import sys
import textwrap
-import xml.etree.cElementTree as et
+import xml.etree.ElementTree as et
hashed_funcs = {}
def emit_udiv(tmp_id, args):
c("uint64_t tmp{0} = {1};".format(tmp_id, args[1]))
c("uint64_t tmp{0} = {1};".format(tmp_id + 1, args[0]))
- c("uint64_t tmp{0} = tmp{1} ? tmp{2} / tmp{1} : 0;".format(tmp_id + 2, tmp_id + 1, tmp_id))
+ if args[0].isdigit():
+ assert int(args[0]) > 0
+ c("uint64_t tmp{0} = tmp{2} / tmp{1};".format(tmp_id + 2, tmp_id + 1, tmp_id))
+ else:
+ c("uint64_t tmp{0} = tmp{1} ? tmp{2} / tmp{1} : 0;".format(tmp_id + 2, tmp_id + 1, tmp_id))
return tmp_id + 3
def emit_umul(tmp_id, args):
hw_vars["$EuSubslicesTotalCount"] = "perf->sys_vars.n_eu_sub_slices"
hw_vars["$EuThreadsCount"] = "perf->sys_vars.eu_threads_count"
hw_vars["$SliceMask"] = "perf->sys_vars.slice_mask"
+# subslice_mask is interchangeable with subslice/dual-subslice since Gen12+
+# only has dual subslices which can be assimilated with 16EUs subslices.
hw_vars["$SubsliceMask"] = "perf->sys_vars.subslice_mask"
+hw_vars["$DualSubsliceMask"] = "perf->sys_vars.subslice_mask"
hw_vars["$GpuTimestampFrequency"] = "perf->sys_vars.timestamp_frequency"
hw_vars["$GpuMinFrequency"] = "perf->sys_vars.gt_min_freq"
hw_vars["$GpuMaxFrequency"] = "perf->sys_vars.gt_max_freq"
read_eq = counter.get('equation')
c("static " + ret_type)
- c(counter.read_sym + "(UNUSED struct gen_perf *perf,\n")
+ c(counter.read_sym + "(UNUSED struct gen_perf_config *perf,\n")
c_indent(len(counter.read_sym) + 1)
c("const struct gen_perf_query_info *query,\n")
c("const uint64_t *accumulator)\n")
ret_type = "uint64_t"
c("static " + ret_type)
- c(counter.max_sym() + "(struct gen_perf *perf)\n")
+ c(counter.max_sym() + "(struct gen_perf_config *perf)\n")
c("{")
c_indent(3)
output_rpn_equation_code(set, counter, max_eq)
c_outdent(4)
+def output_units(unit):
+ return unit.replace(' ', '_').upper()
+
+
def output_counter_report(set, counter, current_offset):
data_type = counter.get('data_type')
data_type_uc = data_type.upper()
c("counter->oa_counter_read_" + data_type + " = " + set.read_funcs[counter.get('symbol_name')] + ";\n")
c("counter->name = \"" + counter.get('name') + "\";\n")
c("counter->desc = \"" + counter.get('description') + "\";\n")
+ c("counter->symbol_name = \"" + counter.get('symbol_name') + "\";\n")
+ c("counter->category = \"" + counter.get('mdapi_group') + "\";\n")
c("counter->type = GEN_PERF_COUNTER_TYPE_" + semantic_type_uc + ";\n")
c("counter->data_type = GEN_PERF_COUNTER_DATA_TYPE_" + data_type_uc + ";\n")
+ c("counter->units = GEN_PERF_COUNTER_UNITS_" + output_units(counter.get('units')) + ";\n")
c("counter->raw_max = " + set.max_values[counter.get('symbol_name')] + ";\n")
current_offset = pot_align(current_offset, sizeof(c_type))
def generate_register_configs(set):
register_configs = set.findall('register_config')
+
for register_config in register_configs:
t = register_types[register_config.get('type')]
output_availability(set, availability, register_config.get('type') + ' register config')
c_indent(3)
- for register in register_config.findall('register'):
- c("query->%s[query->n_%s++] = (struct gen_perf_query_register_prog) { .reg = %s, .val = %s };" %
- (t, t, register.get('address'), register.get('value')))
+ registers = register_config.findall('register')
+ c("static const struct gen_perf_query_register_prog %s[] = {" % t)
+ c_indent(3)
+ for register in registers:
+ c("{ .reg = %s, .val = %s }," % (register.get('address'), register.get('value')))
+ c_outdent(3)
+ c("};")
+ c("query->config.%s = %s;" % (t, t))
+ c("query->config.n_%s = ARRAY_SIZE(%s);" % (t, t))
if availability:
c_outdent(3)
h(textwrap.dedent("""\
#pragma once
- struct gen_perf;
+ struct gen_perf_config;
"""))
#include <drm-uapi/i915_drm.h>
#include "util/hash_table.h"
+ #include "util/ralloc.h"
"""))
counters = set.counters
c("\n")
- register_lengths = compute_register_lengths(set);
- for reg_type, reg_length in register_lengths.items():
- c("static struct gen_perf_query_register_prog {0}_{1}_{2}[{3}];".format(gen.chipset,
- set.underscore_name,
- reg_type, reg_length))
-
- c("\nstatic struct gen_perf_query_counter {0}_{1}_query_counters[{2}];\n".format(gen.chipset, set.underscore_name, len(counters)))
- c("static struct gen_perf_query_info " + gen.chipset + "_" + set.underscore_name + "_query = {\n")
+ c("\nstatic void\n")
+ c("{0}_register_{1}_counter_query(struct gen_perf_config *perf)\n".format(gen.chipset, set.underscore_name))
+ c("{\n")
c_indent(3)
- c(".kind = GEN_PERF_QUERY_TYPE_OA,\n")
- c(".name = \"" + set.name + "\",\n")
- c(".guid = \"" + set.hw_config_guid + "\",\n")
+ c("struct gen_perf_query_info *query = rzalloc(perf, struct gen_perf_query_info);\n")
+ c("\n")
+ c("query->kind = GEN_PERF_QUERY_TYPE_OA;\n")
+ c("query->name = \"" + set.name + "\";\n")
+ c("query->symbol_name = \"" + set.symbol_name + "\";\n")
+ c("query->guid = \"" + set.hw_config_guid + "\";\n")
- c(".counters = {0}_{1}_query_counters,".format(gen.chipset, set.underscore_name))
- c(".n_counters = 0,")
- c(".oa_metrics_set_id = 0, /* determined at runtime, via sysfs */")
+ c("query->counters = rzalloc_array(query, struct gen_perf_query_counter, %u);" % len(counters))
+ c("query->n_counters = 0;")
+ c("query->oa_metrics_set_id = 0; /* determined at runtime, via sysfs */")
if gen.chipset == "hsw":
c(textwrap.dedent("""\
- .oa_format = I915_OA_FORMAT_A45_B8_C8,
-
+ query->oa_format = I915_OA_FORMAT_A45_B8_C8;
/* Accumulation buffer offsets... */
- .gpu_time_offset = 0,
- .a_offset = 1,
- .b_offset = 46,
- .c_offset = 54,
+ query->gpu_time_offset = 0;
+ query->a_offset = 1;
+ query->b_offset = 46;
+ query->c_offset = 54;
"""))
else:
c(textwrap.dedent("""\
- .oa_format = I915_OA_FORMAT_A32u40_A4u32_B8_C8,
-
+ query->oa_format = I915_OA_FORMAT_A32u40_A4u32_B8_C8;
/* Accumulation buffer offsets... */
- .gpu_time_offset = 0,
- .gpu_clock_offset = 1,
- .a_offset = 2,
- .b_offset = 38,
- .c_offset = 46,
+ query->gpu_time_offset = 0;
+ query->gpu_clock_offset = 1;
+ query->a_offset = 2;
+ query->b_offset = 38;
+ query->c_offset = 46;
"""))
- for reg_type, reg_length in register_lengths.items():
- c(".{0} = {1}_{2}_{3},".format(reg_type, gen.chipset, set.underscore_name, reg_type))
- c(".n_{0} = 0, /* Determined at runtime */".format(reg_type))
- c_outdent(3)
- c("};\n")
-
- c("\nstatic void\n")
- c("{0}_register_{1}_counter_query(struct gen_perf *perf)\n".format(gen.chipset, set.underscore_name))
- c("{\n")
- c_indent(3)
-
- c("static struct gen_perf_query_info *query = &" + gen.chipset + "_" + set.underscore_name + "_query;\n")
- c("struct gen_perf_query_counter *counter;\n")
+ c("\n")
+ c("struct gen_perf_query_counter *counter = query->counters;\n")
c("\n")
c("/* Note: we're assuming there can't be any variation in the definition ")
c_outdent(3)
c("}\n")
- h("void gen_oa_register_queries_" + gen.chipset + "(struct gen_perf *perf);\n")
+ h("void gen_oa_register_queries_" + gen.chipset + "(struct gen_perf_config *perf);\n")
c("\nvoid")
- c("gen_oa_register_queries_" + gen.chipset + "(struct gen_perf *perf)")
+ c("gen_oa_register_queries_" + gen.chipset + "(struct gen_perf_config *perf)")
c("{")
c_indent(3)
projects / mesa.git / blobdiff