src/amd/llvm/ac_llvm_util.h

   1 /*
   2  * Copyright 2016 Bas Nieuwenhuizen
   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
  26 #ifndef AC_LLVM_UTIL_H
  27 #define AC_LLVM_UTIL_H
  28
  29 #include <stdbool.h>
  30 #include <llvm-c/TargetMachine.h>
  31 #include <llvm/Config/llvm-config.h>
  32
  33 #include "amd_family.h"
  34
  35 #ifdef __cplusplus
  36 extern "C" {
  37 #endif
  38
  39 #if LLVM_VERSION_MAJOR < 11
  40 #define LLVMFixedVectorTypeKind LLVMVectorTypeKind
  41 #endif
  42
  43 struct ac_compiler_passes;
  44
  45 enum ac_func_attr {
  46         AC_FUNC_ATTR_ALWAYSINLINE = (1 << 0),
  47         AC_FUNC_ATTR_INREG        = (1 << 2),
  48         AC_FUNC_ATTR_NOALIAS      = (1 << 3),
  49         AC_FUNC_ATTR_NOUNWIND     = (1 << 4),
  50         AC_FUNC_ATTR_READNONE     = (1 << 5),
  51         AC_FUNC_ATTR_READONLY     = (1 << 6),
  52         AC_FUNC_ATTR_WRITEONLY    = (1 << 7),
  53         AC_FUNC_ATTR_INACCESSIBLE_MEM_ONLY = (1 << 8),
  54         AC_FUNC_ATTR_CONVERGENT = (1 << 9),
  55
  56         /* Legacy intrinsic that needs attributes on function declarations
  57          * and they must match the internal LLVM definition exactly, otherwise
  58          * intrinsic selection fails.
  59          */
  60         AC_FUNC_ATTR_LEGACY       = (1u << 31),
  61 };
  62
  63 enum ac_target_machine_options {
  64         AC_TM_SUPPORTS_SPILL = (1 << 0),
  65         AC_TM_FORCE_ENABLE_XNACK = (1 << 1),
  66         AC_TM_FORCE_DISABLE_XNACK = (1 << 2),
  67         AC_TM_PROMOTE_ALLOCA_TO_SCRATCH = (1 << 3),
  68         AC_TM_CHECK_IR = (1 << 4),
  69         AC_TM_ENABLE_GLOBAL_ISEL = (1 << 5),
  70         AC_TM_CREATE_LOW_OPT = (1 << 6),
  71         AC_TM_NO_LOAD_STORE_OPT = (1 << 7),
  72         AC_TM_WAVE32 = (1 << 8),
  73 };
  74
  75 enum ac_float_mode {
  76         AC_FLOAT_MODE_DEFAULT,
  77         AC_FLOAT_MODE_DEFAULT_OPENGL,
  78         AC_FLOAT_MODE_DENORM_FLUSH_TO_ZERO,
  79 };
  80
  81 /* Per-thread persistent LLVM objects. */
  82 struct ac_llvm_compiler {
  83         LLVMTargetLibraryInfoRef        target_library_info;
  84         LLVMPassManagerRef              passmgr;
  85
  86         /* Default compiler. */
  87         LLVMTargetMachineRef            tm;
  88         struct ac_compiler_passes       *passes;
  89
  90         /* Wave32 compiler for GFX10. */
  91         LLVMTargetMachineRef            tm_wave32;
  92         struct ac_compiler_passes       *passes_wave32;
  93
  94         /* Optional compiler for faster compilation with fewer optimizations.
  95          * LLVM modules can be created with "tm" too. There is no difference.
  96          */
  97         LLVMTargetMachineRef            low_opt_tm; /* uses -O1 instead of -O2 */
  98         struct ac_compiler_passes       *low_opt_passes;
  99 };
 100
 101 const char *ac_get_llvm_processor_name(enum radeon_family family);
 102 void ac_add_attr_dereferenceable(LLVMValueRef val, uint64_t bytes);
 103 bool ac_is_sgpr_param(LLVMValueRef param);
 104 void ac_add_function_attr(LLVMContextRef ctx, LLVMValueRef function,
 105                           int attr_idx, enum ac_func_attr attr);
 106 void ac_add_func_attributes(LLVMContextRef ctx, LLVMValueRef function,
 107                             unsigned attrib_mask);
 108 void ac_dump_module(LLVMModuleRef module);
 109
 110 LLVMValueRef ac_llvm_get_called_value(LLVMValueRef call);
 111 bool ac_llvm_is_function(LLVMValueRef v);
 112 LLVMModuleRef ac_create_module(LLVMTargetMachineRef tm, LLVMContextRef ctx);
 113
 114 LLVMBuilderRef ac_create_builder(LLVMContextRef ctx,
 115                                  enum ac_float_mode float_mode);
 116 bool ac_disable_inexact_math(LLVMBuilderRef builder);
 117 void ac_restore_inexact_math(LLVMBuilderRef builder, bool value);
 118
 119 void
 120 ac_llvm_add_target_dep_function_attr(LLVMValueRef F,
 121                                      const char *name, unsigned value);
 122 void ac_llvm_set_workgroup_size(LLVMValueRef F, unsigned size);
 123
 124 static inline unsigned
 125 ac_get_load_intr_attribs(bool can_speculate)
 126 {
 127         /* READNONE means writes can't affect it, while READONLY means that
 128          * writes can affect it. */
 129         return can_speculate ? AC_FUNC_ATTR_READNONE :
 130                                AC_FUNC_ATTR_READONLY;
 131 }
 132
 133 unsigned
 134 ac_count_scratch_private_memory(LLVMValueRef function);
 135
 136 LLVMTargetLibraryInfoRef ac_create_target_library_info(const char *triple);
 137 void ac_dispose_target_library_info(LLVMTargetLibraryInfoRef library_info);
 138 void ac_init_llvm_once(void);
 139
 140
 141 bool ac_init_llvm_compiler(struct ac_llvm_compiler *compiler,
 142                            enum radeon_family family,
 143                            enum ac_target_machine_options tm_options);
 144 void ac_destroy_llvm_compiler(struct ac_llvm_compiler *compiler);
 145
 146 struct ac_compiler_passes *ac_create_llvm_passes(LLVMTargetMachineRef tm);
 147 void ac_destroy_llvm_passes(struct ac_compiler_passes *p);
 148 bool ac_compile_module_to_elf(struct ac_compiler_passes *p, LLVMModuleRef module,
 149                               char **pelf_buffer, size_t *pelf_size);
 150 void ac_llvm_add_barrier_noop_pass(LLVMPassManagerRef passmgr);
 151 void ac_enable_global_isel(LLVMTargetMachineRef tm);
 152
 153 static inline bool
 154 ac_has_vec3_support(enum chip_class chip, bool use_format)
 155 {
 156         if (chip == GFX6 && !use_format) {
 157                 /* GFX6 only supports vec3 with load/store format. */
 158                 return false;
 159         }
 160
 161         return LLVM_VERSION_MAJOR >= 9;
 162 }
 163
 164 #ifdef __cplusplus
 165 }
 166 #endif
 167
 168 #endif /* AC_LLVM_UTIL_H */