10b57cec5SDimitry Andric //===--- AMDGPU.cpp - Implement AMDGPU target feature support -------------===// 20b57cec5SDimitry Andric // 30b57cec5SDimitry Andric // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. 40b57cec5SDimitry Andric // See https://llvm.org/LICENSE.txt for license information. 50b57cec5SDimitry Andric // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception 60b57cec5SDimitry Andric // 70b57cec5SDimitry Andric //===----------------------------------------------------------------------===// 80b57cec5SDimitry Andric // 90b57cec5SDimitry Andric // This file implements AMDGPU TargetInfo objects. 100b57cec5SDimitry Andric // 110b57cec5SDimitry Andric //===----------------------------------------------------------------------===// 120b57cec5SDimitry Andric 130b57cec5SDimitry Andric #include "AMDGPU.h" 140b57cec5SDimitry Andric #include "clang/Basic/Builtins.h" 150b57cec5SDimitry Andric #include "clang/Basic/CodeGenOptions.h" 16*bdd1243dSDimitry Andric #include "clang/Basic/Diagnostic.h" 170b57cec5SDimitry Andric #include "clang/Basic/LangOptions.h" 180b57cec5SDimitry Andric #include "clang/Basic/MacroBuilder.h" 190b57cec5SDimitry Andric #include "clang/Basic/TargetBuiltins.h" 200b57cec5SDimitry Andric 210b57cec5SDimitry Andric using namespace clang; 220b57cec5SDimitry Andric using namespace clang::targets; 230b57cec5SDimitry Andric 240b57cec5SDimitry Andric namespace clang { 250b57cec5SDimitry Andric namespace targets { 260b57cec5SDimitry Andric 270b57cec5SDimitry Andric // If you edit the description strings, make sure you update 280b57cec5SDimitry Andric // getPointerWidthV(). 290b57cec5SDimitry Andric 300b57cec5SDimitry Andric static const char *const DataLayoutStringR600 = 310b57cec5SDimitry Andric "e-p:32:32-i64:64-v16:16-v24:32-v32:32-v48:64-v96:128" 32e8d8bef9SDimitry Andric "-v192:256-v256:256-v512:512-v1024:1024-v2048:2048-n32:64-S32-A5-G1"; 330b57cec5SDimitry Andric 340b57cec5SDimitry Andric static const char *const DataLayoutStringAMDGCN = 350b57cec5SDimitry Andric "e-p:64:64-p1:64:64-p2:32:32-p3:32:32-p4:64:64-p5:32:32-p6:32:32" 360b57cec5SDimitry Andric "-i64:64-v16:16-v24:32-v32:32-v48:64-v96:128" 37e8d8bef9SDimitry Andric "-v192:256-v256:256-v512:512-v1024:1024-v2048:2048-n32:64-S32-A5-G1" 380b57cec5SDimitry Andric "-ni:7"; 390b57cec5SDimitry Andric 400b57cec5SDimitry Andric const LangASMap AMDGPUTargetInfo::AMDGPUDefIsGenMap = { 410b57cec5SDimitry Andric Generic, // Default 420b57cec5SDimitry Andric Global, // opencl_global 430b57cec5SDimitry Andric Local, // opencl_local 440b57cec5SDimitry Andric Constant, // opencl_constant 450b57cec5SDimitry Andric Private, // opencl_private 460b57cec5SDimitry Andric Generic, // opencl_generic 47e8d8bef9SDimitry Andric Global, // opencl_global_device 48e8d8bef9SDimitry Andric Global, // opencl_global_host 490b57cec5SDimitry Andric Global, // cuda_device 500b57cec5SDimitry Andric Constant, // cuda_constant 51480093f4SDimitry Andric Local, // cuda_shared 52fe6060f1SDimitry Andric Global, // sycl_global 53fe6060f1SDimitry Andric Global, // sycl_global_device 54fe6060f1SDimitry Andric Global, // sycl_global_host 55fe6060f1SDimitry Andric Local, // sycl_local 56fe6060f1SDimitry Andric Private, // sycl_private 57480093f4SDimitry Andric Generic, // ptr32_sptr 58480093f4SDimitry Andric Generic, // ptr32_uptr 59*bdd1243dSDimitry Andric Generic, // ptr64 60*bdd1243dSDimitry Andric Generic, // hlsl_groupshared 610b57cec5SDimitry Andric }; 620b57cec5SDimitry Andric 630b57cec5SDimitry Andric const LangASMap AMDGPUTargetInfo::AMDGPUDefIsPrivMap = { 640b57cec5SDimitry Andric Private, // Default 650b57cec5SDimitry Andric Global, // opencl_global 660b57cec5SDimitry Andric Local, // opencl_local 670b57cec5SDimitry Andric Constant, // opencl_constant 680b57cec5SDimitry Andric Private, // opencl_private 690b57cec5SDimitry Andric Generic, // opencl_generic 70e8d8bef9SDimitry Andric Global, // opencl_global_device 71e8d8bef9SDimitry Andric Global, // opencl_global_host 720b57cec5SDimitry Andric Global, // cuda_device 730b57cec5SDimitry Andric Constant, // cuda_constant 74480093f4SDimitry Andric Local, // cuda_shared 75fe6060f1SDimitry Andric // SYCL address space values for this map are dummy 76fe6060f1SDimitry Andric Generic, // sycl_global 77fe6060f1SDimitry Andric Generic, // sycl_global_device 78fe6060f1SDimitry Andric Generic, // sycl_global_host 79fe6060f1SDimitry Andric Generic, // sycl_local 80fe6060f1SDimitry Andric Generic, // sycl_private 81480093f4SDimitry Andric Generic, // ptr32_sptr 82480093f4SDimitry Andric Generic, // ptr32_uptr 83*bdd1243dSDimitry Andric Generic, // ptr64 84*bdd1243dSDimitry Andric Generic, // hlsl_groupshared 85480093f4SDimitry Andric 860b57cec5SDimitry Andric }; 870b57cec5SDimitry Andric } // namespace targets 880b57cec5SDimitry Andric } // namespace clang 890b57cec5SDimitry Andric 90*bdd1243dSDimitry Andric static constexpr Builtin::Info BuiltinInfo[] = { 910b57cec5SDimitry Andric #define BUILTIN(ID, TYPE, ATTRS) \ 92*bdd1243dSDimitry Andric {#ID, TYPE, ATTRS, nullptr, HeaderDesc::NO_HEADER, ALL_LANGUAGES}, 930b57cec5SDimitry Andric #define TARGET_BUILTIN(ID, TYPE, ATTRS, FEATURE) \ 94*bdd1243dSDimitry Andric {#ID, TYPE, ATTRS, FEATURE, HeaderDesc::NO_HEADER, ALL_LANGUAGES}, 950b57cec5SDimitry Andric #include "clang/Basic/BuiltinsAMDGPU.def" 960b57cec5SDimitry Andric }; 970b57cec5SDimitry Andric 980b57cec5SDimitry Andric const char *const AMDGPUTargetInfo::GCCRegNames[] = { 990b57cec5SDimitry Andric "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7", "v8", 1000b57cec5SDimitry Andric "v9", "v10", "v11", "v12", "v13", "v14", "v15", "v16", "v17", 1010b57cec5SDimitry Andric "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25", "v26", 1020b57cec5SDimitry Andric "v27", "v28", "v29", "v30", "v31", "v32", "v33", "v34", "v35", 1030b57cec5SDimitry Andric "v36", "v37", "v38", "v39", "v40", "v41", "v42", "v43", "v44", 1040b57cec5SDimitry Andric "v45", "v46", "v47", "v48", "v49", "v50", "v51", "v52", "v53", 1050b57cec5SDimitry Andric "v54", "v55", "v56", "v57", "v58", "v59", "v60", "v61", "v62", 1060b57cec5SDimitry Andric "v63", "v64", "v65", "v66", "v67", "v68", "v69", "v70", "v71", 1070b57cec5SDimitry Andric "v72", "v73", "v74", "v75", "v76", "v77", "v78", "v79", "v80", 1080b57cec5SDimitry Andric "v81", "v82", "v83", "v84", "v85", "v86", "v87", "v88", "v89", 1090b57cec5SDimitry Andric "v90", "v91", "v92", "v93", "v94", "v95", "v96", "v97", "v98", 1100b57cec5SDimitry Andric "v99", "v100", "v101", "v102", "v103", "v104", "v105", "v106", "v107", 1110b57cec5SDimitry Andric "v108", "v109", "v110", "v111", "v112", "v113", "v114", "v115", "v116", 1120b57cec5SDimitry Andric "v117", "v118", "v119", "v120", "v121", "v122", "v123", "v124", "v125", 1130b57cec5SDimitry Andric "v126", "v127", "v128", "v129", "v130", "v131", "v132", "v133", "v134", 1140b57cec5SDimitry Andric "v135", "v136", "v137", "v138", "v139", "v140", "v141", "v142", "v143", 1150b57cec5SDimitry Andric "v144", "v145", "v146", "v147", "v148", "v149", "v150", "v151", "v152", 1160b57cec5SDimitry Andric "v153", "v154", "v155", "v156", "v157", "v158", "v159", "v160", "v161", 1170b57cec5SDimitry Andric "v162", "v163", "v164", "v165", "v166", "v167", "v168", "v169", "v170", 1180b57cec5SDimitry Andric "v171", "v172", "v173", "v174", "v175", "v176", "v177", "v178", "v179", 1190b57cec5SDimitry Andric "v180", "v181", "v182", "v183", "v184", "v185", "v186", "v187", "v188", 1200b57cec5SDimitry Andric "v189", "v190", "v191", "v192", "v193", "v194", "v195", "v196", "v197", 1210b57cec5SDimitry Andric "v198", "v199", "v200", "v201", "v202", "v203", "v204", "v205", "v206", 1220b57cec5SDimitry Andric "v207", "v208", "v209", "v210", "v211", "v212", "v213", "v214", "v215", 1230b57cec5SDimitry Andric "v216", "v217", "v218", "v219", "v220", "v221", "v222", "v223", "v224", 1240b57cec5SDimitry Andric "v225", "v226", "v227", "v228", "v229", "v230", "v231", "v232", "v233", 1250b57cec5SDimitry Andric "v234", "v235", "v236", "v237", "v238", "v239", "v240", "v241", "v242", 1260b57cec5SDimitry Andric "v243", "v244", "v245", "v246", "v247", "v248", "v249", "v250", "v251", 1270b57cec5SDimitry Andric "v252", "v253", "v254", "v255", "s0", "s1", "s2", "s3", "s4", 1280b57cec5SDimitry Andric "s5", "s6", "s7", "s8", "s9", "s10", "s11", "s12", "s13", 1290b57cec5SDimitry Andric "s14", "s15", "s16", "s17", "s18", "s19", "s20", "s21", "s22", 1300b57cec5SDimitry Andric "s23", "s24", "s25", "s26", "s27", "s28", "s29", "s30", "s31", 1310b57cec5SDimitry Andric "s32", "s33", "s34", "s35", "s36", "s37", "s38", "s39", "s40", 1320b57cec5SDimitry Andric "s41", "s42", "s43", "s44", "s45", "s46", "s47", "s48", "s49", 1330b57cec5SDimitry Andric "s50", "s51", "s52", "s53", "s54", "s55", "s56", "s57", "s58", 1340b57cec5SDimitry Andric "s59", "s60", "s61", "s62", "s63", "s64", "s65", "s66", "s67", 1350b57cec5SDimitry Andric "s68", "s69", "s70", "s71", "s72", "s73", "s74", "s75", "s76", 1360b57cec5SDimitry Andric "s77", "s78", "s79", "s80", "s81", "s82", "s83", "s84", "s85", 1370b57cec5SDimitry Andric "s86", "s87", "s88", "s89", "s90", "s91", "s92", "s93", "s94", 1380b57cec5SDimitry Andric "s95", "s96", "s97", "s98", "s99", "s100", "s101", "s102", "s103", 1390b57cec5SDimitry Andric "s104", "s105", "s106", "s107", "s108", "s109", "s110", "s111", "s112", 1400b57cec5SDimitry Andric "s113", "s114", "s115", "s116", "s117", "s118", "s119", "s120", "s121", 1410b57cec5SDimitry Andric "s122", "s123", "s124", "s125", "s126", "s127", "exec", "vcc", "scc", 1420b57cec5SDimitry Andric "m0", "flat_scratch", "exec_lo", "exec_hi", "vcc_lo", "vcc_hi", 1435ffd83dbSDimitry Andric "flat_scratch_lo", "flat_scratch_hi", 1445ffd83dbSDimitry Andric "a0", "a1", "a2", "a3", "a4", "a5", "a6", "a7", "a8", 1455ffd83dbSDimitry Andric "a9", "a10", "a11", "a12", "a13", "a14", "a15", "a16", "a17", 1465ffd83dbSDimitry Andric "a18", "a19", "a20", "a21", "a22", "a23", "a24", "a25", "a26", 1475ffd83dbSDimitry Andric "a27", "a28", "a29", "a30", "a31", "a32", "a33", "a34", "a35", 1485ffd83dbSDimitry Andric "a36", "a37", "a38", "a39", "a40", "a41", "a42", "a43", "a44", 1495ffd83dbSDimitry Andric "a45", "a46", "a47", "a48", "a49", "a50", "a51", "a52", "a53", 1505ffd83dbSDimitry Andric "a54", "a55", "a56", "a57", "a58", "a59", "a60", "a61", "a62", 1515ffd83dbSDimitry Andric "a63", "a64", "a65", "a66", "a67", "a68", "a69", "a70", "a71", 1525ffd83dbSDimitry Andric "a72", "a73", "a74", "a75", "a76", "a77", "a78", "a79", "a80", 1535ffd83dbSDimitry Andric "a81", "a82", "a83", "a84", "a85", "a86", "a87", "a88", "a89", 1545ffd83dbSDimitry Andric "a90", "a91", "a92", "a93", "a94", "a95", "a96", "a97", "a98", 1555ffd83dbSDimitry Andric "a99", "a100", "a101", "a102", "a103", "a104", "a105", "a106", "a107", 1565ffd83dbSDimitry Andric "a108", "a109", "a110", "a111", "a112", "a113", "a114", "a115", "a116", 1575ffd83dbSDimitry Andric "a117", "a118", "a119", "a120", "a121", "a122", "a123", "a124", "a125", 1585ffd83dbSDimitry Andric "a126", "a127", "a128", "a129", "a130", "a131", "a132", "a133", "a134", 1595ffd83dbSDimitry Andric "a135", "a136", "a137", "a138", "a139", "a140", "a141", "a142", "a143", 1605ffd83dbSDimitry Andric "a144", "a145", "a146", "a147", "a148", "a149", "a150", "a151", "a152", 1615ffd83dbSDimitry Andric "a153", "a154", "a155", "a156", "a157", "a158", "a159", "a160", "a161", 1625ffd83dbSDimitry Andric "a162", "a163", "a164", "a165", "a166", "a167", "a168", "a169", "a170", 1635ffd83dbSDimitry Andric "a171", "a172", "a173", "a174", "a175", "a176", "a177", "a178", "a179", 1645ffd83dbSDimitry Andric "a180", "a181", "a182", "a183", "a184", "a185", "a186", "a187", "a188", 1655ffd83dbSDimitry Andric "a189", "a190", "a191", "a192", "a193", "a194", "a195", "a196", "a197", 1665ffd83dbSDimitry Andric "a198", "a199", "a200", "a201", "a202", "a203", "a204", "a205", "a206", 1675ffd83dbSDimitry Andric "a207", "a208", "a209", "a210", "a211", "a212", "a213", "a214", "a215", 1685ffd83dbSDimitry Andric "a216", "a217", "a218", "a219", "a220", "a221", "a222", "a223", "a224", 1695ffd83dbSDimitry Andric "a225", "a226", "a227", "a228", "a229", "a230", "a231", "a232", "a233", 1705ffd83dbSDimitry Andric "a234", "a235", "a236", "a237", "a238", "a239", "a240", "a241", "a242", 1715ffd83dbSDimitry Andric "a243", "a244", "a245", "a246", "a247", "a248", "a249", "a250", "a251", 1725ffd83dbSDimitry Andric "a252", "a253", "a254", "a255" 1730b57cec5SDimitry Andric }; 1740b57cec5SDimitry Andric 1750b57cec5SDimitry Andric ArrayRef<const char *> AMDGPUTargetInfo::getGCCRegNames() const { 176*bdd1243dSDimitry Andric return llvm::ArrayRef(GCCRegNames); 1770b57cec5SDimitry Andric } 1780b57cec5SDimitry Andric 1790b57cec5SDimitry Andric bool AMDGPUTargetInfo::initFeatureMap( 1800b57cec5SDimitry Andric llvm::StringMap<bool> &Features, DiagnosticsEngine &Diags, StringRef CPU, 1810b57cec5SDimitry Andric const std::vector<std::string> &FeatureVec) const { 182*bdd1243dSDimitry Andric const bool IsNullCPU = CPU.empty(); 183*bdd1243dSDimitry Andric bool IsWave32Capable = false; 1840b57cec5SDimitry Andric 1850b57cec5SDimitry Andric using namespace llvm::AMDGPU; 1860b57cec5SDimitry Andric 1870b57cec5SDimitry Andric // XXX - What does the member GPU mean if device name string passed here? 1880b57cec5SDimitry Andric if (isAMDGCN(getTriple())) { 1890b57cec5SDimitry Andric switch (llvm::AMDGPU::parseArchAMDGCN(CPU)) { 19081ad6265SDimitry Andric case GK_GFX1103: 19181ad6265SDimitry Andric case GK_GFX1102: 19281ad6265SDimitry Andric case GK_GFX1101: 19381ad6265SDimitry Andric case GK_GFX1100: 194*bdd1243dSDimitry Andric IsWave32Capable = true; 19581ad6265SDimitry Andric Features["ci-insts"] = true; 19681ad6265SDimitry Andric Features["dot5-insts"] = true; 19781ad6265SDimitry Andric Features["dot7-insts"] = true; 19881ad6265SDimitry Andric Features["dot8-insts"] = true; 199*bdd1243dSDimitry Andric Features["dot9-insts"] = true; 20081ad6265SDimitry Andric Features["dl-insts"] = true; 20181ad6265SDimitry Andric Features["16-bit-insts"] = true; 20281ad6265SDimitry Andric Features["dpp"] = true; 20381ad6265SDimitry Andric Features["gfx8-insts"] = true; 20481ad6265SDimitry Andric Features["gfx9-insts"] = true; 20581ad6265SDimitry Andric Features["gfx10-insts"] = true; 20681ad6265SDimitry Andric Features["gfx10-3-insts"] = true; 20781ad6265SDimitry Andric Features["gfx11-insts"] = true; 20881ad6265SDimitry Andric break; 20981ad6265SDimitry Andric case GK_GFX1036: 210fe6060f1SDimitry Andric case GK_GFX1035: 211fe6060f1SDimitry Andric case GK_GFX1034: 212e8d8bef9SDimitry Andric case GK_GFX1033: 213e8d8bef9SDimitry Andric case GK_GFX1032: 214e8d8bef9SDimitry Andric case GK_GFX1031: 2155ffd83dbSDimitry Andric case GK_GFX1030: 216*bdd1243dSDimitry Andric IsWave32Capable = true; 2175ffd83dbSDimitry Andric Features["ci-insts"] = true; 2185ffd83dbSDimitry Andric Features["dot1-insts"] = true; 2195ffd83dbSDimitry Andric Features["dot2-insts"] = true; 2205ffd83dbSDimitry Andric Features["dot5-insts"] = true; 2215ffd83dbSDimitry Andric Features["dot6-insts"] = true; 222fe6060f1SDimitry Andric Features["dot7-insts"] = true; 2235ffd83dbSDimitry Andric Features["dl-insts"] = true; 2245ffd83dbSDimitry Andric Features["16-bit-insts"] = true; 2255ffd83dbSDimitry Andric Features["dpp"] = true; 2265ffd83dbSDimitry Andric Features["gfx8-insts"] = true; 2275ffd83dbSDimitry Andric Features["gfx9-insts"] = true; 2285ffd83dbSDimitry Andric Features["gfx10-insts"] = true; 2295ffd83dbSDimitry Andric Features["gfx10-3-insts"] = true; 2305ffd83dbSDimitry Andric Features["s-memrealtime"] = true; 231fe6060f1SDimitry Andric Features["s-memtime-inst"] = true; 2325ffd83dbSDimitry Andric break; 2330b57cec5SDimitry Andric case GK_GFX1012: 2340b57cec5SDimitry Andric case GK_GFX1011: 2350b57cec5SDimitry Andric Features["dot1-insts"] = true; 2360b57cec5SDimitry Andric Features["dot2-insts"] = true; 2370b57cec5SDimitry Andric Features["dot5-insts"] = true; 2380b57cec5SDimitry Andric Features["dot6-insts"] = true; 239fe6060f1SDimitry Andric Features["dot7-insts"] = true; 240*bdd1243dSDimitry Andric [[fallthrough]]; 241fe6060f1SDimitry Andric case GK_GFX1013: 2420b57cec5SDimitry Andric case GK_GFX1010: 243*bdd1243dSDimitry Andric IsWave32Capable = true; 2440b57cec5SDimitry Andric Features["dl-insts"] = true; 2450b57cec5SDimitry Andric Features["ci-insts"] = true; 2460b57cec5SDimitry Andric Features["16-bit-insts"] = true; 2470b57cec5SDimitry Andric Features["dpp"] = true; 2480b57cec5SDimitry Andric Features["gfx8-insts"] = true; 2490b57cec5SDimitry Andric Features["gfx9-insts"] = true; 2500b57cec5SDimitry Andric Features["gfx10-insts"] = true; 2510b57cec5SDimitry Andric Features["s-memrealtime"] = true; 252fe6060f1SDimitry Andric Features["s-memtime-inst"] = true; 2530b57cec5SDimitry Andric break; 25481ad6265SDimitry Andric case GK_GFX940: 25581ad6265SDimitry Andric Features["gfx940-insts"] = true; 256fcaf7f86SDimitry Andric Features["fp8-insts"] = true; 257*bdd1243dSDimitry Andric [[fallthrough]]; 258fe6060f1SDimitry Andric case GK_GFX90A: 259fe6060f1SDimitry Andric Features["gfx90a-insts"] = true; 260*bdd1243dSDimitry Andric [[fallthrough]]; 2610b57cec5SDimitry Andric case GK_GFX908: 2620b57cec5SDimitry Andric Features["dot3-insts"] = true; 2630b57cec5SDimitry Andric Features["dot4-insts"] = true; 2640b57cec5SDimitry Andric Features["dot5-insts"] = true; 2650b57cec5SDimitry Andric Features["dot6-insts"] = true; 2665ffd83dbSDimitry Andric Features["mai-insts"] = true; 267*bdd1243dSDimitry Andric [[fallthrough]]; 2680b57cec5SDimitry Andric case GK_GFX906: 2690b57cec5SDimitry Andric Features["dl-insts"] = true; 2700b57cec5SDimitry Andric Features["dot1-insts"] = true; 2710b57cec5SDimitry Andric Features["dot2-insts"] = true; 272fe6060f1SDimitry Andric Features["dot7-insts"] = true; 273*bdd1243dSDimitry Andric [[fallthrough]]; 274e8d8bef9SDimitry Andric case GK_GFX90C: 2750b57cec5SDimitry Andric case GK_GFX909: 2760b57cec5SDimitry Andric case GK_GFX904: 2770b57cec5SDimitry Andric case GK_GFX902: 2780b57cec5SDimitry Andric case GK_GFX900: 2790b57cec5SDimitry Andric Features["gfx9-insts"] = true; 280*bdd1243dSDimitry Andric [[fallthrough]]; 2810b57cec5SDimitry Andric case GK_GFX810: 282e8d8bef9SDimitry Andric case GK_GFX805: 2830b57cec5SDimitry Andric case GK_GFX803: 2840b57cec5SDimitry Andric case GK_GFX802: 2850b57cec5SDimitry Andric case GK_GFX801: 2860b57cec5SDimitry Andric Features["gfx8-insts"] = true; 2870b57cec5SDimitry Andric Features["16-bit-insts"] = true; 2880b57cec5SDimitry Andric Features["dpp"] = true; 2890b57cec5SDimitry Andric Features["s-memrealtime"] = true; 290*bdd1243dSDimitry Andric [[fallthrough]]; 291e8d8bef9SDimitry Andric case GK_GFX705: 2920b57cec5SDimitry Andric case GK_GFX704: 2930b57cec5SDimitry Andric case GK_GFX703: 2940b57cec5SDimitry Andric case GK_GFX702: 2950b57cec5SDimitry Andric case GK_GFX701: 2960b57cec5SDimitry Andric case GK_GFX700: 2970b57cec5SDimitry Andric Features["ci-insts"] = true; 298*bdd1243dSDimitry Andric [[fallthrough]]; 299e8d8bef9SDimitry Andric case GK_GFX602: 3000b57cec5SDimitry Andric case GK_GFX601: 3010b57cec5SDimitry Andric case GK_GFX600: 302fe6060f1SDimitry Andric Features["s-memtime-inst"] = true; 3030b57cec5SDimitry Andric break; 3040b57cec5SDimitry Andric case GK_NONE: 305a7dea167SDimitry Andric break; 3060b57cec5SDimitry Andric default: 3070b57cec5SDimitry Andric llvm_unreachable("Unhandled GPU!"); 3080b57cec5SDimitry Andric } 3090b57cec5SDimitry Andric } else { 3100b57cec5SDimitry Andric if (CPU.empty()) 3110b57cec5SDimitry Andric CPU = "r600"; 3120b57cec5SDimitry Andric 3130b57cec5SDimitry Andric switch (llvm::AMDGPU::parseArchR600(CPU)) { 3140b57cec5SDimitry Andric case GK_CAYMAN: 3150b57cec5SDimitry Andric case GK_CYPRESS: 3160b57cec5SDimitry Andric case GK_RV770: 3170b57cec5SDimitry Andric case GK_RV670: 3180b57cec5SDimitry Andric // TODO: Add fp64 when implemented. 3190b57cec5SDimitry Andric break; 3200b57cec5SDimitry Andric case GK_TURKS: 3210b57cec5SDimitry Andric case GK_CAICOS: 3220b57cec5SDimitry Andric case GK_BARTS: 3230b57cec5SDimitry Andric case GK_SUMO: 3240b57cec5SDimitry Andric case GK_REDWOOD: 3250b57cec5SDimitry Andric case GK_JUNIPER: 3260b57cec5SDimitry Andric case GK_CEDAR: 3270b57cec5SDimitry Andric case GK_RV730: 3280b57cec5SDimitry Andric case GK_RV710: 3290b57cec5SDimitry Andric case GK_RS880: 3300b57cec5SDimitry Andric case GK_R630: 3310b57cec5SDimitry Andric case GK_R600: 3320b57cec5SDimitry Andric break; 3330b57cec5SDimitry Andric default: 3340b57cec5SDimitry Andric llvm_unreachable("Unhandled GPU!"); 3350b57cec5SDimitry Andric } 3360b57cec5SDimitry Andric } 3370b57cec5SDimitry Andric 338*bdd1243dSDimitry Andric if (!TargetInfo::initFeatureMap(Features, Diags, CPU, FeatureVec)) 339*bdd1243dSDimitry Andric return false; 340*bdd1243dSDimitry Andric 341*bdd1243dSDimitry Andric // FIXME: Not diagnosing wavefrontsize32 on wave64 only targets. 342*bdd1243dSDimitry Andric const bool HaveWave32 = 343*bdd1243dSDimitry Andric (IsWave32Capable || IsNullCPU) && Features.count("wavefrontsize32"); 344*bdd1243dSDimitry Andric const bool HaveWave64 = Features.count("wavefrontsize64"); 345*bdd1243dSDimitry Andric 346*bdd1243dSDimitry Andric // TODO: Should move this logic into TargetParser 347*bdd1243dSDimitry Andric if (HaveWave32 && HaveWave64) { 348*bdd1243dSDimitry Andric Diags.Report(diag::err_invalid_feature_combination) 349*bdd1243dSDimitry Andric << "'wavefrontsize32' and 'wavefrontsize64' are mutually exclusive"; 350*bdd1243dSDimitry Andric return false; 351*bdd1243dSDimitry Andric } 352*bdd1243dSDimitry Andric 353*bdd1243dSDimitry Andric // Don't assume any wavesize with an unknown subtarget. 354*bdd1243dSDimitry Andric if (!IsNullCPU) { 355*bdd1243dSDimitry Andric // Default to wave32 if available, or wave64 if not 356*bdd1243dSDimitry Andric if (!HaveWave32 && !HaveWave64) { 357*bdd1243dSDimitry Andric StringRef DefaultWaveSizeFeature = 358*bdd1243dSDimitry Andric IsWave32Capable ? "wavefrontsize32" : "wavefrontsize64"; 359*bdd1243dSDimitry Andric Features.insert(std::make_pair(DefaultWaveSizeFeature, true)); 360*bdd1243dSDimitry Andric } 361*bdd1243dSDimitry Andric } 362*bdd1243dSDimitry Andric 363*bdd1243dSDimitry Andric return true; 3640b57cec5SDimitry Andric } 3650b57cec5SDimitry Andric 3660b57cec5SDimitry Andric void AMDGPUTargetInfo::fillValidCPUList( 3670b57cec5SDimitry Andric SmallVectorImpl<StringRef> &Values) const { 3680b57cec5SDimitry Andric if (isAMDGCN(getTriple())) 3690b57cec5SDimitry Andric llvm::AMDGPU::fillValidArchListAMDGCN(Values); 3700b57cec5SDimitry Andric else 3710b57cec5SDimitry Andric llvm::AMDGPU::fillValidArchListR600(Values); 3720b57cec5SDimitry Andric } 3730b57cec5SDimitry Andric 3740b57cec5SDimitry Andric void AMDGPUTargetInfo::setAddressSpaceMap(bool DefaultIsPrivate) { 3750b57cec5SDimitry Andric AddrSpaceMap = DefaultIsPrivate ? &AMDGPUDefIsPrivMap : &AMDGPUDefIsGenMap; 3760b57cec5SDimitry Andric } 3770b57cec5SDimitry Andric 3780b57cec5SDimitry Andric AMDGPUTargetInfo::AMDGPUTargetInfo(const llvm::Triple &Triple, 3790b57cec5SDimitry Andric const TargetOptions &Opts) 3800b57cec5SDimitry Andric : TargetInfo(Triple), 3810b57cec5SDimitry Andric GPUKind(isAMDGCN(Triple) ? 3820b57cec5SDimitry Andric llvm::AMDGPU::parseArchAMDGCN(Opts.CPU) : 3830b57cec5SDimitry Andric llvm::AMDGPU::parseArchR600(Opts.CPU)), 3840b57cec5SDimitry Andric GPUFeatures(isAMDGCN(Triple) ? 3850b57cec5SDimitry Andric llvm::AMDGPU::getArchAttrAMDGCN(GPUKind) : 3860b57cec5SDimitry Andric llvm::AMDGPU::getArchAttrR600(GPUKind)) { 3870b57cec5SDimitry Andric resetDataLayout(isAMDGCN(getTriple()) ? DataLayoutStringAMDGCN 3880b57cec5SDimitry Andric : DataLayoutStringR600); 3890b57cec5SDimitry Andric 3900b57cec5SDimitry Andric setAddressSpaceMap(Triple.getOS() == llvm::Triple::Mesa3D || 3910b57cec5SDimitry Andric !isAMDGCN(Triple)); 3920b57cec5SDimitry Andric UseAddrSpaceMapMangling = true; 3930b57cec5SDimitry Andric 394*bdd1243dSDimitry Andric if (isAMDGCN(Triple)) { 395*bdd1243dSDimitry Andric // __bf16 is always available as a load/store only type on AMDGCN. 396*bdd1243dSDimitry Andric BFloat16Width = BFloat16Align = 16; 397*bdd1243dSDimitry Andric BFloat16Format = &llvm::APFloat::BFloat(); 398*bdd1243dSDimitry Andric } 399*bdd1243dSDimitry Andric 4000b57cec5SDimitry Andric HasLegalHalfType = true; 4010b57cec5SDimitry Andric HasFloat16 = true; 402e8d8bef9SDimitry Andric WavefrontSize = GPUFeatures & llvm::AMDGPU::FEATURE_WAVE32 ? 32 : 64; 403e8d8bef9SDimitry Andric AllowAMDGPUUnsafeFPAtomics = Opts.AllowAMDGPUUnsafeFPAtomics; 4040b57cec5SDimitry Andric 405*bdd1243dSDimitry Andric // Set pointer width and alignment for the generic address space. 406*bdd1243dSDimitry Andric PointerWidth = PointerAlign = getPointerWidthV(LangAS::Default); 4070b57cec5SDimitry Andric if (getMaxPointerWidth() == 64) { 4080b57cec5SDimitry Andric LongWidth = LongAlign = 64; 4090b57cec5SDimitry Andric SizeType = UnsignedLong; 4100b57cec5SDimitry Andric PtrDiffType = SignedLong; 4110b57cec5SDimitry Andric IntPtrType = SignedLong; 4120b57cec5SDimitry Andric } 4130b57cec5SDimitry Andric 4140b57cec5SDimitry Andric MaxAtomicPromoteWidth = MaxAtomicInlineWidth = 64; 4150b57cec5SDimitry Andric } 4160b57cec5SDimitry Andric 417fe6060f1SDimitry Andric void AMDGPUTargetInfo::adjust(DiagnosticsEngine &Diags, LangOptions &Opts) { 418fe6060f1SDimitry Andric TargetInfo::adjust(Diags, Opts); 4190b57cec5SDimitry Andric // ToDo: There are still a few places using default address space as private 4200b57cec5SDimitry Andric // address space in OpenCL, which needs to be cleaned up, then Opts.OpenCL 4210b57cec5SDimitry Andric // can be removed from the following line. 4220b57cec5SDimitry Andric setAddressSpaceMap(/*DefaultIsPrivate=*/Opts.OpenCL || 4230b57cec5SDimitry Andric !isAMDGCN(getTriple())); 4240b57cec5SDimitry Andric } 4250b57cec5SDimitry Andric 4260b57cec5SDimitry Andric ArrayRef<Builtin::Info> AMDGPUTargetInfo::getTargetBuiltins() const { 427*bdd1243dSDimitry Andric return llvm::ArrayRef(BuiltinInfo, 428*bdd1243dSDimitry Andric clang::AMDGPU::LastTSBuiltin - Builtin::FirstTSBuiltin); 4290b57cec5SDimitry Andric } 4300b57cec5SDimitry Andric 4310b57cec5SDimitry Andric void AMDGPUTargetInfo::getTargetDefines(const LangOptions &Opts, 4320b57cec5SDimitry Andric MacroBuilder &Builder) const { 4330b57cec5SDimitry Andric Builder.defineMacro("__AMD__"); 4340b57cec5SDimitry Andric Builder.defineMacro("__AMDGPU__"); 4350b57cec5SDimitry Andric 4360b57cec5SDimitry Andric if (isAMDGCN(getTriple())) 4370b57cec5SDimitry Andric Builder.defineMacro("__AMDGCN__"); 4380b57cec5SDimitry Andric else 4390b57cec5SDimitry Andric Builder.defineMacro("__R600__"); 4400b57cec5SDimitry Andric 4410b57cec5SDimitry Andric if (GPUKind != llvm::AMDGPU::GK_NONE) { 4420b57cec5SDimitry Andric StringRef CanonName = isAMDGCN(getTriple()) ? 4430b57cec5SDimitry Andric getArchNameAMDGCN(GPUKind) : getArchNameR600(GPUKind); 4440b57cec5SDimitry Andric Builder.defineMacro(Twine("__") + Twine(CanonName) + Twine("__")); 44581ad6265SDimitry Andric // Emit macros for gfx family e.g. gfx906 -> __GFX9__, gfx1030 -> __GFX10___ 44681ad6265SDimitry Andric if (isAMDGCN(getTriple())) { 44781ad6265SDimitry Andric assert(CanonName.startswith("gfx") && "Invalid amdgcn canonical name"); 44881ad6265SDimitry Andric Builder.defineMacro(Twine("__") + Twine(CanonName.drop_back(2).upper()) + 44981ad6265SDimitry Andric Twine("__")); 45081ad6265SDimitry Andric } 451e8d8bef9SDimitry Andric if (isAMDGCN(getTriple())) { 452e8d8bef9SDimitry Andric Builder.defineMacro("__amdgcn_processor__", 453e8d8bef9SDimitry Andric Twine("\"") + Twine(CanonName) + Twine("\"")); 454e8d8bef9SDimitry Andric Builder.defineMacro("__amdgcn_target_id__", 45581ad6265SDimitry Andric Twine("\"") + Twine(*getTargetID()) + Twine("\"")); 456e8d8bef9SDimitry Andric for (auto F : getAllPossibleTargetIDFeatures(getTriple(), CanonName)) { 457e8d8bef9SDimitry Andric auto Loc = OffloadArchFeatures.find(F); 458e8d8bef9SDimitry Andric if (Loc != OffloadArchFeatures.end()) { 459e8d8bef9SDimitry Andric std::string NewF = F.str(); 460e8d8bef9SDimitry Andric std::replace(NewF.begin(), NewF.end(), '-', '_'); 461e8d8bef9SDimitry Andric Builder.defineMacro(Twine("__amdgcn_feature_") + Twine(NewF) + 462e8d8bef9SDimitry Andric Twine("__"), 463e8d8bef9SDimitry Andric Loc->second ? "1" : "0"); 464e8d8bef9SDimitry Andric } 465e8d8bef9SDimitry Andric } 466e8d8bef9SDimitry Andric } 4670b57cec5SDimitry Andric } 4680b57cec5SDimitry Andric 46981ad6265SDimitry Andric if (AllowAMDGPUUnsafeFPAtomics) 47081ad6265SDimitry Andric Builder.defineMacro("__AMDGCN_UNSAFE_FP_ATOMICS__"); 47181ad6265SDimitry Andric 4720b57cec5SDimitry Andric // TODO: __HAS_FMAF__, __HAS_LDEXPF__, __HAS_FP64__ are deprecated and will be 4730b57cec5SDimitry Andric // removed in the near future. 4740b57cec5SDimitry Andric if (hasFMAF()) 4750b57cec5SDimitry Andric Builder.defineMacro("__HAS_FMAF__"); 4760b57cec5SDimitry Andric if (hasFastFMAF()) 4770b57cec5SDimitry Andric Builder.defineMacro("FP_FAST_FMAF"); 4780b57cec5SDimitry Andric if (hasLDEXPF()) 4790b57cec5SDimitry Andric Builder.defineMacro("__HAS_LDEXPF__"); 4800b57cec5SDimitry Andric if (hasFP64()) 4810b57cec5SDimitry Andric Builder.defineMacro("__HAS_FP64__"); 4820b57cec5SDimitry Andric if (hasFastFMA()) 4830b57cec5SDimitry Andric Builder.defineMacro("FP_FAST_FMA"); 484e8d8bef9SDimitry Andric 485e8d8bef9SDimitry Andric Builder.defineMacro("__AMDGCN_WAVEFRONT_SIZE", Twine(WavefrontSize)); 4860b57cec5SDimitry Andric } 4870b57cec5SDimitry Andric 4880b57cec5SDimitry Andric void AMDGPUTargetInfo::setAuxTarget(const TargetInfo *Aux) { 4890b57cec5SDimitry Andric assert(HalfFormat == Aux->HalfFormat); 4900b57cec5SDimitry Andric assert(FloatFormat == Aux->FloatFormat); 4910b57cec5SDimitry Andric assert(DoubleFormat == Aux->DoubleFormat); 4920b57cec5SDimitry Andric 4930b57cec5SDimitry Andric // On x86_64 long double is 80-bit extended precision format, which is 4940b57cec5SDimitry Andric // not supported by AMDGPU. 128-bit floating point format is also not 4950b57cec5SDimitry Andric // supported by AMDGPU. Therefore keep its own format for these two types. 4960b57cec5SDimitry Andric auto SaveLongDoubleFormat = LongDoubleFormat; 4970b57cec5SDimitry Andric auto SaveFloat128Format = Float128Format; 49881ad6265SDimitry Andric auto SaveLongDoubleWidth = LongDoubleWidth; 49981ad6265SDimitry Andric auto SaveLongDoubleAlign = LongDoubleAlign; 5000b57cec5SDimitry Andric copyAuxTarget(Aux); 5010b57cec5SDimitry Andric LongDoubleFormat = SaveLongDoubleFormat; 5020b57cec5SDimitry Andric Float128Format = SaveFloat128Format; 50381ad6265SDimitry Andric LongDoubleWidth = SaveLongDoubleWidth; 50481ad6265SDimitry Andric LongDoubleAlign = SaveLongDoubleAlign; 5055ffd83dbSDimitry Andric // For certain builtin types support on the host target, claim they are 5065ffd83dbSDimitry Andric // support to pass the compilation of the host code during the device-side 5075ffd83dbSDimitry Andric // compilation. 5085ffd83dbSDimitry Andric // FIXME: As the side effect, we also accept `__float128` uses in the device 5095ffd83dbSDimitry Andric // code. To rejct these builtin types supported in the host target but not in 5105ffd83dbSDimitry Andric // the device target, one approach would support `device_builtin` attribute 5115ffd83dbSDimitry Andric // so that we could tell the device builtin types from the host ones. The 5125ffd83dbSDimitry Andric // also solves the different representations of the same builtin type, such 5135ffd83dbSDimitry Andric // as `size_t` in the MSVC environment. 5145ffd83dbSDimitry Andric if (Aux->hasFloat128Type()) { 5155ffd83dbSDimitry Andric HasFloat128 = true; 5165ffd83dbSDimitry Andric Float128Format = DoubleFormat; 5175ffd83dbSDimitry Andric } 5180b57cec5SDimitry Andric } 519