1 # Copyright (c) 2015-2017 Intel Corporation
3 # Permission is hereby granted, free of charge, to any person obtaining a
4 # copy of this software and associated documentation files (the "Software"),
5 # to deal in the Software without restriction, including without limitation
6 # the rights to use, copy, modify, merge, publish, distribute, sublicense,
7 # and/or sell copies of the Software, and to permit persons to whom the
8 # Software is furnished to do so, subject to the following conditions:
10 # The above copyright notice and this permission notice (including the next
11 # paragraph) shall be included in all copies or substantial portions of the
14 # THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
15 # IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16 # FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
17 # THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
18 # LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
19 # FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
27 import xml
.etree
.ElementTree
as et
35 code
= ' '.join(map(str,args
))
36 for line
in code
.splitlines():
37 text
= ''.rjust(_c_indent
) + line
38 c_file
.write(text
.rstrip() + "\n")
40 # indented, but no trailing newline...
41 def c_line_start(code
):
42 c_file
.write(''.rjust(_c_indent
) + code
)
48 _c_indent
= _c_indent
+ n
51 _c_indent
= _c_indent
- n
57 code
= ' '.join(map(str,args
))
58 for line
in code
.splitlines():
59 text
= ''.rjust(_h_indent
) + line
60 header_file
.write(text
.rstrip() + "\n")
64 _h_indent
= _h_indent
+ n
67 _h_indent
= _h_indent
- n
70 def emit_fadd(tmp_id
, args
):
71 c("double tmp{0} = {1} + {2};".format(tmp_id
, args
[1], args
[0]))
74 # Be careful to check for divide by zero...
75 def emit_fdiv(tmp_id
, args
):
76 c("double tmp{0} = {1};".format(tmp_id
, args
[1]))
77 c("double tmp{0} = {1};".format(tmp_id
+ 1, args
[0]))
78 c("double tmp{0} = tmp{1} ? tmp{2} / tmp{1} : 0;".format(tmp_id
+ 2, tmp_id
+ 1, tmp_id
))
81 def emit_fmax(tmp_id
, args
):
82 c("double tmp{0} = {1};".format(tmp_id
, args
[1]))
83 c("double tmp{0} = {1};".format(tmp_id
+ 1, args
[0]))
84 c("double tmp{0} = MAX(tmp{1}, tmp{2});".format(tmp_id
+ 2, tmp_id
, tmp_id
+ 1))
87 def emit_fmul(tmp_id
, args
):
88 c("double tmp{0} = {1} * {2};".format(tmp_id
, args
[1], args
[0]))
91 def emit_fsub(tmp_id
, args
):
92 c("double tmp{0} = {1} - {2};".format(tmp_id
, args
[1], args
[0]))
95 def emit_read(tmp_id
, args
):
96 type = args
[1].lower()
97 c("uint64_t tmp{0} = accumulator[query->{1}_offset + {2}];".format(tmp_id
, type, args
[0]))
100 def emit_uadd(tmp_id
, args
):
101 c("uint64_t tmp{0} = {1} + {2};".format(tmp_id
, args
[1], args
[0]))
104 # Be careful to check for divide by zero...
105 def emit_udiv(tmp_id
, args
):
106 c("uint64_t tmp{0} = {1};".format(tmp_id
, args
[1]))
107 c("uint64_t tmp{0} = {1};".format(tmp_id
+ 1, args
[0]))
108 if args
[0].isdigit():
109 assert int(args
[0]) > 0
110 c("uint64_t tmp{0} = tmp{2} / tmp{1};".format(tmp_id
+ 2, tmp_id
+ 1, tmp_id
))
112 c("uint64_t tmp{0} = tmp{1} ? tmp{2} / tmp{1} : 0;".format(tmp_id
+ 2, tmp_id
+ 1, tmp_id
))
115 def emit_umul(tmp_id
, args
):
116 c("uint64_t tmp{0} = {1} * {2};".format(tmp_id
, args
[1], args
[0]))
119 def emit_usub(tmp_id
, args
):
120 c("uint64_t tmp{0} = {1} - {2};".format(tmp_id
, args
[1], args
[0]))
123 def emit_umin(tmp_id
, args
):
124 c("uint64_t tmp{0} = MIN({1}, {2});".format(tmp_id
, args
[1], args
[0]))
127 def emit_lshft(tmp_id
, args
):
128 c("uint64_t tmp{0} = {1} << {2};".format(tmp_id
, args
[1], args
[0]))
131 def emit_rshft(tmp_id
, args
):
132 c("uint64_t tmp{0} = {1} >> {2};".format(tmp_id
, args
[1], args
[0]))
135 def emit_and(tmp_id
, args
):
136 c("uint64_t tmp{0} = {1} & {2};".format(tmp_id
, args
[1], args
[0]))
140 # (n operands, emitter)
141 ops
["FADD"] = (2, emit_fadd
)
142 ops
["FDIV"] = (2, emit_fdiv
)
143 ops
["FMAX"] = (2, emit_fmax
)
144 ops
["FMUL"] = (2, emit_fmul
)
145 ops
["FSUB"] = (2, emit_fsub
)
146 ops
["READ"] = (2, emit_read
)
147 ops
["UADD"] = (2, emit_uadd
)
148 ops
["UDIV"] = (2, emit_udiv
)
149 ops
["UMUL"] = (2, emit_umul
)
150 ops
["USUB"] = (2, emit_usub
)
151 ops
["UMIN"] = (2, emit_umin
)
152 ops
["<<"] = (2, emit_lshft
)
153 ops
[">>"] = (2, emit_rshft
)
154 ops
["AND"] = (2, emit_and
)
158 return "(" + subexp
+ ")"
162 def splice_bitwise_and(args
):
163 return brkt(args
[1]) + " & " + brkt(args
[0])
165 def splice_logical_and(args
):
166 return brkt(args
[1]) + " && " + brkt(args
[0])
168 def splice_ult(args
):
169 return brkt(args
[1]) + " < " + brkt(args
[0])
171 def splice_ugte(args
):
172 return brkt(args
[1]) + " >= " + brkt(args
[0])
175 # (n operands, splicer)
176 exp_ops
["AND"] = (2, splice_bitwise_and
)
177 exp_ops
["UGTE"] = (2, splice_ugte
)
178 exp_ops
["ULT"] = (2, splice_ult
)
179 exp_ops
["&&"] = (2, splice_logical_and
)
183 hw_vars
["$EuCoresTotalCount"] = "perf->sys_vars.n_eus"
184 hw_vars
["$EuSlicesTotalCount"] = "perf->sys_vars.n_eu_slices"
185 hw_vars
["$EuSubslicesTotalCount"] = "perf->sys_vars.n_eu_sub_slices"
186 hw_vars
["$EuThreadsCount"] = "perf->sys_vars.eu_threads_count"
187 hw_vars
["$SliceMask"] = "perf->sys_vars.slice_mask"
188 # subslice_mask is interchangeable with subslice/dual-subslice since Gen12+
189 # only has dual subslices which can be assimilated with 16EUs subslices.
190 hw_vars
["$SubsliceMask"] = "perf->sys_vars.subslice_mask"
191 hw_vars
["$DualSubsliceMask"] = "perf->sys_vars.subslice_mask"
192 hw_vars
["$GpuTimestampFrequency"] = "perf->sys_vars.timestamp_frequency"
193 hw_vars
["$GpuMinFrequency"] = "perf->sys_vars.gt_min_freq"
194 hw_vars
["$GpuMaxFrequency"] = "perf->sys_vars.gt_max_freq"
195 hw_vars
["$SkuRevisionId"] = "perf->sys_vars.revision"
197 def output_rpn_equation_code(set, counter
, equation
):
198 c("/* RPN equation: " + equation
+ " */")
199 tokens
= equation
.split()
206 while stack
and stack
[-1] in ops
:
208 argc
, callback
= ops
[op
]
210 for i
in range(0, argc
):
211 operand
= stack
.pop()
212 if operand
[0] == "$":
213 if operand
in hw_vars
:
214 operand
= hw_vars
[operand
]
215 elif operand
in set.counter_vars
:
216 reference
= set.counter_vars
[operand
]
217 operand
= set.read_funcs
[operand
[1:]] + "(perf, query, accumulator)"
219 raise Exception("Failed to resolve variable " + operand
+ " in equation " + equation
+ " for " + set.name
+ " :: " + counter
.get('name'));
222 tmp_id
= callback(tmp_id
, args
)
224 tmp
= "tmp{0}".format(tmp_id
- 1)
228 raise Exception("Spurious empty rpn code for " + set.name
+ " :: " +
229 counter
.get('name') + ".\nThis is probably due to some unhandled RPN function, in the equation \"" +
235 value
= hw_vars
[value
]
236 if value
in set.counter_vars
:
237 value
= set.read_funcs
[value
[1:]] + "(perf, query, accumulator)"
239 c("\nreturn " + value
+ ";")
241 def splice_rpn_expression(set, counter
, expression
):
242 tokens
= expression
.split()
247 while stack
and stack
[-1] in exp_ops
:
249 argc
, callback
= exp_ops
[op
]
251 for i
in range(0, argc
):
252 operand
= stack
.pop()
253 if operand
[0] == "$":
254 if operand
in hw_vars
:
255 operand
= hw_vars
[operand
]
257 raise Exception("Failed to resolve variable " + operand
+ " in expression " + expression
+ " for " + set.name
+ " :: " + counter
.get('name'));
260 subexp
= callback(args
)
265 raise Exception("Spurious empty rpn expression for " + set.name
+ " :: " +
266 counter
.get('name') + ".\nThis is probably due to some unhandled RPN operation, in the expression \"" +
271 def output_counter_read(gen
, set, counter
):
273 c("/* {0} :: {1} */".format(set.name
, counter
.get('name')))
275 if counter
.read_hash
in hashed_funcs
:
276 c("#define %s \\" % counter
.read_sym
)
278 c("%s" % hashed_funcs
[counter
.read_hash
])
281 ret_type
= counter
.get('data_type')
282 if ret_type
== "uint64":
283 ret_type
= "uint64_t"
285 read_eq
= counter
.get('equation')
287 c("static " + ret_type
)
288 c(counter
.read_sym
+ "(UNUSED struct gen_perf_config *perf,\n")
289 c_indent(len(counter
.read_sym
) + 1)
290 c("const struct gen_perf_query_info *query,\n")
291 c("const uint64_t *accumulator)\n")
292 c_outdent(len(counter
.read_sym
) + 1)
296 output_rpn_equation_code(set, counter
, read_eq
)
300 hashed_funcs
[counter
.read_hash
] = counter
.read_sym
303 def output_counter_max(gen
, set, counter
):
304 max_eq
= counter
.get('max_equation')
306 if not counter
.has_max_func():
310 c("/* {0} :: {1} */".format(set.name
, counter
.get('name')))
312 if counter
.max_hash
in hashed_funcs
:
313 c("#define %s \\" % counter
.max_sym())
315 c("%s" % hashed_funcs
[counter
.max_hash
])
318 ret_type
= counter
.get('data_type')
319 if ret_type
== "uint64":
320 ret_type
= "uint64_t"
322 c("static " + ret_type
)
323 c(counter
.max_sym() + "(struct gen_perf_config *perf)\n")
326 output_rpn_equation_code(set, counter
, max_eq
)
330 hashed_funcs
[counter
.max_hash
] = counter
.max_sym()
333 c_type_sizes
= { "uint32_t": 4, "uint64_t": 8, "float": 4, "double": 8, "bool": 4 }
335 return c_type_sizes
[c_type
]
337 def pot_align(base
, pot_alignment
):
338 return (base
+ pot_alignment
- 1) & ~
(pot_alignment
- 1);
340 semantic_type_map
= {
345 def output_availability(set, availability
, counter_name
):
346 expression
= splice_rpn_expression(set, counter_name
, availability
)
347 lines
= expression
.split(' && ')
350 c("if (" + lines
[0] + ") {")
352 c("if (" + lines
[0] + " &&")
354 for i
in range(1, (n_lines
- 1)):
356 c(lines
[(n_lines
- 1)] + ") {")
360 def output_units(unit
):
361 return unit
.replace(' ', '_').upper()
364 def output_counter_report(set, counter
, current_offset
):
365 data_type
= counter
.get('data_type')
366 data_type_uc
= data_type
.upper()
370 c_type
= c_type
+ "_t"
372 semantic_type
= counter
.get('semantic_type')
373 if semantic_type
in semantic_type_map
:
374 semantic_type
= semantic_type_map
[semantic_type
]
376 semantic_type_uc
= semantic_type
.upper()
380 availability
= counter
.get('availability')
382 output_availability(set, availability
, counter
.get('name'))
385 c("counter = &query->counters[query->n_counters++];\n")
386 c("counter->oa_counter_read_" + data_type
+ " = " + set.read_funcs
[counter
.get('symbol_name')] + ";\n")
387 c("counter->name = \"" + counter
.get('name') + "\";\n")
388 c("counter->desc = \"" + counter
.get('description') + "\";\n")
389 c("counter->symbol_name = \"" + counter
.get('symbol_name') + "\";\n")
390 c("counter->category = \"" + counter
.get('mdapi_group') + "\";\n")
391 c("counter->type = GEN_PERF_COUNTER_TYPE_" + semantic_type_uc
+ ";\n")
392 c("counter->data_type = GEN_PERF_COUNTER_DATA_TYPE_" + data_type_uc
+ ";\n")
393 c("counter->units = GEN_PERF_COUNTER_UNITS_" + output_units(counter
.get('units')) + ";\n")
394 c("counter->raw_max = " + set.max_values
[counter
.get('symbol_name')] + ";\n")
396 current_offset
= pot_align(current_offset
, sizeof(c_type
))
397 c("counter->offset = " + str(current_offset
) + ";\n")
403 return current_offset
+ sizeof(c_type
)
409 'OA': 'b_counter_regs',
412 def compute_register_lengths(set):
413 register_lengths
= {}
414 register_configs
= set.findall('register_config')
415 for register_config
in register_configs
:
416 t
= register_types
[register_config
.get('type')]
417 if t
not in register_lengths
:
418 register_lengths
[t
] = len(register_config
.findall('register'))
420 register_lengths
[t
] += len(register_config
.findall('register'))
422 return register_lengths
425 def generate_register_configs(set):
426 register_configs
= set.findall('register_config')
428 for register_config
in register_configs
:
429 t
= register_types
[register_config
.get('type')]
431 availability
= register_config
.get('availability')
433 output_availability(set, availability
, register_config
.get('type') + ' register config')
436 registers
= register_config
.findall('register')
437 c("static const struct gen_perf_query_register_prog %s[] = {" % t
)
439 for register
in registers
:
440 c("{ .reg = %s, .val = %s }," % (register
.get('address'), register
.get('value')))
443 c("query->config.%s = %s;" % (t
, t
))
444 c("query->config.n_%s = ARRAY_SIZE(%s);" % (t
, t
))
452 # Wraps a <counter> element from the oa-*.xml files.
454 def __init__(self
, set, xml
):
457 self
.read_hash
= None
460 self
.read_sym
= "{0}__{1}__{2}__read".format(self
.set.gen
.chipset
,
461 self
.set.underscore_name
,
462 self
.xml
.get('underscore_name'))
465 return self
.xml
.get(prop
)
467 # Compute the hash of a counter's equation by expanding (including all the
468 # sub-equations it depends on)
469 def compute_hashes(self
):
470 if self
.read_hash
is not None:
473 def replace_token(token
):
476 if token
not in self
.set.counter_vars
:
478 self
.set.counter_vars
[token
].compute_hashes()
479 return self
.set.counter_vars
[token
].read_hash
481 read_eq
= self
.xml
.get('equation')
482 self
.read_hash
= ' '.join(map(replace_token
, read_eq
.split()))
484 max_eq
= self
.xml
.get('max_equation')
486 self
.max_hash
= ' '.join(map(replace_token
, max_eq
.split()))
488 def has_max_func(self
):
489 max_eq
= self
.xml
.get('max_equation')
499 for token
in max_eq
.split():
500 if token
[0] == '$' and token
not in hw_vars
:
505 assert self
.has_max_func()
506 return "{0}__{1}__{2}__max".format(self
.set.gen
.chipset
,
507 self
.set.underscore_name
,
508 self
.xml
.get('underscore_name'))
511 max_eq
= self
.xml
.get('max_equation')
513 return "0 /* undefined */"
516 return "{0}".format(float(max_eq
))
520 for token
in max_eq
.split():
521 if token
[0] == '$' and token
not in hw_vars
:
522 return "0 /* unsupported (varies over time) */"
524 return "{0}__{1}__{2}__max(perf)".format(self
.set.gen
.chipset
,
525 self
.set.underscore_name
,
526 self
.xml
.get('underscore_name'))
528 # Wraps a <set> element from the oa-*.xml files.
530 def __init__(self
, gen
, xml
):
534 self
.counter_vars
= {}
538 xml_counters
= self
.xml
.findall("counter")
540 for xml_counter
in xml_counters
:
541 counter
= Counter(self
, xml_counter
)
542 self
.counters
.append(counter
)
543 self
.counter_vars
["$" + counter
.get('symbol_name')] = counter
544 self
.read_funcs
[counter
.get('symbol_name')] = counter
.read_sym
545 self
.max_values
[counter
.get('symbol_name')] = counter
.max_value()
547 for counter
in self
.counters
:
548 counter
.compute_hashes()
551 def hw_config_guid(self
):
552 return self
.xml
.get('hw_config_guid')
556 return self
.xml
.get('name')
559 def symbol_name(self
):
560 return self
.xml
.get('symbol_name')
563 def underscore_name(self
):
564 return self
.xml
.get('underscore_name')
566 def findall(self
, path
):
567 return self
.xml
.findall(path
)
569 def find(self
, path
):
570 return self
.xml
.find(path
)
573 # Wraps an entire oa-*.xml file.
575 def __init__(self
, filename
):
576 self
.filename
= filename
577 self
.xml
= et
.parse(self
.filename
)
578 self
.chipset
= self
.xml
.find('.//set').get('chipset').lower()
581 for xml_set
in self
.xml
.findall(".//set"):
582 self
.sets
.append(Set(self
, xml_set
))
589 parser
= argparse
.ArgumentParser()
590 parser
.add_argument("--header", help="Header file to write", required
=True)
591 parser
.add_argument("--code", help="C file to write", required
=True)
592 parser
.add_argument("xml_files", nargs
='+', help="List of xml metrics files to process")
594 args
= parser
.parse_args()
596 c_file
= open(args
.code
, 'w')
597 header_file
= open(args
.header
, 'w')
600 for xml_file
in args
.xml_files
:
601 gens
.append(Gen(xml_file
))
604 copyright
= textwrap
.dedent("""\
605 /* Autogenerated file, DO NOT EDIT manually! generated by {}
607 * Copyright (c) 2015 Intel Corporation
609 * Permission is hereby granted, free of charge, to any person obtaining a
610 * copy of this software and associated documentation files (the "Software"),
611 * to deal in the Software without restriction, including without limitation
612 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
613 * and/or sell copies of the Software, and to permit persons to whom the
614 * Software is furnished to do so, subject to the following conditions:
616 * The above copyright notice and this permission notice (including the next
617 * paragraph) shall be included in all copies or substantial portions of the
620 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
621 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
622 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
623 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
624 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
625 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
626 * DEALINGS IN THE SOFTWARE.
629 """).format(os
.path
.basename(__file__
))
632 h(textwrap
.dedent("""\
635 struct gen_perf_config;
640 c(textwrap
.dedent("""\
644 #include <drm-uapi/i915_drm.h>
646 #include "util/hash_table.h"
647 #include "util/ralloc.h"
651 c("#include \"" + os
.path
.basename(args
.header
) + "\"")
653 c(textwrap
.dedent("""\
654 #include "perf/gen_perf.h"
657 #define MIN(a, b) ((a < b) ? (a) : (b))
658 #define MAX(a, b) ((a > b) ? (a) : (b))
663 # Print out all equation functions.
666 for counter
in set.counters
:
667 output_counter_read(gen
, set, counter
)
668 output_counter_max(gen
, set, counter
)
670 # Print out all metric sets registration functions for each set in each
674 counters
= set.counters
678 c("{0}_register_{1}_counter_query(struct gen_perf_config *perf)\n".format(gen
.chipset
, set.underscore_name
))
682 c("struct gen_perf_query_info *query = rzalloc(perf, struct gen_perf_query_info);\n")
684 c("query->kind = GEN_PERF_QUERY_TYPE_OA;\n")
685 c("query->name = \"" + set.name
+ "\";\n")
686 c("query->symbol_name = \"" + set.symbol_name
+ "\";\n")
687 c("query->guid = \"" + set.hw_config_guid
+ "\";\n")
689 c("query->counters = rzalloc_array(query, struct gen_perf_query_counter, %u);" % len(counters
))
690 c("query->n_counters = 0;")
691 c("query->oa_metrics_set_id = 0; /* determined at runtime, via sysfs */")
693 if gen
.chipset
== "hsw":
694 c(textwrap
.dedent("""\
695 query->oa_format = I915_OA_FORMAT_A45_B8_C8;
696 /* Accumulation buffer offsets... */
697 query->gpu_time_offset = 0;
699 query->b_offset = 46;
700 query->c_offset = 54;
703 c(textwrap
.dedent("""\
704 query->oa_format = I915_OA_FORMAT_A32u40_A4u32_B8_C8;
705 /* Accumulation buffer offsets... */
706 query->gpu_time_offset = 0;
707 query->gpu_clock_offset = 1;
709 query->b_offset = 38;
710 query->c_offset = 46;
715 c("struct gen_perf_query_counter *counter = query->counters;\n")
718 c("/* Note: we're assuming there can't be any variation in the definition ")
719 c(" * of a query between contexts so it's ok to describe a query within a ")
720 c(" * global variable which only needs to be initialized once... */")
721 c("\nif (!query->data_size) {")
724 generate_register_configs(set)
727 for counter
in counters
:
728 offset
= output_counter_report(set, counter
, offset
)
731 c("\nquery->data_size = counter->offset + gen_perf_query_counter_get_size(counter);\n")
736 c("\n_mesa_hash_table_insert(perf->oa_metrics_table, query->guid, query);")
741 h("void gen_oa_register_queries_" + gen
.chipset
+ "(struct gen_perf_config *perf);\n")
744 c("gen_oa_register_queries_" + gen
.chipset
+ "(struct gen_perf_config *perf)")
749 c("{0}_register_{1}_counter_query(perf);".format(gen
.chipset
, set.underscore_name
))
755 if __name__
== '__main__':