Merge pull request #771 from zakkak/2024-07-17-july-cpu-2024-sync

[24.0] Sync with upstream for July 2024 CPU

ci/ci_common/bench-common.libsonnet

@@ -31,6 +31,17 @@
       else true
   },
 
+  # max number of threads to use for benchmarking in general
+  # the goal being to limit parallelism on very large servers which may not be respresentative of real-world scenarios
+  bench_max_threads:: {
+   restrict_threads:: 36
+  },
+
+  bench_no_thread_cap:: {
+    restrict_threads:: null,
+    should_use_hwloc:: false
+  },
+
   bench_hw:: {
     _bench_machine:: {
       targets+: ["bench"],
@@ -40,15 +51,16 @@
       numa_nodes:: [],
       is_numa:: std.length(self.numa_nodes) > 0,
       num_threads:: error "num_threads must bet set!",
+      hyperthreading:: true,
       threads_per_node:: if self.is_numa then self.num_threads / std.length(self.numa_nodes) else self.num_threads,
     },
 
-    x52:: common.linux_amd64 + self._bench_machine + {
-      machine_name:: "x52",
-      capabilities+: ["tmpfs25g"],
+    e3:: common.linux_amd64 + self._bench_machine + {
+      machine_name:: "e3",
+      capabilities: ["e3", "tmpfs25g", "linux", "amd64"],
       numa_nodes:: [0, 1],
-      default_numa_node:: 0,
-      num_threads:: 72
+      default_numa_node:: 1,
+      num_threads:: 256
     },
     e4_8_64:: common.linux_amd64 + self._bench_machine + {
       machine_name:: "e4_8_64",
@@ -67,31 +79,29 @@
     xgene3:: common.linux_aarch64 + self._bench_machine + {
       machine_name:: "xgene3",
       capabilities+: [],
-      num_threads:: 32
+      num_threads:: 32,
+      hyperthreading:: false
     },
     a12c:: common.linux_aarch64 + self._bench_machine + {
       machine_name:: "a12c",
       capabilities+: ["tmpfs25g"],
       numa_nodes:: [0, 1],
       default_numa_node:: 0,
-      num_threads:: 160
+      num_threads:: 160,
+      hyperthreading:: false
     }
   },
 
-  hwlocIfNuma(numa, cmd, node=0)::
-    if numa then
+  hwloc_cmd(cmd, num_threads, node, hyperthreading, max_threads_per_node)::
+    if num_threads == null then
       ["hwloc-bind", "--cpubind", "node:"+node, "--membind", "node:"+node, "--"] + cmd
     else
-      cmd,
-
-  parallelHwloc(cmd_node0, cmd_node1)::
-    // Returns a list of commands that will run cmd_nod0 on NUMA node 0
-    // concurrently with cmd_node1 on NUMA node 1 and then wait for both to complete.
-    [
-      $.hwlocIfNuma(true, cmd_node0, node=0) + ["&"],
-      $.hwlocIfNuma(true, cmd_node1, node=1) + ["&"],
-      ["wait"]
-    ],
+      local threads = if num_threads != null then num_threads else max_threads_per_node;
+      assert if hyperthreading then threads % 2 == 0 else true: "It is required to bind to an even number of threads on hyperthreaded machines. Got requested "+threads+" threads";
+      assert threads <= max_threads_per_node: "Benchmarking must run on a single NUMA node for stability reasons. Got requested "+threads+" threads but the machine has only "+max_threads_per_node+" threads per node";      local cores = if hyperthreading then "0-"+((threads/2)-1)+".pu:0-1" else "0-"+(threads-1)+".pu:0";
+      local cpu_bind = if hyperthreading then "node:"+node+".core:"+cores else "node:"+node+".core:"+cores+".pu:0";
+      ["hwloc-bind", "--cpubind", cpu_bind, "--membind", "node:"+node, "--"] + cmd
+  ,
 
   // building block used to generate fork builds
   many_forks_benchmarking:: common.build_base + {

common.json

@@ -4,7 +4,7 @@
     "Jsonnet files should not include this file directly but use ci/common.jsonnet instead."
   ],
 
-  "mx_version": "7.4.1",
+  "mx_version": "7.4.1.1",
 
   "COMMENT.jdks": "When adding or removing JDKs keep in sync with JDKs in ci/common.jsonnet",
   "jdks": {
@@ -44,13 +44,13 @@
     "labsjdk-ee-21Debug": {"name": "labsjdk",   "version": "ee-21.0.1+11-jvmci-23.1-b26-debug", "platformspecific": true },
     "labsjdk-ee-21-llvm": {"name": "labsjdk",   "version": "ee-21.0.1+11-jvmci-23.1-b26-sulong", "platformspecific": true },
 
-    "oraclejdk-latest":       {"name": "jpg-jdk",   "version": "22",      "build_id": "jdk-22.0.1+8", "platformspecific": true, "extrabundles": ["static-libs"]},
-    "labsjdk-ce-latest":      {"name": "labsjdk",   "version": "ce-22.0.1+8-jvmci-b01", "platformspecific": true },
-    "labsjdk-ce-latestDebug": {"name": "labsjdk",   "version": "ce-22.0.1+8-jvmci-b01-debug", "platformspecific": true },
-    "labsjdk-ce-latest-llvm": {"name": "labsjdk",   "version": "ce-22.0.1+8-jvmci-b01-sulong", "platformspecific": true },
-    "labsjdk-ee-latest":      {"name": "labsjdk",   "version": "ee-22.0.1+8-jvmci-b01", "platformspecific": true },
-    "labsjdk-ee-latestDebug": {"name": "labsjdk",   "version": "ee-22.0.1+8-jvmci-b01-debug", "platformspecific": true },
-    "labsjdk-ee-latest-llvm": {"name": "labsjdk",   "version": "ee-22.0.1+8-jvmci-b01-sulong", "platformspecific": true }
+    "oraclejdk-latest":       {"name": "jpg-jdk",   "version": "22",      "build_id": "jdk-22.0.2+9", "platformspecific": true, "extrabundles": ["static-libs"]},
+    "labsjdk-ce-latest":      {"name": "labsjdk",   "version": "ce-22.0.2+9-jvmci-b01", "platformspecific": true },
+    "labsjdk-ce-latestDebug": {"name": "labsjdk",   "version": "ce-22.0.2+9-jvmci-b01-debug", "platformspecific": true },
+    "labsjdk-ce-latest-llvm": {"name": "labsjdk",   "version": "ce-22.0.2+9-jvmci-b01-sulong", "platformspecific": true },
+    "labsjdk-ee-latest":      {"name": "labsjdk",   "version": "ee-22.0.2+9-jvmci-b01", "platformspecific": true },
+    "labsjdk-ee-latestDebug": {"name": "labsjdk",   "version": "ee-22.0.2+9-jvmci-b01-debug", "platformspecific": true },
+    "labsjdk-ee-latest-llvm": {"name": "labsjdk",   "version": "ee-22.0.2+9-jvmci-b01-sulong", "platformspecific": true }
   },
 
   "eclipse": {

compiler/ci/ci_common/benchmark-builders.jsonnet

@@ -11,41 +11,35 @@
 
   local main_builds = std.flattenArrays([
     [
-    c.daily + c.opt_post_merge + hw.x52 + jdk + cc.libgraal + bench.dacapo + PR_bench_libgraal,
-    c.weekly                   + hw.x52 + jdk + cc.libgraal + bench.dacapo_size_variants,
-    c.daily + c.opt_post_merge + hw.x52 + jdk + cc.libgraal + bench.scala_dacapo + PR_bench_libgraal,
-    c.weekly                   + hw.x52 + jdk + cc.libgraal + bench.scala_dacapo_size_variants,
-    c.daily + c.opt_post_merge + hw.x52 + jdk + cc.libgraal + bench.renaissance + PR_bench_libgraal,
-    c.daily + c.opt_post_merge + hw.x52 + jdk + cc.libgraal + bench.specjvm2008 + PR_bench_libgraal,
-    c.weekly                   + hw.x52 + jdk + cc.libgraal + bench.specjbb2015,
-    c.daily + c.opt_post_merge + hw.x52 + jdk + cc.libgraal + bench.awfy + PR_bench_libgraal,
-    c.daily                    + hw.x52 + jdk + cc.libgraal + bench.microservice_benchmarks,
-    c.daily                    + hw.x52 + jdk + cc.libgraal + bench.micros_graal_whitebox,
-    c.daily                    + hw.x52 + jdk + cc.libgraal + bench.micros_graal_dist,
-    c.daily                    + hw.x52 + jdk + cc.libgraal + bench.micros_misc_graal_dist,
-    c.daily                    + hw.x52 + jdk + cc.libgraal + bench.micros_shootout_graal_dist,
-    c.daily                    + hw.e4_8_64 + jdk + cc.libgraal + bench.awfy + {job_prefix:: "bench-e4vm-compiler"},
-    c.daily                    + hw.e4_8_64 + jdk + cc.libgraal + bench.dacapo + {job_prefix:: "bench-e4vm-compiler"},
-    c.daily                    + hw.e4_8_64 + jdk + cc.libgraal + bench.scala_dacapo + {job_prefix:: "bench-e4vm-compiler"},
-    c.daily                    + hw.e4_8_64 + jdk + cc.libgraal + bench.renaissance + {job_prefix:: "bench-e4vm-compiler"},
-    c.daily                    + hw.e4_8_64 + jdk + cc.libgraal + bench.specjvm2008 + {job_prefix:: "bench-e4vm-compiler"},
-    c.daily                    + hw.e4_8_64 + jdk + cc.libgraal + bench.microservice_benchmarks + {job_prefix:: "bench-e4vm-compiler"},
+    c.daily + c.opt_post_merge + hw.e3 + jdk + cc.libgraal + bench.dacapo + PR_bench_libgraal,
+    c.daily + c.opt_post_merge + hw.e3 + jdk + cc.libgraal + bench.scala_dacapo + PR_bench_libgraal,
+    c.daily + c.opt_post_merge + hw.e3 + jdk + cc.libgraal + bench.renaissance + PR_bench_libgraal,
+    c.daily + c.opt_post_merge + hw.e3 + jdk + cc.libgraal + bench.specjvm2008 + PR_bench_libgraal,
+    c.on_demand                + hw.e3 + jdk + cc.libgraal + bench.dacapo_size_variants,
+    c.on_demand                + hw.e3 + jdk + cc.libgraal + bench.scala_dacapo_size_variants,
+    c.monthly                  + hw.e3 + jdk + cc.libgraal + bench.specjbb2015,
+    c.daily + c.opt_post_merge + hw.e3 + jdk + cc.libgraal + bench.awfy + PR_bench_libgraal,
+    c.daily                    + hw.e3 + jdk + cc.libgraal + bench.microservice_benchmarks,
+    c.weekly                   + hw.e3 + jdk + cc.libgraal + bench.micros_graal_whitebox,
+    c.weekly                   + hw.e3 + jdk + cc.libgraal + bench.micros_graal_dist,
+    c.weekly                   + hw.e3 + jdk + cc.libgraal + bench.micros_misc_graal_dist,
+    c.weekly                   + hw.e3 + jdk + cc.libgraal + bench.micros_shootout_graal_dist,
     ]
   for jdk in cc.product_jdks
   ]),
 
   local profiling_builds = std.flattenArrays([
     [
-    c.weekly + hw.x52 + jdk + cc.libgraal + suite + cc.enable_profiling   + { job_prefix:: "bench-compiler-profiling" },
-    c.weekly + hw.x52 + jdk + cc.libgraal + suite + cc.footprint_tracking + { job_prefix:: "bench-compiler-footprint" }
+    c.monthly + hw.e3 + jdk + cc.libgraal + suite + cc.enable_profiling   + { job_prefix:: "bench-compiler-profiling" },
+    c.monthly + hw.e3 + jdk + cc.libgraal + suite + cc.footprint_tracking + { job_prefix:: "bench-compiler-footprint" }
     ]
   for jdk in cc.product_jdks
   for suite in bench.groups.main_suites
   ]),
 
   local weekly_amd64_forks_builds = std.flattenArrays([
-    bc.generate_fork_builds(c.weekly  + hw.x52  + jdk + cc.libgraal + suite, subdir='compiler') +
-    bc.generate_fork_builds(c.monthly + hw.x52  + jdk + cc.jargraal + suite, subdir='compiler')
+    bc.generate_fork_builds(c.weekly  + hw.e3  + jdk + cc.libgraal + suite, subdir='compiler') +
+    bc.generate_fork_builds(c.monthly + hw.e3  + jdk + cc.jargraal + suite, subdir='compiler')
   for jdk in cc.product_jdks
   for suite in bench.groups.weekly_forks_suites
   ]),
@@ -70,7 +64,7 @@
   ],
 
   local zgc_builds = [
-    c.weekly + hw.x52 + jdk + cc.libgraal + cc.zgc_mode + suite,
+    c.weekly + hw.e3 + jdk + cc.libgraal + cc.zgc_mode + suite,
   for jdk in cc.product_jdks
   for suite in bench.groups.main_suites + [bench.specjbb2015]
   ],
@@ -83,13 +77,13 @@
   ],
 
   local no_tiered_builds = [
-    c.weekly + hw.x52 + jdk + cc.libgraal + cc.no_tiered_comp + suite,
+    c.monthly + hw.e3 + jdk + cc.libgraal + cc.no_tiered_comp + suite,
   for jdk in cc.product_jdks
   for suite in bench.groups.main_suites
   ],
 
   local no_profile_info_builds = [
-    c.weekly + hw.x52 + jdk + cc.libgraal + cc.no_profile_info + suite,
+    c.monthly + hw.e3 + jdk + cc.libgraal + cc.no_profile_info + suite,
   for jdk in cc.product_jdks
   for suite in bench.groups.main_suites
   ],

compiler/ci/ci_common/benchmark-suites.libsonnet

@@ -24,7 +24,7 @@
 
   // suite definitions
   // *****************
-  awfy: cc.compiler_benchmark + c.heap.small + {
+  awfy: cc.compiler_benchmark + c.heap.small + bc.bench_max_threads + {
     suite:: "awfy",
     run+: [
       self.benchmark_cmd + ["awfy:*", "--"] + self.extra_vm_args
@@ -36,7 +36,7 @@
     max_jdk_version:: null
   },
 
-  dacapo: cc.compiler_benchmark + c.heap.default + {
+  dacapo: cc.compiler_benchmark + c.heap.default + bc.bench_max_threads + {
     suite:: "dacapo",
     run+: [
       self.benchmark_cmd + ["dacapo:*", "--"] + self.extra_vm_args
@@ -48,7 +48,7 @@
     max_jdk_version:: null
   },
 
-  dacapo_size_variants: cc.compiler_benchmark + c.heap.default + {
+  dacapo_size_variants: cc.compiler_benchmark + c.heap.default + bc.bench_max_threads + {
     suite:: "dacapo-size-variants",
     run+: [
       self.benchmark_cmd + ["dacapo-small:*", "--"] + self.extra_vm_args,
@@ -64,7 +64,7 @@
     max_jdk_version:: null
   },
 
-  scala_dacapo: cc.compiler_benchmark + c.heap.default + {
+  scala_dacapo: cc.compiler_benchmark + c.heap.default + bc.bench_max_threads + {
     suite:: "scala-dacapo",
     run+: [
       self.benchmark_cmd + ["scala-dacapo:*", "--"] + self.extra_vm_args
@@ -76,7 +76,7 @@
     max_jdk_version:: null
   },
 
-  scala_dacapo_size_variants: cc.compiler_benchmark + c.heap.default + {
+  scala_dacapo_size_variants: cc.compiler_benchmark + c.heap.default + bc.bench_max_threads + {
     suite:: "scala-dacapo-size-variants",
     run+: [
       self.benchmark_cmd + ["scala-dacapo-tiny:*", "--"] + self.extra_vm_args,
@@ -97,7 +97,7 @@
     max_jdk_version:: null
   },
 
-  renaissance_template(suite_version=null, suite_name="renaissance", max_jdk_version=null):: cc.compiler_benchmark + c.heap.default + {
+  renaissance_template(suite_version=null, suite_name="renaissance", max_jdk_version=null):: cc.compiler_benchmark + c.heap.default + bc.bench_max_threads + {
     suite:: suite_name,
     local suite_version_args = if suite_version != null then ["--bench-suite-version=" + suite_version] else [],
     run+: [
@@ -110,12 +110,9 @@
     max_jdk_version:: max_jdk_version
   },
 
-  renaissance: self.renaissance_template() + {
-    # [JDK-8303076] [GR-44499] requires extra stack size for C1
-    extra_vm_args+:: if self.platform == "c1" then ["-Xss1090K"] else []
-  },
+  renaissance: self.renaissance_template(),
 
-  specjbb2015: cc.compiler_benchmark + c.heap.large_with_large_young_gen + {
+  specjbb2015: cc.compiler_benchmark + c.heap.large_with_large_young_gen + bc.bench_max_threads + {
     suite:: "specjbb2015",
     downloads+: {
       "SPECJBB2015": { name: "specjbb2015", version: "1.03" }
@@ -130,7 +127,7 @@
     max_jdk_version:: null
   },
 
-  specjvm2008: cc.compiler_benchmark + c.heap.default + {
+  specjvm2008: cc.compiler_benchmark + c.heap.default + bc.bench_max_threads + {
     suite:: "specjvm2008",
     downloads+: {
       "SPECJVM2008": { name: "specjvm2008", version: "1.01" }
@@ -146,7 +143,7 @@
   },
 
   // Microservice benchmarks
-  microservice_benchmarks: cc.compiler_benchmark + {
+  microservice_benchmarks: cc.compiler_benchmark + bc.bench_no_thread_cap + {  # no thread cap here since hwloc is handled at the mx level for microservices
     suite:: "microservices",
     packages+: {
       "pip:psutil": "==5.8.0"
@@ -159,42 +156,26 @@
     local hwlocBind_16C_32T = ["--hwloc-bind=--cpubind node:0.core:0-15.pu:0-1 --membind node:0"],
     run+: [
       # shopcart-wrk
-      self.benchmark_cmd + ["shopcart-wrk:mixed-tiny"]                   + hwlocBind_1C_1T   + ["--"] + self.extra_vm_args + ["-Xms32m",   "-Xmx112m",  "-XX:ActiveProcessorCount=1",  "-XX:MaxDirectMemorySize=256m"],
-      bench_upload,
-      self.benchmark_cmd + ["shopcart-wrk:mixed-small"]                  + hwlocBind_2C_2T   + ["--"] + self.extra_vm_args + ["-Xms64m",   "-Xmx224m",  "-XX:ActiveProcessorCount=2",  "-XX:MaxDirectMemorySize=512m"],
-      bench_upload,
-      self.benchmark_cmd + ["shopcart-wrk:mixed-medium"]                 + hwlocBind_4C_4T   + ["--"] + self.extra_vm_args + ["-Xms128m",  "-Xmx512m",  "-XX:ActiveProcessorCount=4",  "-XX:MaxDirectMemorySize=1024m"],
-      bench_upload,
       self.benchmark_cmd + ["shopcart-wrk:mixed-large"]                  + hwlocBind_16C_16T + ["--"] + self.extra_vm_args + ["-Xms512m",  "-Xmx3072m", "-XX:ActiveProcessorCount=16", "-XX:MaxDirectMemorySize=4096m"],
       bench_upload,
-      self.benchmark_cmd + ["shopcart-wrk:mixed-huge"]                   + hwlocBind_16C_32T + ["--"] + self.extra_vm_args + ["-Xms1024m", "-Xmx8192m", "-XX:ActiveProcessorCount=32", "-XX:MaxDirectMemorySize=8192m"],
-      bench_upload,
 
       # petclinic-wrk
-      self.benchmark_cmd + ["petclinic-wrk:mixed-tiny"]                  + hwlocBind_1C_1T   + ["--"] + self.extra_vm_args + ["-Xms32m",   "-Xmx100m",  "-XX:ActiveProcessorCount=1"],
-      bench_upload,
-      self.benchmark_cmd + ["petclinic-wrk:mixed-small"]                 + hwlocBind_2C_2T   + ["--"] + self.extra_vm_args + ["-Xms40m",   "-Xmx144m",  "-XX:ActiveProcessorCount=2"],
-      bench_upload,
-      self.benchmark_cmd + ["petclinic-wrk:mixed-medium"]                + hwlocBind_4C_4T   + ["--"] + self.extra_vm_args + ["-Xms80m",   "-Xmx256m",  "-XX:ActiveProcessorCount=4"],
-      bench_upload,
       self.benchmark_cmd + ["petclinic-wrk:mixed-large"]                 + hwlocBind_16C_16T + ["--"] + self.extra_vm_args + ["-Xms128m",  "-Xmx512m", "-XX:ActiveProcessorCount=16"],
       bench_upload,
-      self.benchmark_cmd + ["petclinic-wrk:mixed-huge"]                  + hwlocBind_16C_32T + ["--"] + self.extra_vm_args + ["-Xms640m",  "-Xmx3072m", "-XX:ActiveProcessorCount=32"],
-      bench_upload,
 
       # helloworld-wrk
       self.benchmark_cmd + ["micronaut-helloworld-wrk:helloworld"]       + hwlocBind_1C_1T   + ["--"] + self.extra_vm_args + ["-Xms8m",    "-Xmx64m",   "-XX:ActiveProcessorCount=1", "-XX:MaxDirectMemorySize=256m"],
       bench_upload,
       self.benchmark_cmd + ["spring-helloworld-wrk:helloworld"]          + hwlocBind_1C_1T   + ["--"] + self.extra_vm_args + ["-Xms8m",    "-Xmx64m",   "-XX:ActiveProcessorCount=1", "-XX:MaxDirectMemorySize=256m"],
       bench_upload
     ],
-    timelimit: "7:00:00",
+    timelimit: "4:00:00",
     min_jdk_version:: 11,
     max_jdk_version:: null
   },
 
   // JMH microbenchmarks
-  micros_graal_whitebox: cc.compiler_benchmark + c.heap.default + {
+  micros_graal_whitebox: cc.compiler_benchmark + c.heap.default + bc.bench_max_threads + {
     suite:: "micros-graal-whitebox",
     run+: [
       self.benchmark_cmd + ["jmh-whitebox:*", "--"] + self.extra_vm_args + ["--", "jdk.graal.compiler"]
@@ -204,7 +185,7 @@
     max_jdk_version:: null
   },
 
-  micros_graal_dist: cc.compiler_benchmark + c.heap.default + {
+  micros_graal_dist: cc.compiler_benchmark + c.heap.default + bc.bench_max_threads + {
     suite:: "micros-graal-dist",
     run+: [
       self.benchmark_cmd + ["jmh-dist:GRAAL_COMPILER_MICRO_BENCHMARKS", "--"] + self.extra_vm_args
@@ -214,7 +195,7 @@
     max_jdk_version:: null
   },
 
-  micros_misc_graal_dist: cc.compiler_benchmark + c.heap.default + {
+  micros_misc_graal_dist: cc.compiler_benchmark + c.heap.default + bc.bench_max_threads + {
     suite:: "micros-misc-graal-dist",
     run+: [
       self.benchmark_cmd + ["jmh-dist:GRAAL_BENCH_MISC", "--"] + self.extra_vm_args
@@ -224,7 +205,7 @@
     max_jdk_version:: null
   },
 
-  micros_shootout_graal_dist: cc.compiler_benchmark + c.heap.default {
+  micros_shootout_graal_dist: cc.compiler_benchmark + c.heap.default + bc.bench_max_threads + {
     suite:: "micros-shootout-graal-dist",
     run+: [
       self.benchmark_cmd + ["jmh-dist:GRAAL_BENCH_SHOOTOUT", "--"] + self.extra_vm_args

compiler/ci/ci_common/compiler-common.libsonnet

@@ -71,7 +71,8 @@
       "${BENCH_RESULTS_FILE_PATH}",
       "--machine-name=${MACHINE_NAME}"] +
       (if std.objectHasAll(self.environment, 'MX_TRACKER') then ["--tracker=" + self.environment['MX_TRACKER']] else ["--tracker=rss"]),
-    benchmark_cmd:: bench_common.hwlocIfNuma(self.should_use_hwloc, self.plain_benchmark_cmd, node=self.default_numa_node),
+    restrict_threads:: null,  # can be overridden to restrict the benchmark to the given number of threads. If null, will use one full NUMA node
+    benchmark_cmd:: if self.should_use_hwloc then bench_common.hwloc_cmd(self.plain_benchmark_cmd, self.restrict_threads, self.default_numa_node, self.hyperthreading, self.threads_per_node) else self.plain_benchmark_cmd,
     min_heap_size:: if std.objectHasAll(self.environment, 'XMS') then ["-Xms${XMS}"] else [],
     max_heap_size:: if std.objectHasAll(self.environment, 'XMX') then ["-Xmx${XMX}"] else [],
     extra_vm_args::