aco: change thread_local memory resource to pointer

Apparently the TLS constructor doesn't work well if RADV
is instantiated multiple times and/or used by a program with
already existing threads.

Fixes: a128d444cb ('aco: use monotonic_buffer_resource for instructions')
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19219>

src/amd/compiler/aco_interface.cpp

@@ -257,8 +257,6 @@ aco_compile_shader(const struct aco_compiler_options* options,
    if (program->collect_statistics)
       stats_size = aco::num_statistics * sizeof(uint32_t);
 
-   aco::instruction_buffer.release();
-
    (*build_binary)(binary,
                    shaders[shader_count - 1]->info.stage,
                    args->is_gs_copy_shader,
@@ -310,8 +308,6 @@ aco_compile_vs_prolog(const struct aco_compiler_options* options,
    if (get_disasm)
       disasm = get_disasm_string(program.get(), code, exec_size);
 
-   aco::instruction_buffer.release();
-
    (*build_prolog)(binary,
                    config.num_sgprs,
                    config.num_vgprs,
@@ -357,8 +353,6 @@ aco_compile_ps_epilog(const struct aco_compiler_options* options,
    if (get_disasm)
       disasm = get_disasm_string(program.get(), code, exec_size);
 
-   aco::instruction_buffer.release();
-
    (*build_epilog)(binary,
                    config.num_sgprs,
                    config.num_vgprs,

src/amd/compiler/aco_ir.cpp

@@ -32,7 +32,7 @@
 
 namespace aco {
 
-thread_local aco::monotonic_buffer_resource instruction_buffer(65536);
+thread_local aco::monotonic_buffer_resource* instruction_buffer = nullptr;
 
 uint64_t debug_flags = 0;
 
@@ -76,6 +76,7 @@ init_program(Program* program, Stage stage, const struct aco_shader_info* info,
              enum amd_gfx_level gfx_level, enum radeon_family family, bool wgp_mode,
              ac_shader_config* config)
 {
+   instruction_buffer = &program->m;
    program->stage = stage;
    program->config = config;
    program->info = *info;

src/amd/compiler/aco_ir.h

@@ -1778,7 +1778,7 @@ struct Pseudo_reduction_instruction : public Instruction {
 static_assert(sizeof(Pseudo_reduction_instruction) == sizeof(Instruction) + 4,
               "Unexpected padding");
 
-extern thread_local aco::monotonic_buffer_resource instruction_buffer;
+extern thread_local aco::monotonic_buffer_resource* instruction_buffer;
 
 struct instr_deleter_functor {
    /* Don't yet free any instructions. They will be de-allocated
@@ -1796,7 +1796,7 @@ create_instruction(aco_opcode opcode, Format format, uint32_t num_operands,
 {
    std::size_t size =
       sizeof(T) + num_operands * sizeof(Operand) + num_definitions * sizeof(Definition);
-   void* data = instruction_buffer.allocate(size, alignof(uint32_t));
+   void* data = instruction_buffer->allocate(size, alignof(uint32_t));
    memset(data, 0, size);
    T* inst = (T*)data;
 
@@ -2155,6 +2155,7 @@ enum class CompilationProgress {
 
 class Program final {
 public:
+   aco::monotonic_buffer_resource m{65536};
    std::vector<Block> blocks;
    std::vector<RegClass> temp_rc = {s1};
    RegisterDemand max_reg_demand = RegisterDemand();