_UXTH_A1 1189 HPI_SXTAB_SXTAB16_SXTAH_UXTAB_UXTAB16_UXTAH_A1(), 1190 1191 HPI_SXTAB_T1(), 1192 HPI_SXTAB16_T1(), 1193 HPI_SXTAH_T1(), 1194 HPI_SXTB_T2(), 1195 HPI_SXTB16_T1(), 1196 HPI_SXTH_T2(), 1197 1198 HPI_PKH_A1(), 1199 HPI_PKH_T1(), 1200 HPI_SBFX_UBFX_A1(), 1201 HPI_SEL_A1(), 1202 HPI_RBIT_A1(), 1203 HPI_REV_REV16_A1(), 1204 HPI_REVSH_A1(), 1205 HPI_USAD8_USADA8_A1(), 1206 HPI_BFI_A1(), 1207 HPI_BFI_T1(), 1208 1209 HPI_CMN_register_A1(), 1210 HPI_CMN_immediate_A1(), 1211 HPI_CMP_register_A1(), 1212 HPI_CMP_immediate_A1(), 1213 1214 HPI_DataProcessingNoShift(), 1215 HPI_DataProcessingMovShiftr(), 1216 HPI_DataProcessingMayShift(), 1217 1218 HPI_Cxxx_A64(), 1219 1220 HPI_DefaultA64Int(), 1221 HPI_DefaultInt()] 1222 opLat = 3 1223 1224class HPI_Int2FU(MinorFU): 1225 opClasses = minorMakeOpClassSet(['IntAlu']) 1226 # IMPORTANT! Keep the order below, add new entries at the head 1227 timings = [ 1228 HPI_SSAT_USAT_no_shift_A1(), 1229 HPI_SSAT_USAT_shift_A1(), 1230 HPI_SSAT16_USAT16_A1(), 1231 HPI_QADD_QSUB_A1(), 1232 HPI_QADD_QSUB_T1(), 1233 HPI_QADD_ETC_A1(), 1234 HPI_QASX_QSAX_UQASX_UQSAX_T1(), 1235 HPI_QADD_ETC_T1(), 1236 HPI_USUB_ETC_A1(), 1237 HPI_ADD_ETC_A1(), 1238 HPI_ADD_ETC_T1(), 1239 HPI_QDADD_QDSUB_A1(), 1240 HPI_QDADD_QDSUB_T1(), 1241 HPI_SASX_SHASX_UASX_UHASX_A1(), 1242 HPI_SHSAX_SSAX_UHSAX_USAX_A1(), 1243 HPI_SXTB_SXTB16_SXTH_UXTB_UXTB16_UXTH_A1(), 1244 1245 # Must be after HPI_SXTB_SXTB16_SXTH_UXTB_UXTB16_UXTH_A1 1246 HPI_SXTAB_SXTAB16_SXTAH_UXTAB_UXTAB16_UXTAH_A1(), 1247 1248 HPI_SXTAB_T1(), 1249 HPI_SXTAB16_T1(), 1250 HPI_SXTAH_T1(), 1251 HPI_SXTB_T2(), 1252 HPI_SXTB16_T1(), 1253 HPI_SXTH_T2(), 1254 1255 HPI_PKH_A1(), 1256 HPI_PKH_T1(), 1257 HPI_SBFX_UBFX_A1(), 1258 HPI_SEL_A1_Suppress(), 1259 HPI_RBIT_A1(), 1260 HPI_REV_REV16_A1(), 1261 HPI_REVSH_A1(), 1262 HPI_USAD8_USADA8_A1_Suppress(), 1263 HPI_BFI_A1(), 1264 HPI_BFI_T1(), 1265 1266 HPI_CMN_register_A1(), # Need to check for shift 1267 HPI_CMN_immediate_A1(), 1268 HPI_CMP_register_A1(), # Need to check for shift 1269 HPI_CMP_immediate_A1(), 1270 1271 HPI_DataProcessingNoShift(), 1272 HPI_DataProcessingAllowShifti(), 1273 # HPI_DataProcessingAllowMovShiftr(), 1274 1275 # Data processing ops that match SuppressShift but are not 1276 # to be suppressed here 1277 HPI_CLZ_A1(), 1278 HPI_CLZ_T1(), 1279 HPI_DataProcessingSuppressShift(), 1280 # Can you dual issue a branch? 1281 # HPI_DataProcessingSuppressBranch(), 1282 HPI_Cxxx_A64(), 1283 1284 HPI_DefaultA64Int(), 1285 HPI_DefaultInt()] 1286 opLat = 3 1287 1288class HPI_IntMulFU(MinorFU): 1289 opClasses = minorMakeOpClassSet(['IntMult']) 1290 timings = [ 1291 HPI_MLA_A1(), HPI_MLA_T1(), 1292 HPI_MLS_A1(), HPI_MLS_T1(), 1293 HPI_SMLABB_A1(), HPI_SMLABB_T1(), 1294 HPI_SMLAWB_A1(), HPI_SMLAWB_T1(), 1295 HPI_SMLAD_A1(), HPI_SMLAD_T1(), 1296 HPI_SMMUL_A1(), HPI_SMMUL_T1(), 1297 # SMMUL_A1 must be before SMMLA_A1 1298 HPI_SMMLA_A1(), HPI_SMMLA_T1(), 1299 HPI_SMMLS_A1(), HPI_SMMLS_T1(), 1300 HPI_UMAAL_A1(), HPI_UMAAL_T1(), 1301 1302 HPI_MADD_A64(), 1303 HPI_DefaultA64Mul(), 1304 HPI_DefaultMul()] 1305 opLat = 3 1306 cantForwardFromFUIndices = [0, 1, 5] # Int1, Int2, Mem 1307 1308class HPI_IntDivFU(MinorFU): 1309 opClasses = minorMakeOpClassSet(['IntDiv']) 1310 timings = [HPI_SDIV_A1(), HPI_UDIV_A1(), 1311 HPI_SDIV_A64()] 1312 issueLat = 3 1313 opLat = 3 1314 1315class HPI_MemFU(MinorFU): 1316 opClasses = minorMakeOpClassSet(['MemRead', 'MemWrite', 'FloatMemRead', 1317 'FloatMemWrite']) 1318 timings = [HPI_DefaultMem(), HPI_DefaultMem64()] 1319 opLat = 1 1320 cantForwardFromFUIndices = [5] # Mem (this FU) 1321 1322class HPI_MiscFU(MinorFU): 1323 opClasses = minorMakeOpClassSet(['IprAccess', 'InstPrefetch']) 1324 opLat = 1 1325 1326class HPI_FUPool(MinorFUPool): 1327 funcUnits = [HPI_IntFU(), # 0 1328 HPI_Int2FU(), # 1 1329 HPI_IntMulFU(), # 2 1330 HPI_IntDivFU(), # 3 1331 HPI_FloatSimdFU(), # 4 1332 HPI_MemFU(), # 5 1333 HPI_MiscFU() # 6 1334 ] 1335 1336class HPI_DTB(ArmTLB): 1337 size = 256 1338 1339class HPI_ITB(ArmTLB): 1340 size = 256 1341 1342class HPI_WalkCache(Cache): 1343 data_latency = 4 1344 tag_latency = 4 1345 response_latency = 4 1346 mshrs = 6 1347 tgts_per_mshr = 8 1348 size = '1kB' 1349 assoc = 8 1350 write_buffers = 16 1351 1352class HPI_BP(TournamentBP): 1353 localPredictorSize = 64 1354 localCtrBits = 2 1355 localHistoryTableSize = 64 1356 globalPredictorSize = 1024 1357 globalCtrBits = 2 1358 choicePredictorSize = 1024 1359 choiceCtrBits = 2 1360 BTBEntries = 128 1361 BTBTagSize = 18 1362 RASSize = 8 1363 instShiftAmt = 2 1364 1365class HPI_ICache(Cache): 1366 data_latency = 1 1367 tag_latency = 1 1368 response_latency = 1 1369 mshrs = 2 1370 tgts_per_mshr = 8 1371 size = '32kB' 1372 assoc = 2 1373 # No prefetcher, this is handled by the core 1374 1375class HPI_DCache(Cache): 1376 data_latency = 1 1377 tag_latency = 1 1378 response_latency = 1 1379 mshrs = 4 1380 tgts_per_mshr = 8 1381 size = '32kB' 1382 assoc = 4 1383 write_buffers = 4 1384 prefetcher = StridePrefetcher( 1385 queue_size=4, 1386 degree=4) 1387 1388class HPI_L2(Cache): 1389 data_latency = 13 1390 tag_latency = 13 1391 response_latency = 5 1392 mshrs = 4 1393 tgts_per_mshr = 8 1394 size = '1024kB' 1395 assoc = 16 1396 write_buffers = 16 1397 # prefetcher FIXME 1398 1399class HPI(MinorCPU): 1400 # Inherit the doc string from the module to avoid repeating it 1401 # here. 1402 __doc__ = __doc__ 1403 1404 fetch1LineSnapWidth = 0 1405 fetch1LineWidth = 0 1406 fetch1FetchLimit = 1 1407 fetch1ToFetch2ForwardDelay = 1 1408 fetch1ToFetch2BackwardDelay = 1 1409 1410 fetch2InputBufferSize = 2 1411 fetch2ToDecodeForwardDelay = 1 1412 fetch2CycleInput = True 1413 1414 decodeInputBufferSize = 3 1415 decodeToExecuteForwardDelay = 1 1416 decodeInputWidth = 2 1417 decodeCycleInput = True 1418 1419 executeInputWidth = 2 1420 executeCycleInput = True 1421 executeIssueLimit = 2 1422 1423 # Only allow one ld/st to be issued but the second ld/st FU allows 1424 # back-to-back loads 1425 executeMemoryIssueLimit = 1 1426 1427 executeCommitLimit = 2 1428 executeMemoryCommitLimit = 1 1429 executeInputBufferSize = 7 1430 1431 executeMaxAccessesInMemory = 2 1432 1433 executeLSQMaxStoreBufferStoresPerCycle = 2 1434 executeLSQRequestsQueueSize = 1 1435 executeLSQTransfersQueueSize = 2 1436 executeLSQStoreBufferSize = 5 1437 executeBranchDelay = 1 1438 executeFuncUnits = HPI_FUPool() 1439 executeSetTraceTimeOnCommit = True 1440 executeSetTraceTimeOnIssue = False 1441 1442 executeAllowEarlyMemoryIssue = True 1443 1444 enableIdling = True 1445 1446 branchPred = HPI_BP() 1447 1448 itb = HPI_ITB() 1449 dtb = HPI_DTB() 1450 1451__all__ = [ 1452 "HPI_BP", 1453 "HPI_ITB", "HPI_DTB", "HPI_WalkCache", 1454 "HPI_ICache", "HPI_DCache", "HPI_L2", 1455 "HPI", 1456]