/*
* pipe_context
*/
static void si_destroy_context(struct pipe_context *context)
{
struct si_context *sctx = (struct si_context *)context;
int i;
si_dec_framebuffer_counters(&sctx->framebuffer.state);
si_release_all_descriptors(sctx);
if (sctx->ce_suballocator)
u_suballocator_destroy(sctx->ce_suballocator);
pipe_resource_reference(&sctx->esgs_ring, NULL);
pipe_resource_reference(&sctx->gsvs_ring, NULL);
pipe_resource_reference(&sctx->tf_ring, NULL);
pipe_resource_reference(&sctx->tess_offchip_ring, NULL);
pipe_resource_reference(&sctx->null_const_buf.buffer, NULL);
r600_resource_reference(&sctx->border_color_buffer, NULL);
free(sctx->border_color_table);
r600_resource_reference(&sctx->scratch_buffer, NULL);
r600_resource_reference(&sctx->compute_scratch_buffer, NULL);
sctx->b.ws->fence_reference(&sctx->last_gfx_fence, NULL);
si_pm4_free_state(sctx, sctx->init_config, ~0);
if (sctx->init_config_gs_rings)
si_pm4_free_state(sctx, sctx->init_config_gs_rings, ~0);
for (i = 0; i < ARRAY_SIZE(sctx->vgt_shader_config); i++)
si_pm4_delete_state(sctx, vgt_shader_config, sctx->vgt_shader_config[i]);
if (sctx->fixed_func_tcs_shader.cso)
sctx->b.b.delete_tcs_state(&sctx->b.b, sctx->fixed_func_tcs_shader.cso);
if (sctx->custom_dsa_flush)
sctx->b.b.delete_depth_stencil_alpha_state(&sctx->b.b, sctx->custom_dsa_flush);
if (sctx->custom_blend_resolve)
sctx->b.b.delete_blend_state(&sctx->b.b, sctx->custom_blend_resolve);
if (sctx->custom_blend_decompress)
sctx->b.b.delete_blend_state(&sctx->b.b, sctx->custom_blend_decompress);
if (sctx->custom_blend_fastclear)
sctx->b.b.delete_blend_state(&sctx->b.b, sctx->custom_blend_fastclear);
if (sctx->custom_blend_dcc_decompress)
sctx->b.b.delete_blend_state(&sctx->b.b, sctx->custom_blend_dcc_decompress);
util_unreference_framebuffer_state(&sctx->framebuffer.state);
if (sctx->blitter)
util_blitter_destroy(sctx->blitter);
r600_common_context_cleanup(&sctx->b);
LLVMDisposeTargetMachine(sctx->tm);
r600_resource_reference(&sctx->trace_buf, NULL);
r600_resource_reference(&sctx->last_trace_buf, NULL);
free(sctx->last_ib);
if (sctx->last_bo_list) {
for (i = 0; i < sctx->last_bo_count; i++)
pb_reference(&sctx->last_bo_list[i].buf, NULL);
free(sctx->last_bo_list);
}
FREE(sctx);
}
/*
* pipe_context
*/
static void si_destroy_context(struct pipe_context *context)
{
struct si_context *sctx = (struct si_context *)context;
int i;
/* Unreference the framebuffer normally to disable related logic
* properly.
*/
struct pipe_framebuffer_state fb = {};
context->set_framebuffer_state(context, &fb);
si_release_all_descriptors(sctx);
if (sctx->ce_suballocator)
u_suballocator_destroy(sctx->ce_suballocator);
pipe_resource_reference(&sctx->esgs_ring, NULL);
pipe_resource_reference(&sctx->gsvs_ring, NULL);
pipe_resource_reference(&sctx->tf_ring, NULL);
pipe_resource_reference(&sctx->tess_offchip_ring, NULL);
pipe_resource_reference(&sctx->null_const_buf.buffer, NULL);
r600_resource_reference(&sctx->border_color_buffer, NULL);
free(sctx->border_color_table);
r600_resource_reference(&sctx->scratch_buffer, NULL);
r600_resource_reference(&sctx->compute_scratch_buffer, NULL);
si_pm4_free_state(sctx, sctx->init_config, ~0);
if (sctx->init_config_gs_rings)
si_pm4_free_state(sctx, sctx->init_config_gs_rings, ~0);
for (i = 0; i < ARRAY_SIZE(sctx->vgt_shader_config); i++)
si_pm4_delete_state(sctx, vgt_shader_config, sctx->vgt_shader_config[i]);
if (sctx->fixed_func_tcs_shader.cso)
sctx->b.b.delete_tcs_state(&sctx->b.b, sctx->fixed_func_tcs_shader.cso);
if (sctx->custom_dsa_flush)
sctx->b.b.delete_depth_stencil_alpha_state(&sctx->b.b, sctx->custom_dsa_flush);
if (sctx->custom_blend_resolve)
sctx->b.b.delete_blend_state(&sctx->b.b, sctx->custom_blend_resolve);
if (sctx->custom_blend_decompress)
sctx->b.b.delete_blend_state(&sctx->b.b, sctx->custom_blend_decompress);
if (sctx->custom_blend_fastclear)
sctx->b.b.delete_blend_state(&sctx->b.b, sctx->custom_blend_fastclear);
if (sctx->custom_blend_dcc_decompress)
sctx->b.b.delete_blend_state(&sctx->b.b, sctx->custom_blend_dcc_decompress);
if (sctx->blitter)
util_blitter_destroy(sctx->blitter);
r600_common_context_cleanup(&sctx->b);
LLVMDisposeTargetMachine(sctx->tm);
r600_resource_reference(&sctx->trace_buf, NULL);
r600_resource_reference(&sctx->last_trace_buf, NULL);
radeon_clear_saved_cs(&sctx->last_gfx);
FREE(sctx);
}
static void codegen_cleanup(compile_t* c)
{
while(c->frame != NULL)
pop_frame(c);
LLVMDisposeBuilder(c->builder);
LLVMDisposeModule(c->module);
LLVMContextDispose(c->context);
LLVMDisposeTargetMachine(c->machine);
reach_free(c->reachable);
}
static void codegen_cleanup(compile_t* c)
{
while(c->frame != NULL)
pop_frame(c);
LLVMDIBuilderDestroy(c->di);
LLVMDisposeBuilder(c->builder);
LLVMDisposeModule(c->module);
LLVMContextDispose(c->context);
LLVMDisposeTargetMachine(c->machine);
tbaa_metadatas_free(c->tbaa_mds);
reach_free(c->reach);
}
/*
* pipe_context
*/
static void si_destroy_context(struct pipe_context *context)
{
struct si_context *sctx = (struct si_context *)context;
int i;
si_release_all_descriptors(sctx);
pipe_resource_reference(&sctx->esgs_ring, NULL);
pipe_resource_reference(&sctx->gsvs_ring, NULL);
pipe_resource_reference(&sctx->tf_ring, NULL);
pipe_resource_reference(&sctx->null_const_buf.buffer, NULL);
r600_resource_reference(&sctx->border_color_table, NULL);
r600_resource_reference(&sctx->scratch_buffer, NULL);
sctx->b.ws->fence_reference(&sctx->last_gfx_fence, NULL);
si_pm4_free_state(sctx, sctx->init_config, ~0);
si_pm4_delete_state(sctx, gs_rings, sctx->gs_rings);
si_pm4_delete_state(sctx, tf_ring, sctx->tf_state);
for (i = 0; i < Elements(sctx->vgt_shader_config); i++)
si_pm4_delete_state(sctx, vgt_shader_config, sctx->vgt_shader_config[i]);
if (sctx->pstipple_sampler_state)
sctx->b.b.delete_sampler_state(&sctx->b.b, sctx->pstipple_sampler_state);
if (sctx->dummy_pixel_shader)
sctx->b.b.delete_fs_state(&sctx->b.b, sctx->dummy_pixel_shader);
if (sctx->fixed_func_tcs_shader)
sctx->b.b.delete_tcs_state(&sctx->b.b, sctx->fixed_func_tcs_shader);
if (sctx->custom_dsa_flush)
sctx->b.b.delete_depth_stencil_alpha_state(&sctx->b.b, sctx->custom_dsa_flush);
if (sctx->custom_blend_resolve)
sctx->b.b.delete_blend_state(&sctx->b.b, sctx->custom_blend_resolve);
if (sctx->custom_blend_decompress)
sctx->b.b.delete_blend_state(&sctx->b.b, sctx->custom_blend_decompress);
if (sctx->custom_blend_fastclear)
sctx->b.b.delete_blend_state(&sctx->b.b, sctx->custom_blend_fastclear);
util_unreference_framebuffer_state(&sctx->framebuffer.state);
if (sctx->blitter)
util_blitter_destroy(sctx->blitter);
si_pm4_cleanup(sctx);
r600_common_context_cleanup(&sctx->b);
#if HAVE_LLVM >= 0x0306
LLVMDisposeTargetMachine(sctx->tm);
#endif
FREE(sctx);
}
/*
* machine_finish
*
* Context cleanup.
*/
void
machine_finish (machinedef_t *mach)
{
machine_ctx_t m;
if (mach == 0) return;
m = machine_context(mach);
if (m == 0) return;
if (m->target_machine != 0) LLVMDisposeTargetMachine(m->target_machine);
if (m->llvmctx != 0) LLVMContextDispose(m->llvmctx);
if (m->outfile != 0) free(m->outfile);
free(m);
} /* machine_finish */
static void si_destroy_screen(struct pipe_screen* pscreen)
{
struct si_screen *sscreen = (struct si_screen *)pscreen;
struct si_shader_part *parts[] = {
sscreen->vs_prologs,
sscreen->vs_epilogs,
sscreen->tcs_epilogs,
sscreen->gs_prologs,
sscreen->ps_prologs,
sscreen->ps_epilogs
};
unsigned i;
if (!sscreen)
return;
if (!sscreen->b.ws->unref(sscreen->b.ws))
return;
if (util_queue_is_initialized(&sscreen->shader_compiler_queue))
util_queue_destroy(&sscreen->shader_compiler_queue);
for (i = 0; i < ARRAY_SIZE(sscreen->tm); i++)
if (sscreen->tm[i])
LLVMDisposeTargetMachine(sscreen->tm[i]);
/* Free shader parts. */
for (i = 0; i < ARRAY_SIZE(parts); i++) {
while (parts[i]) {
struct si_shader_part *part = parts[i];
parts[i] = part->next;
radeon_shader_binary_clean(&part->binary);
FREE(part);
}
}
pipe_mutex_destroy(sscreen->shader_parts_mutex);
si_destroy_shader_cache(sscreen);
r600_destroy_common_screen(&sscreen->b);
}
/**
* Compile an LLVM module to machine code.
*
* @returns 0 for success, 1 for failure
*/
unsigned radeon_llvm_compile(LLVMModuleRef M, struct radeon_shader_binary *binary,
const char *gpu_family, unsigned dump, LLVMTargetMachineRef tm)
{
char cpu[CPU_STRING_LEN];
char fs[FS_STRING_LEN];
char *err;
bool dispose_tm = false;
LLVMContextRef llvm_ctx;
unsigned rval = 0;
LLVMMemoryBufferRef out_buffer;
unsigned buffer_size;
const char *buffer_data;
char triple[TRIPLE_STRING_LEN];
LLVMBool mem_err;
if (!tm) {
strncpy(triple, "r600--", TRIPLE_STRING_LEN);
LLVMTargetRef target = radeon_llvm_get_r600_target(triple);
if (!target) {
return 1;
}
strncpy(cpu, gpu_family, CPU_STRING_LEN);
memset(fs, 0, sizeof(fs));
if (dump) {
strncpy(fs, "+DumpCode", FS_STRING_LEN);
}
tm = LLVMCreateTargetMachine(target, triple, cpu, fs,
LLVMCodeGenLevelDefault, LLVMRelocDefault,
LLVMCodeModelDefault);
dispose_tm = true;
}
if (dump) {
LLVMDumpModule(M);
}
/* Setup Diagnostic Handler*/
llvm_ctx = LLVMGetModuleContext(M);
#if HAVE_LLVM >= 0x0305
LLVMContextSetDiagnosticHandler(llvm_ctx, radeonDiagnosticHandler, &rval);
#endif
rval = 0;
/* Compile IR*/
mem_err = LLVMTargetMachineEmitToMemoryBuffer(tm, M, LLVMObjectFile, &err,
&out_buffer);
/* Process Errors/Warnings */
if (mem_err) {
fprintf(stderr, "%s: %s", __FUNCTION__, err);
FREE(err);
LLVMDisposeTargetMachine(tm);
return 1;
}
if (0 != rval) {
fprintf(stderr, "%s: Processing Diag Flag\n", __FUNCTION__);
}
/* Extract Shader Code*/
buffer_size = LLVMGetBufferSize(out_buffer);
buffer_data = LLVMGetBufferStart(out_buffer);
radeon_elf_read(buffer_data, buffer_size, binary, dump);
/* Clean up */
LLVMDisposeMemoryBuffer(out_buffer);
if (dispose_tm) {
LLVMDisposeTargetMachine(tm);
}
return rval;
}
static void llvm_finalize_target_machine(value Machine) {
LLVMDisposeTargetMachine(TargetMachine_val(Machine));
}
/**
* Compile an LLVM module to machine code.
*
* @returns 0 for success, 1 for failure
*/
unsigned radeon_llvm_compile(LLVMModuleRef M, struct radeon_llvm_binary *binary,
const char * gpu_family, unsigned dump) {
LLVMTargetRef target;
LLVMTargetMachineRef tm;
char cpu[CPU_STRING_LEN];
char fs[FS_STRING_LEN];
char *err;
LLVMMemoryBufferRef out_buffer;
unsigned buffer_size;
const char *buffer_data;
char triple[TRIPLE_STRING_LEN];
char *elf_buffer;
Elf *elf;
Elf_Scn *section = NULL;
size_t section_str_index;
LLVMBool r;
init_r600_target();
target = get_r600_target();
if (!target) {
return 1;
}
strncpy(cpu, gpu_family, CPU_STRING_LEN);
memset(fs, 0, sizeof(fs));
if (dump) {
LLVMDumpModule(M);
strncpy(fs, "+DumpCode", FS_STRING_LEN);
}
strncpy(triple, "r600--", TRIPLE_STRING_LEN);
tm = LLVMCreateTargetMachine(target, triple, cpu, fs,
LLVMCodeGenLevelDefault, LLVMRelocDefault,
LLVMCodeModelDefault);
r = LLVMTargetMachineEmitToMemoryBuffer(tm, M, LLVMObjectFile, &err,
&out_buffer);
if (r) {
fprintf(stderr, "%s", err);
FREE(err);
return 1;
}
buffer_size = LLVMGetBufferSize(out_buffer);
buffer_data = LLVMGetBufferStart(out_buffer);
/* One of the libelf implementations
* (http://www.mr511.de/software/english.htm) requires calling
* elf_version() before elf_memory().
*/
elf_version(EV_CURRENT);
elf_buffer = MALLOC(buffer_size);
memcpy(elf_buffer, buffer_data, buffer_size);
elf = elf_memory(elf_buffer, buffer_size);
elf_getshdrstrndx(elf, §ion_str_index);
binary->disassembled = 0;
while ((section = elf_nextscn(elf, section))) {
const char *name;
Elf_Data *section_data = NULL;
GElf_Shdr section_header;
if (gelf_getshdr(section, §ion_header) != §ion_header) {
fprintf(stderr, "Failed to read ELF section header\n");
return 1;
}
name = elf_strptr(elf, section_str_index, section_header.sh_name);
if (!strcmp(name, ".text")) {
section_data = elf_getdata(section, section_data);
binary->code_size = section_data->d_size;
binary->code = MALLOC(binary->code_size * sizeof(unsigned char));
memcpy(binary->code, section_data->d_buf, binary->code_size);
} else if (!strcmp(name, ".AMDGPU.config")) {
section_data = elf_getdata(section, section_data);
binary->config_size = section_data->d_size;
binary->config = MALLOC(binary->config_size * sizeof(unsigned char));
memcpy(binary->config, section_data->d_buf, binary->config_size);
} else if (dump && !strcmp(name, ".AMDGPU.disasm")) {
binary->disassembled = 1;
section_data = elf_getdata(section, section_data);
fprintf(stderr, "\nShader Disassembly:\n\n");
fprintf(stderr, "%.*s\n", (int)section_data->d_size,
(char *)section_data->d_buf);
}
}
LLVMDisposeMemoryBuffer(out_buffer);
LLVMDisposeTargetMachine(tm);
return 0;
}
/*
* machine_init
*
* Initializes the machine context.
*/
machinedef_t *
machine_init (const char *mspec)
{
machine_ctx_t m;
machinedef_t *mach;
char *err;
LLVMTargetRef target;
char *machspec = (char *) mspec;
unsigned long allosize;
if (machspec == 0) {
machspec = LLVM_DEFAULT_TARGET_TRIPLE;
}
allosize = (sizeof(struct machine_ctx_s) +
sizeof(struct machinedef_s) +
strlen(machspec) + 1);
m = malloc(allosize);
if (m == 0) return 0;
memset(m, 0, allosize);
m->triple = ((char *) m) + (sizeof(struct machine_ctx_s) +
sizeof(struct machinedef_s));
memcpy(m->triple, machspec, strlen(machspec));
LLVM_NATIVE_TARGETINFO();
LLVM_NATIVE_TARGET();
LLVM_NATIVE_TARGETMC();
LLVM_NATIVE_ASMPRINTER();
LLVM_NATIVE_ASMPARSER();
err = 0;
target = HelperLookupTarget(machspec, &err);
if (target == 0) {
if (err != 0) free(err);
free(m);
return 0;
}
m->target_machine = LLVMCreateTargetMachine(target, (char *)machspec, "", "",
LLVMCodeGenLevelDefault,
LLVMRelocPIC, LLVMCodeModelDefault);
if (m->target_machine == 0) {
free(m);
return 0;
}
HelperSetAsmVerbosity(m->target_machine, 1);
m->llvmctx = LLVMContextCreate();
if (m->llvmctx == 0) {
LLVMDisposeTargetMachine(m->target_machine);
free(m);
return 0;
}
m->is_macho = (strstr(machspec, "darwin") != 0); // XXX
mach = (machinedef_t *)(m + 1);
mach->machctx = m;
mach->bpaddr = sizeof(int *) * 8;
mach->bpval = sizeof(long) * 8;
mach->bpunit = 8;
mach->charsize_count = 1;
mach->charsizes[0] = 8;
mach->flags = MACH_M_SIGNEXT | MACH_M_LTC_INIT;
mach->max_align = 4;
mach->reg_count = 16;
return mach;
} /* machine_init */
/**
* Compile an LLVM module to machine code.
*
* @returns 0 for success, 1 for failure
*/
unsigned radeon_llvm_compile(LLVMModuleRef M, struct radeon_shader_binary *binary,
const char *gpu_family,
LLVMTargetMachineRef tm,
struct pipe_debug_callback *debug)
{
struct radeon_llvm_diagnostics diag;
char cpu[CPU_STRING_LEN];
char fs[FS_STRING_LEN];
char *err;
bool dispose_tm = false;
LLVMContextRef llvm_ctx;
LLVMMemoryBufferRef out_buffer;
unsigned buffer_size;
const char *buffer_data;
char triple[TRIPLE_STRING_LEN];
LLVMBool mem_err;
diag.debug = debug;
diag.retval = 0;
if (!tm) {
strncpy(triple, "r600--", TRIPLE_STRING_LEN);
LLVMTargetRef target = radeon_llvm_get_r600_target(triple);
if (!target) {
return 1;
}
strncpy(cpu, gpu_family, CPU_STRING_LEN);
memset(fs, 0, sizeof(fs));
strncpy(fs, "+DumpCode", FS_STRING_LEN);
tm = LLVMCreateTargetMachine(target, triple, cpu, fs,
LLVMCodeGenLevelDefault, LLVMRelocDefault,
LLVMCodeModelDefault);
dispose_tm = true;
}
/* Setup Diagnostic Handler*/
llvm_ctx = LLVMGetModuleContext(M);
LLVMContextSetDiagnosticHandler(llvm_ctx, radeonDiagnosticHandler, &diag);
/* Compile IR*/
mem_err = LLVMTargetMachineEmitToMemoryBuffer(tm, M, LLVMObjectFile, &err,
&out_buffer);
/* Process Errors/Warnings */
if (mem_err) {
fprintf(stderr, "%s: %s", __FUNCTION__, err);
pipe_debug_message(debug, SHADER_INFO,
"LLVM emit error: %s", err);
FREE(err);
diag.retval = 1;
goto out;
}
/* Extract Shader Code*/
buffer_size = LLVMGetBufferSize(out_buffer);
buffer_data = LLVMGetBufferStart(out_buffer);
radeon_elf_read(buffer_data, buffer_size, binary);
/* Clean up */
LLVMDisposeMemoryBuffer(out_buffer);
out:
if (dispose_tm) {
LLVMDisposeTargetMachine(tm);
}
if (diag.retval != 0)
pipe_debug_message(debug, SHADER_INFO, "LLVM compile failed");
return diag.retval;
}