aco: add various compiler statistics

Adds these statistics:
- hash of code and constant data
- number of instructions
- number of copies from pseudo-instructions
- number of branches
- estimate of cycles spent not waiting in s_waitcnt
- number of vmem/smem "clauses"
- sgpr/vgpr usage before scheduling

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Acked-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/2965>

src/amd/compiler/aco_interface.cpp

@@ -54,6 +54,18 @@ static void init()
 }
 }
 
+static radv_compiler_statistic_info statistic_infos[] = {
+   [aco::statistic_hash] = {"Hash", "CRC32 hash of code and constant data"},
+   [aco::statistic_instructions] = {"Instructions", "Instruction count"},
+   [aco::statistic_copies] = {"Copies", "Copy instructions created for pseudo-instructions"},
+   [aco::statistic_branches] = {"Branches", "Branch instructions"},
+   [aco::statistic_cycles] = {"Busy Cycles", "Estimate of busy cycles"},
+   [aco::statistic_vmem_clauses] = {"VMEM Clause", "Number of VMEM clauses (includes 1-sized clauses)"},
+   [aco::statistic_smem_clauses] = {"SMEM Clause", "Number of SMEM clauses (includes 1-sized clauses)"},
+   [aco::statistic_sgpr_presched] = {"Pre-Sched SGPRs", "SGPR usage before scheduling"},
+   [aco::statistic_vgpr_presched] = {"Pre-Sched VGPRs", "VGPR usage before scheduling"},
+};
+
 void aco_compile_shader(unsigned shader_count,
                         struct nir_shader *const *shaders,
                         struct radv_shader_binary **binary,
@@ -64,6 +76,10 @@ void aco_compile_shader(unsigned shader_count,
    ac_shader_config config = {0};
    std::unique_ptr<aco::Program> program{new aco::Program};
 
+   program->collect_statistics = args->options->record_ir;
+   if (program->collect_statistics)
+      memset(program->statistics, 0, sizeof(program->statistics));
+
    /* Instruction Selection */
    if (args->is_gs_copy_shader)
       aco::select_gs_copy_shader(program.get(), shaders[0], &config, args);
@@ -94,6 +110,9 @@ void aco_compile_shader(unsigned shader_count,
    aco::live live_vars = aco::live_var_analysis(program.get(), args->options);
    aco::spill(program.get(), live_vars, args->options);
 
+   if (program->collect_statistics)
+      aco::collect_presched_stats(program.get());
+
    //std::cerr << "Before Schedule:\n";
    //aco_print_program(program.get(), stderr);
    aco::schedule_program(program.get(), live_vars);
@@ -139,10 +158,16 @@ void aco_compile_shader(unsigned shader_count,
    //std::cerr << "After Insert-Waitcnt:\n";
    //aco_print_program(program.get(), stderr);
 
+   if (program->collect_statistics)
+      aco::collect_preasm_stats(program.get());
+
    /* Assembly */
    std::vector<uint32_t> code;
    unsigned exec_size = aco::emit_program(program.get(), code);
 
+   if (program->collect_statistics)
+      aco::collect_postasm_stats(program.get(), code);
+
    bool get_disasm = args->options->dump_shader || args->options->record_ir;
 
    size_t size = llvm_ir.size();
@@ -156,6 +181,11 @@ void aco_compile_shader(unsigned shader_count,
       size += disasm.size();
    }
 
+   size_t stats_size = 0;
+   if (program->collect_statistics)
+      stats_size = sizeof(radv_compiler_statistics) + aco::num_statistics * sizeof(uint32_t);
+   size += stats_size;
+
    size += code.size() * sizeof(uint32_t) + sizeof(radv_shader_binary_legacy);
    /* We need to calloc to prevent unintialized data because this will be used
     * directly for the disk cache. Uninitialized data can appear because of
@@ -168,9 +198,15 @@ void aco_compile_shader(unsigned shader_count,
    legacy_binary->base.is_gs_copy_shader = args->is_gs_copy_shader;
    legacy_binary->base.total_size = size;
 
-   legacy_binary->stats_size = 0;
+   if (program->collect_statistics) {
+      radv_compiler_statistics *statistics = (radv_compiler_statistics *)legacy_binary->data;
+      statistics->count = aco::num_statistics;
+      statistics->infos = statistic_infos;
+      memcpy(statistics->values, program->statistics, aco::num_statistics * sizeof(uint32_t));
+   }
+   legacy_binary->stats_size = stats_size;
 
-   memcpy(legacy_binary->data, code.data(), code.size() * sizeof(uint32_t));
+   memcpy(legacy_binary->data + legacy_binary->stats_size, code.data(), code.size() * sizeof(uint32_t));
    legacy_binary->exec_size = exec_size;
    legacy_binary->code_size = code.size() * sizeof(uint32_t);
 
@@ -178,10 +214,10 @@ void aco_compile_shader(unsigned shader_count,
    legacy_binary->disasm_size = 0;
    legacy_binary->ir_size = llvm_ir.size();
 
-   llvm_ir.copy((char*) legacy_binary->data + legacy_binary->code_size, llvm_ir.size());
+   llvm_ir.copy((char*) legacy_binary->data + legacy_binary->stats_size + legacy_binary->code_size, llvm_ir.size());
 
    if (get_disasm) {
-      disasm.copy((char*) legacy_binary->data + legacy_binary->code_size + llvm_ir.size(), disasm.size());
+      disasm.copy((char*) legacy_binary->data + legacy_binary->stats_size + legacy_binary->code_size + llvm_ir.size(), disasm.size());
       legacy_binary->disasm_size = disasm.size();
    }