src/amd/common/ac_llvm_util.c

   1 /*
   2  * Copyright 2014 Advanced Micro Devices, Inc.
   3  *
   4  * Permission is hereby granted, free of charge, to any person obtaining a
   5  * copy of this software and associated documentation files (the
   6  * "Software"), to deal in the Software without restriction, including
   7  * without limitation the rights to use, copy, modify, merge, publish,
   8  * distribute, sub license, and/or sell copies of the Software, and to
   9  * permit persons to whom the Software is furnished to do so, subject to
  10  * the following conditions:
  11  *
  12  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  13  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  14  * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
  15  * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM,
  16  * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
  17  * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
  18  * USE OR OTHER DEALINGS IN THE SOFTWARE.
  19  *
  20  * The above copyright notice and this permission notice (including the
  21  * next paragraph) shall be included in all copies or substantial portions
  22  * of the Software.
  23  *
  24  */
  25 /* based on pieces from si_pipe.c and radeon_llvm_emit.c */
  26 #include "ac_llvm_util.h"
  27 #include "ac_llvm_build.h"
  28 #include "util/bitscan.h"
  29 #include <llvm-c/Core.h>
  30 #include <llvm-c/Support.h>
  31 #include "c11/threads.h"
  32 #include "util/u_math.h"
  33
  34 #include <assert.h>
  35 #include <stdio.h>
  36 #include <string.h>
  37
  38 static void ac_init_llvm_target()
  39 {
  40         LLVMInitializeAMDGPUTargetInfo();
  41         LLVMInitializeAMDGPUTarget();
  42         LLVMInitializeAMDGPUTargetMC();
  43         LLVMInitializeAMDGPUAsmPrinter();
  44
  45         /* For inline assembly. */
  46         LLVMInitializeAMDGPUAsmParser();
  47
  48         /* Workaround for bug in llvm 4.0 that causes image intrinsics
  49          * to disappear.
  50          * https://reviews.llvm.org/D26348
  51          *
  52          * "mesa" is the prefix for error messages.
  53          */
  54         const char *argv[2] = { "mesa", "-simplifycfg-sink-common=false" };
  55         LLVMParseCommandLineOptions(2, argv, NULL);
  56 }
  57
  58 static once_flag ac_init_llvm_target_once_flag = ONCE_FLAG_INIT;
  59
  60 LLVMTargetRef ac_get_llvm_target(const char *triple)
  61 {
  62         LLVMTargetRef target = NULL;
  63         char *err_message = NULL;
  64
  65         call_once(&ac_init_llvm_target_once_flag, ac_init_llvm_target);
  66
  67         if (LLVMGetTargetFromTriple(triple, &target, &err_message)) {
  68                 fprintf(stderr, "Cannot find target for triple %s ", triple);
  69                 if (err_message) {
  70                         fprintf(stderr, "%s\n", err_message);
  71                 }
  72                 LLVMDisposeMessage(err_message);
  73                 return NULL;
  74         }
  75         return target;
  76 }
  77
  78 const char *ac_get_llvm_processor_name(enum radeon_family family)
  79 {
  80         switch (family) {
  81         case CHIP_TAHITI:
  82                 return "tahiti";
  83         case CHIP_PITCAIRN:
  84                 return "pitcairn";
  85         case CHIP_VERDE:
  86                 return "verde";
  87         case CHIP_OLAND:
  88                 return "oland";
  89         case CHIP_HAINAN:
  90                 return "hainan";
  91         case CHIP_BONAIRE:
  92                 return "bonaire";
  93         case CHIP_KABINI:
  94                 return "kabini";
  95         case CHIP_KAVERI:
  96                 return "kaveri";
  97         case CHIP_HAWAII:
  98                 return "hawaii";
  99         case CHIP_MULLINS:
 100                 return "mullins";
 101         case CHIP_TONGA:
 102                 return "tonga";
 103         case CHIP_ICELAND:
 104                 return "iceland";
 105         case CHIP_CARRIZO:
 106                 return "carrizo";
 107         case CHIP_FIJI:
 108                 return "fiji";
 109         case CHIP_STONEY:
 110                 return "stoney";
 111         case CHIP_POLARIS10:
 112                 return "polaris10";
 113         case CHIP_POLARIS11:
 114         case CHIP_POLARIS12:
 115         case CHIP_VEGAM:
 116                 return "polaris11";
 117         case CHIP_VEGA10:
 118         case CHIP_VEGA12:
 119         case CHIP_RAVEN:
 120                 return "gfx900";
 121         default:
 122                 return "";
 123         }
 124 }
 125
 126 LLVMTargetMachineRef ac_create_target_machine(enum radeon_family family,
 127                                               enum ac_target_machine_options tm_options,
 128                                               const char **out_triple)
 129 {
 130         assert(family >= CHIP_TAHITI);
 131         char features[256];
 132         const char *triple = (tm_options & AC_TM_SUPPORTS_SPILL) ? "amdgcn-mesa-mesa3d" : "amdgcn--";
 133         LLVMTargetRef target = ac_get_llvm_target(triple);
 134
 135         snprintf(features, sizeof(features),
 136                  "+DumpCode,+vgpr-spilling,-fp32-denormals,+fp64-denormals%s%s%s%s",
 137                  tm_options & AC_TM_SISCHED ? ",+si-scheduler" : "",
 138                  tm_options & AC_TM_FORCE_ENABLE_XNACK ? ",+xnack" : "",
 139                  tm_options & AC_TM_FORCE_DISABLE_XNACK ? ",-xnack" : "",
 140                  tm_options & AC_TM_PROMOTE_ALLOCA_TO_SCRATCH ? ",-promote-alloca" : "");
 141
 142         LLVMTargetMachineRef tm = LLVMCreateTargetMachine(
 143                                      target,
 144                                      triple,
 145                                      ac_get_llvm_processor_name(family),
 146                                      features,
 147                                      LLVMCodeGenLevelDefault,
 148                                      LLVMRelocDefault,
 149                                      LLVMCodeModelDefault);
 150
 151         if (out_triple)
 152                 *out_triple = triple;
 153         return tm;
 154 }
 155
 156 static const char *attr_to_str(enum ac_func_attr attr)
 157 {
 158    switch (attr) {
 159    case AC_FUNC_ATTR_ALWAYSINLINE: return "alwaysinline";
 160    case AC_FUNC_ATTR_INREG: return "inreg";
 161    case AC_FUNC_ATTR_NOALIAS: return "noalias";
 162    case AC_FUNC_ATTR_NOUNWIND: return "nounwind";
 163    case AC_FUNC_ATTR_READNONE: return "readnone";
 164    case AC_FUNC_ATTR_READONLY: return "readonly";
 165    case AC_FUNC_ATTR_WRITEONLY: return "writeonly";
 166    case AC_FUNC_ATTR_INACCESSIBLE_MEM_ONLY: return "inaccessiblememonly";
 167    case AC_FUNC_ATTR_CONVERGENT: return "convergent";
 168    default:
 169            fprintf(stderr, "Unhandled function attribute: %x\n", attr);
 170            return 0;
 171    }
 172 }
 173
 174 void
 175 ac_add_function_attr(LLVMContextRef ctx, LLVMValueRef function,
 176                      int attr_idx, enum ac_func_attr attr)
 177 {
 178    const char *attr_name = attr_to_str(attr);
 179    unsigned kind_id = LLVMGetEnumAttributeKindForName(attr_name,
 180                                                       strlen(attr_name));
 181    LLVMAttributeRef llvm_attr = LLVMCreateEnumAttribute(ctx, kind_id, 0);
 182
 183    if (LLVMIsAFunction(function))
 184       LLVMAddAttributeAtIndex(function, attr_idx, llvm_attr);
 185    else
 186       LLVMAddCallSiteAttribute(function, attr_idx, llvm_attr);
 187 }
 188
 189 void ac_add_func_attributes(LLVMContextRef ctx, LLVMValueRef function,
 190                             unsigned attrib_mask)
 191 {
 192         attrib_mask |= AC_FUNC_ATTR_NOUNWIND;
 193         attrib_mask &= ~AC_FUNC_ATTR_LEGACY;
 194
 195         while (attrib_mask) {
 196                 enum ac_func_attr attr = 1u << u_bit_scan(&attrib_mask);
 197                 ac_add_function_attr(ctx, function, -1, attr);
 198         }
 199 }
 200
 201 void
 202 ac_dump_module(LLVMModuleRef module)
 203 {
 204         char *str = LLVMPrintModuleToString(module);
 205         fprintf(stderr, "%s", str);
 206         LLVMDisposeMessage(str);
 207 }
 208
 209 void
 210 ac_llvm_add_target_dep_function_attr(LLVMValueRef F,
 211                                      const char *name, unsigned value)
 212 {
 213         char str[16];
 214
 215         snprintf(str, sizeof(str), "0x%x", value);
 216         LLVMAddTargetDependentFunctionAttr(F, name, str);
 217 }
 218
 219 unsigned
 220 ac_count_scratch_private_memory(LLVMValueRef function)
 221 {
 222         unsigned private_mem_vgprs = 0;
 223
 224         /* Process all LLVM instructions. */
 225         LLVMBasicBlockRef bb = LLVMGetFirstBasicBlock(function);
 226         while (bb) {
 227                 LLVMValueRef next = LLVMGetFirstInstruction(bb);
 228
 229                 while (next) {
 230                         LLVMValueRef inst = next;
 231                         next = LLVMGetNextInstruction(next);
 232
 233                         if (LLVMGetInstructionOpcode(inst) != LLVMAlloca)
 234                                 continue;
 235
 236                         LLVMTypeRef type = LLVMGetElementType(LLVMTypeOf(inst));
 237                         /* No idea why LLVM aligns allocas to 4 elements. */
 238                         unsigned alignment = LLVMGetAlignment(inst);
 239                         unsigned dw_size = align(ac_get_type_size(type) / 4, alignment);
 240                         private_mem_vgprs += dw_size;
 241                 }
 242                 bb = LLVMGetNextBasicBlock(bb);
 243         }
 244
 245         return private_mem_vgprs;
 246 }