2 * Copyright 2016 Bas Nieuwenhuizen
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:
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.
20 * The above copyright notice and this permission notice (including the
21 * next paragraph) shall be included in all copies or substantial portions
26 #ifndef AC_LLVM_UTIL_H
27 #define AC_LLVM_UTIL_H
29 #include "amd_family.h"
30 #include <llvm-c/TargetMachine.h>
31 #include <llvm/Config/llvm-config.h>
39 struct ac_compiler_passes
;
40 struct ac_llvm_context
;
44 AC_FUNC_ATTR_ALWAYSINLINE
= (1 << 0),
45 AC_FUNC_ATTR_INREG
= (1 << 2),
46 AC_FUNC_ATTR_NOALIAS
= (1 << 3),
47 AC_FUNC_ATTR_NOUNWIND
= (1 << 4),
48 AC_FUNC_ATTR_READNONE
= (1 << 5),
49 AC_FUNC_ATTR_READONLY
= (1 << 6),
50 AC_FUNC_ATTR_WRITEONLY
= (1 << 7),
51 AC_FUNC_ATTR_INACCESSIBLE_MEM_ONLY
= (1 << 8),
52 AC_FUNC_ATTR_CONVERGENT
= (1 << 9),
54 /* Legacy intrinsic that needs attributes on function declarations
55 * and they must match the internal LLVM definition exactly, otherwise
56 * intrinsic selection fails.
58 AC_FUNC_ATTR_LEGACY
= (1u << 31),
61 enum ac_target_machine_options
63 AC_TM_SUPPORTS_SPILL
= (1 << 0),
64 AC_TM_FORCE_ENABLE_XNACK
= (1 << 1),
65 AC_TM_FORCE_DISABLE_XNACK
= (1 << 2),
66 AC_TM_PROMOTE_ALLOCA_TO_SCRATCH
= (1 << 3),
67 AC_TM_CHECK_IR
= (1 << 4),
68 AC_TM_ENABLE_GLOBAL_ISEL
= (1 << 5),
69 AC_TM_CREATE_LOW_OPT
= (1 << 6),
70 AC_TM_WAVE32
= (1 << 7),
75 AC_FLOAT_MODE_DEFAULT
,
76 AC_FLOAT_MODE_DEFAULT_OPENGL
,
77 AC_FLOAT_MODE_DENORM_FLUSH_TO_ZERO
,
80 /* Per-thread persistent LLVM objects. */
81 struct ac_llvm_compiler
{
82 LLVMTargetLibraryInfoRef target_library_info
;
83 LLVMPassManagerRef passmgr
;
85 /* Default compiler. */
86 LLVMTargetMachineRef tm
;
87 struct ac_compiler_passes
*passes
;
89 /* Wave32 compiler for GFX10. */
90 LLVMTargetMachineRef tm_wave32
;
91 struct ac_compiler_passes
*passes_wave32
;
93 /* Optional compiler for faster compilation with fewer optimizations.
94 * LLVM modules can be created with "tm" too. There is no difference.
96 LLVMTargetMachineRef low_opt_tm
; /* uses -O1 instead of -O2 */
97 struct ac_compiler_passes
*low_opt_passes
;
100 const char *ac_get_llvm_processor_name(enum radeon_family family
);
101 void ac_add_attr_dereferenceable(LLVMValueRef val
, uint64_t bytes
);
102 void ac_add_attr_alignment(LLVMValueRef val
, uint64_t bytes
);
103 bool ac_is_sgpr_param(LLVMValueRef param
);
104 void ac_add_function_attr(LLVMContextRef ctx
, LLVMValueRef function
, int attr_idx
,
105 enum ac_func_attr attr
);
106 void ac_add_func_attributes(LLVMContextRef ctx
, LLVMValueRef function
, unsigned attrib_mask
);
107 void ac_dump_module(LLVMModuleRef module
);
109 LLVMValueRef
ac_llvm_get_called_value(LLVMValueRef call
);
110 bool ac_llvm_is_function(LLVMValueRef v
);
111 LLVMModuleRef
ac_create_module(LLVMTargetMachineRef tm
, LLVMContextRef ctx
);
113 LLVMBuilderRef
ac_create_builder(LLVMContextRef ctx
, enum ac_float_mode float_mode
);
114 void ac_enable_signed_zeros(struct ac_llvm_context
*ctx
);
115 void ac_disable_signed_zeros(struct ac_llvm_context
*ctx
);
117 void ac_llvm_add_target_dep_function_attr(LLVMValueRef F
, const char *name
, unsigned value
);
118 void ac_llvm_set_workgroup_size(LLVMValueRef F
, unsigned size
);
120 static inline unsigned ac_get_load_intr_attribs(bool can_speculate
)
122 /* READNONE means writes can't affect it, while READONLY means that
123 * writes can affect it. */
124 return can_speculate
? AC_FUNC_ATTR_READNONE
: AC_FUNC_ATTR_READONLY
;
127 unsigned ac_count_scratch_private_memory(LLVMValueRef function
);
129 LLVMTargetLibraryInfoRef
ac_create_target_library_info(const char *triple
);
130 void ac_dispose_target_library_info(LLVMTargetLibraryInfoRef library_info
);
131 void ac_init_shared_llvm_once(void); /* Do not use directly, use ac_init_llvm_once */
132 void ac_init_llvm_once(void);
134 bool ac_init_llvm_compiler(struct ac_llvm_compiler
*compiler
, enum radeon_family family
,
135 enum ac_target_machine_options tm_options
);
136 void ac_destroy_llvm_compiler(struct ac_llvm_compiler
*compiler
);
138 struct ac_compiler_passes
*ac_create_llvm_passes(LLVMTargetMachineRef tm
);
139 void ac_destroy_llvm_passes(struct ac_compiler_passes
*p
);
140 bool ac_compile_module_to_elf(struct ac_compiler_passes
*p
, LLVMModuleRef module
,
141 char **pelf_buffer
, size_t *pelf_size
);
142 void ac_llvm_add_barrier_noop_pass(LLVMPassManagerRef passmgr
);
143 void ac_enable_global_isel(LLVMTargetMachineRef tm
);
145 static inline bool ac_has_vec3_support(enum chip_class chip
, bool use_format
)
147 if (chip
== GFX6
&& !use_format
) {
148 /* GFX6 only supports vec3 with load/store format. */
152 return LLVM_VERSION_MAJOR
>= 9;
159 #endif /* AC_LLVM_UTIL_H */