/**
* Create a new pipe_screen object
* Note: we're not presently subclassing pipe_screen (no llvmpipe_screen).
*/
struct pipe_screen *
llvmpipe_create_screen(struct sw_winsys *winsys)
{
struct llvmpipe_screen *screen;
util_cpu_detect();
#ifdef DEBUG
LP_DEBUG = debug_get_flags_option("LP_DEBUG", lp_debug_flags, 0 );
#endif
LP_PERF = debug_get_flags_option("LP_PERF", lp_perf_flags, 0 );
screen = CALLOC_STRUCT(llvmpipe_screen);
if (!screen)
return NULL;
if (!lp_jit_screen_init(screen)) {
FREE(screen);
return NULL;
}
screen->winsys = winsys;
screen->base.destroy = llvmpipe_destroy_screen;
screen->base.get_name = llvmpipe_get_name;
screen->base.get_vendor = llvmpipe_get_vendor;
screen->base.get_param = llvmpipe_get_param;
screen->base.get_shader_param = llvmpipe_get_shader_param;
screen->base.get_paramf = llvmpipe_get_paramf;
screen->base.is_format_supported = llvmpipe_is_format_supported;
screen->base.context_create = llvmpipe_create_context;
screen->base.flush_frontbuffer = llvmpipe_flush_frontbuffer;
screen->base.fence_reference = llvmpipe_fence_reference;
screen->base.fence_signalled = llvmpipe_fence_signalled;
screen->base.fence_finish = llvmpipe_fence_finish;
screen->base.get_timestamp = llvmpipe_get_timestamp;
llvmpipe_init_screen_resource_funcs(&screen->base);
screen->num_threads = util_cpu_caps.nr_cpus > 1 ? util_cpu_caps.nr_cpus : 0;
#ifdef PIPE_SUBSYSTEM_EMBEDDED
screen->num_threads = 0;
#endif
screen->num_threads = debug_get_num_option("LP_NUM_THREADS", screen->num_threads);
screen->num_threads = MIN2(screen->num_threads, LP_MAX_THREADS);
screen->rast = lp_rast_create(screen->num_threads);
if (!screen->rast) {
lp_jit_screen_cleanup(screen);
FREE(screen);
return NULL;
}
pipe_mutex_init(screen->rast_mutex);
util_format_s3tc_init();
return &screen->base;
}
void
lp_build_init(void)
{
#ifdef DEBUG
gallivm_debug = debug_get_option_gallivm_debug();
#endif
lp_set_target_options();
LLVMInitializeNativeTarget();
LLVMLinkInJIT();
if (!lp_build_module)
lp_build_module = LLVMModuleCreateWithName("gallivm");
if (!lp_build_provider)
lp_build_provider = LLVMCreateModuleProviderForExistingModule(lp_build_module);
if (!lp_build_engine) {
enum LLVM_CodeGenOpt_Level optlevel;
char *error = NULL;
if (gallivm_debug & GALLIVM_DEBUG_NO_OPT) {
optlevel = None;
}
else {
optlevel = Default;
}
if (LLVMCreateJITCompiler(&lp_build_engine, lp_build_provider,
(unsigned)optlevel, &error)) {
_debug_printf("%s\n", error);
LLVMDisposeMessage(error);
assert(0);
}
#if defined(DEBUG) || defined(PROFILE)
lp_register_oprofile_jit_event_listener(lp_build_engine);
#endif
}
if (!lp_build_target)
lp_build_target = LLVMGetExecutionEngineTargetData(lp_build_engine);
if (!lp_build_pass) {
lp_build_pass = LLVMCreateFunctionPassManager(lp_build_provider);
LLVMAddTargetData(lp_build_target, lp_build_pass);
if ((gallivm_debug & GALLIVM_DEBUG_NO_OPT) == 0) {
/* These are the passes currently listed in llvm-c/Transforms/Scalar.h,
* but there are more on SVN. */
/* TODO: Add more passes */
LLVMAddCFGSimplificationPass(lp_build_pass);
LLVMAddPromoteMemoryToRegisterPass(lp_build_pass);
LLVMAddConstantPropagationPass(lp_build_pass);
if(util_cpu_caps.has_sse4_1) {
/* FIXME: There is a bug in this pass, whereby the combination of fptosi
* and sitofp (necessary for trunc/floor/ceil/round implementation)
* somehow becomes invalid code.
*/
LLVMAddInstructionCombiningPass(lp_build_pass);
}
LLVMAddGVNPass(lp_build_pass);
} else {
/* We need at least this pass to prevent the backends to fail in
* unexpected ways.
*/
LLVMAddPromoteMemoryToRegisterPass(lp_build_pass);
}
}
util_cpu_detect();
#if 0
/* For simulating less capable machines */
util_cpu_caps.has_sse3 = 0;
util_cpu_caps.has_ssse3 = 0;
util_cpu_caps.has_sse4_1 = 0;
#endif
}
