Qthreads 1.22 (chapel-lang#26744)

This updates the bundled copy of qthreads to 1.22. It also switches to using the new qthreads cmake build config. Currently this is using the latest 1.22 release candidate and shouldn't be merged until the 1.22 release is finalized there. I'm submitting it now for testing so there's some time to fix anything that comes up before the release. [Contributed by @insertinterestingnamehere, reviewed and merged by @jabraham17]
mppf · Feb 28, 2025 · 8637c0a · 8637c0a
2 parents cb4cae9 + 441bb44
commit 8637c0a
Show file tree

Hide file tree

Showing 311 changed files with 3,537 additions and 76,059 deletions.
diff --git a/third-party/qthread/Makefile b/third-party/qthread/Makefile
@@ -5,6 +5,8 @@ endif
 CHPL_MAKE_HOST_TARGET = --target
 include $(CHPL_MAKE_HOME)/make/Makefile.base
 
+CMAKE ?= cmake
+
 export CHPL_MAKE_LOCALE_MODEL
 
 #
@@ -15,33 +17,18 @@ QTHREAD_INSTALL_DIR = $(QTHREAD_ABS_DIR)/$(QTHREAD_INSTALL_SUBDIR)
 QTHREAD_BUILD_DIR = $(QTHREAD_ABS_DIR)/$(QTHREAD_BUILD_SUBDIR)
 QTHREAD_DIR = $(QTHREAD_ABS_DIR)
 
-#
-# Cray builds are cross-compilations
-#
-ifneq (, $(filter cray-x%,$(CHPL_MAKE_TARGET_PLATFORM)))
-CHPL_QTHREAD_CFG_OPTIONS += --host=$(CHPL_MAKE_TARGET_ARCH)-cle-linux-gnu
-else ifneq (, $(filter cray-cs hpe-cray-ex,$(CHPL_MAKE_TARGET_PLATFORM)))
-CHPL_QTHREAD_CFG_OPTIONS += --host=$(CHPL_MAKE_TARGET_ARCH)-unknown-linux-gnu
-endif
-
 ifneq ($(CHPL_MAKE_HWLOC),none)
   # Have Qthreads get its hwloc topology from the Chapel runtime,
   # unless directed not to.
   TOPOLOGY=binders
   ifeq (, $(call isTrue, $(CHPL_QTHREAD_DONT_GET_TOPO_FROM_RT)))
     CFLAGS_NEEDS_RT_INCLUDES = y
-    CHPL_QTHREAD_CFG_OPTIONS += --with-hwloc-get-topology-function="chpl_topo_getHwlocTopology()"
+    CHPL_QTHREAD_CFG_OPTIONS += -DQTHREADS_HWLOC_GET_TOPOLOGY_FUNCTION="chpl_topo_getHwlocTopology()"
   endif
   ifeq ($(CHPL_MAKE_HWLOC),bundled)
-    CHPL_QTHREAD_CFG_OPTIONS += --with-hwloc=$(HWLOC_INSTALL_DIR)
-    ifdef CHPL_HWLOC_PREFIX
-      CHPL_QTHREAD_CFG_OPTIONS += --with-hwloc-symbol-prefix=$(CHPL_HWLOC_PREFIX)hwloc_
-    endif
-    # don't bother checking if we can link against hwloc
-    CHPL_QTHREAD_CFG_OPTIONS += --disable-hwloc-configure-checks
+    CHPL_QTHREAD_CFG_OPTIONS += -DHWLOC_INSTALL_DIR=$(HWLOC_INSTALL_DIR)
   else ifeq ($(CHPL_MAKE_HWLOC),system)
-    CHPL_QTHREAD_CFG_OPTIONS += --with-hwloc=$(shell $(CHPL_MAKE_PYTHON) $(CHPL_MAKE_HOME)/util/chplenv/chpl_hwloc.py --prefix)
-    CHPL_QTHREAD_CFG_OPTIONS += --disable-hwloc-configure-checks
+    CHPL_QTHREAD_CFG_OPTIONS += -DHWLOC_INSTALL_DIR=$(shell $(CHPL_MAKE_PYTHON) $(CHPL_MAKE_HOME)/util/chplenv/chpl_hwloc.py --prefix)
   endif
 endif
 
@@ -50,7 +37,7 @@ TOPOLOGY = $(CHPL_QTHREAD_TOPOLOGY)
 endif
 
 ifneq (, $(TOPOLOGY))
-CHPL_QTHREAD_CFG_OPTIONS += --with-topology=$(TOPOLOGY)
+CHPL_QTHREAD_CFG_OPTIONS += -DQTHREADS_TOPOLOGY=$(TOPOLOGY)
 endif
 
 # Used to create qthread-chapel.h
@@ -62,7 +49,7 @@ endif
 
 # Have qthreads use Chapel's allocator, unless directed not to
 ifeq (, $(call isTrue, $(CHPL_QTHREAD_NO_CHPL_ALLOC)))
-  CHPL_QTHREAD_CFG_OPTIONS += --with-alloc=chapel
+  CHPL_QTHREAD_CFG_OPTIONS += -DQTHREADS_ALLOC=chapel
   CFLAGS_NEEDS_RT_INCLUDES = y
 endif
 
@@ -120,17 +107,13 @@ endif
 HAVE_GUARD_PAGES = 0
 ifeq (, $(call isTrue, $(CHPL_QTHREAD_NO_GUARD_PAGES)))
 HAVE_GUARD_PAGES = 1
-CHPL_QTHREAD_CFG_OPTIONS += --enable-guard-pages
+CHPL_QTHREAD_CFG_OPTIONS += -DQTHREADS_GUARD_PAGES=ON
 endif
 
-# optionally enable printing compilation commands, etc. (to help diagnose
-# build problems)
 ifneq (, $(CHPL_QTHREAD_LOUD_RULES))
-CHPL_QTHREAD_CFG_OPTIONS += --disable-silent-rules
+CHPL_QTHREAD_CFG_OPTIONS += -DCMAKE_VERBOSE_MAKEFILE=ON
 endif
 
-CHPL_QTHREAD_CFG_OPTIONS += --enable-static --disable-shared
-
 # determine which scheduler to use based on locale model.  Override with a user
 # provided option if they requested one.
 SCHEDULER = nemesis
@@ -144,33 +127,17 @@ endif
 ifneq (, $(CHPL_QTHREAD_SCHEDULER))
 SCHEDULER = $(CHPL_QTHREAD_SCHEDULER)
 endif
-CHPL_QTHREAD_CFG_OPTIONS += --with-scheduler=$(SCHEDULER)
+CHPL_QTHREAD_CFG_OPTIONS += -DQTHREADS_SCHEDULER=$(SCHEDULER)
 
 # only do this on M1+ Macs
 ifeq ($(CHPL_MAKE_TARGET_ARCH),arm64)
 ifeq ($(CHPL_MAKE_TARGET_PLATFORM),darwin)
-  CHPL_QTHREAD_CFG_OPTIONS += --with-cacheline-width=128
+  CHPL_QTHREAD_CFG_OPTIONS += -DQTHREADS_CACHELINE_SIZE_ESTIMATE=128
 endif
 endif
 
 # reduce performance penalty in cases where numChapelTasks < numQthreadWorkers
-CHPL_QTHREAD_CFG_OPTIONS += --enable-condwait-queue
-
-# pgi doesn't support the proper atomic intrinsics that are required for
-# spinlocks to be fast, so disable them for pgi, and PrgEnv-pgi
-ifneq (, $(findstring pgi,$(CHPL_MAKE_TARGET_COMPILER)))
-CHPL_QTHREAD_CFG_OPTIONS += --disable-internal-spinlock
-endif
-
-# Address sanitizer doesn't work with asm context switching so try
-# makecontext/swapcontext (which may have false-positives, but doesn't
-# immediately fail)
-ifneq (,$(CHPL_QTHREAD_SANITIZER_SUPPORT))
-  CHPL_QTHREAD_CFG_OPTIONS += --disable-fastcontext
-  ifeq ($(CHPL_MAKE_TARGET_ARCH),arm64)
-    CFLAGS += -D_XOPEN_SOURCE
-  endif
-endif
+CHPL_QTHREAD_CFG_OPTIONS += -DQTHREADS_CONDWAIT_QUEUE=ON
 
 ifeq ($(CHPL_LIB_PIC),pic)
 CFLAGS += $(SHARED_LIB_CFLAGS)
@@ -193,35 +160,28 @@ clobber: FORCE
 
 
 qthread-config: FORCE
-	@test -z "$(FAILBUILD)" || ( echo 'compileline failed!' && false )
-
-#
-# These first few lines touch a bunch of autoconf-oriented files in a
-# certain order to prevent autoconf from running again; otherwise, we
-# ran into issues if a user's autoconf environment was not as far
-# ahead in version numbers as that which was used when packaging the
-# Qthreads release
-#
-	cd $(QTHREAD_SUBDIR) && touch -c configure.ac
-	cd $(QTHREAD_SUBDIR) && find . -name "*.m4" | xargs touch
-	sleep 1
-	cd $(QTHREAD_SUBDIR) && touch -c aclocal.m4
-	sleep 1
-	cd $(QTHREAD_SUBDIR) && touch configure
-	cd $(QTHREAD_SUBDIR) && find . -name "*.in" | xargs touch
-
-#
-# Then configure
-#
 	mkdir -p $(QTHREAD_BUILD_DIR)
-	cd $(QTHREAD_BUILD_DIR) \
-	&& $(QTHREAD_SUBDIR)/configure CC='$(CC)' CFLAGS='$(CFLAGS)' \
-	       CXX='$(CXX)'  CXXFLAGS='$(CFLAGS)' LDFLAGS='$(RUNTIME_LFLAGS) $(LDFLAGS)' \
-	       --prefix=$(QTHREAD_INSTALL_DIR) $(CHPL_QTHREAD_CFG_OPTIONS)
+	$(CMAKE) -B $(QTHREAD_BUILD_DIR) -S $(QTHREAD_SUBDIR) \
+	$(CHPL_QTHREAD_CFG_OPTIONS) \
+	-DQTHREADS_BUILD_TESTS=OFF \
+	-DCMAKE_INSTALL_PREFIX='$(QTHREAD_INSTALL_DIR)' \
+	-DCMAKE_INSTALL_LIBDIR='$(QTHREAD_LIB_DIR)' \
+	-DBUILD_SHARED_LIBS=OFF \
+	-DCMAKE_C_COMPILER='$(CC)' \
+	-DCMAKE_C_FLAGS='$(CFLAGS)' \
+	-DCMAKE_CXX_COMPILER='$(CXX)' \
+	-DCMAKE_CXX_FLAGS='$(CXXFLAGS)' \
+	-DCMAKE_MAKE_PROGRAM='$(MAKE)'
 
 qthread-build: FORCE
-	cd $(QTHREAD_BUILD_DIR) && $(MAKE)
-	cd $(QTHREAD_BUILD_DIR) && $(MAKE) install
+	if [ -f $(QTHREAD_BUILD_DIR)/Makefile ]; then cd $(QTHREAD_BUILD_DIR) && $(MAKE); else $(CMAKE) --build $(QTHREAD_BUILD_DIR); fi
+
+qthread-install: FORCE
+	if [ -f $(QTHREAD_BUILD_DIR)/Makefile ]; then cd $(QTHREAD_BUILD_DIR) && $(MAKE) install; else $(CMAKE) --install $(QTHREAD_BUILD_DIR); fi
+
+FORCE:
+
+.NOTPARALLEL:
 
 #
 # The two variables here answer different questions even though they're
@@ -258,10 +218,7 @@ qthread-chapel-h: FORCE
 	     $(USE_TOPOLOGY_BINDERS) \
 	     >> $(QTHREAD_INSTALL_DIR)/include/qthread-chapel.h
 
-qthread: qthread-config qthread-build qthread-chapel-h
-
-qthread-reconfig:
-	cd $(QTHREAD_SUBDIR) && autoreconf -f -i
+qthread: qthread-config qthread-build qthread-install qthread-chapel-h
 
 FORCE:
 

diff --git a/third-party/qthread/README b/third-party/qthread/README
@@ -2,7 +2,7 @@
 Qthreads README for Chapel
 ==========================
 
-This copy of Qthreads 1.21 is being released with Chapel for
+This copy of Qthreads 1.22 is being released with Chapel for
 convenience and was obtained from:
 
   https://github.com/sandialabs/qthreads
@@ -11,95 +11,19 @@ Any Chapel issues that seem to be related to Qthreads should be directed
 to the Chapel team at https://chapel-lang.org/bugs.html.
 
 
-Upgrading qthread versions
+Upgrading versions
 ============================
 
-The directory $CHPL_HOME/third-party/qthread/qthread-src contains the
-un-tarballed qthread package contents. Version updates should be done as
-follows, assuming the CWD is $CHPL_HOME/third-party/qthread/:
+The directory $CHPL_HOME/third-party/qthread/qthread-src contains
+a copy of the Qthreads repository. Version updates should be done
+as follows, assuming the CWD is $CHPL_HOME/third-party/qthread/:
 
-1. download the latest qthread version: e.g. `wget https://github.com/sandialabs/qthreads/releases/download/1.21/qthreads-1.21.tar.gz`
-2. `tar xf qthreads-1.21.tar.gz`
-3. `git rm -r qthread-src`
-4. `mv qthreads-1.21 qthread-src`
-5. `git add --force qthread-src` (--force to ignore our .gitignore)
-6. update the version number mentioned above
-7. test
-8. commit, PR, merge, etc
+1. download the latest qthread version: e.g. `wget https://github.com/sandialabs/qthreads/archive/refs/tags/1.22.tar.gz -O qthread-src.tar.gz`
+2. `git rm -r qthread-src`
+3. `tar xf qthread-src.tar.gz`
+4. `git add --force qthread-src` (--force to ignore our .gitignore)
+5. update the version number mentioned above
+6. test
+7. commit, PR, merge, etc
 
-Chapel modifications to Qthreads
-================================
 
-The modifications that we have made to the official Qthreads release are
-as follows:
-
-* Do not #include "qthread.hpp" ifdef CHPL_AVOID_CPP_CODE
-  https://github.com/chapel-lang/chapel/pull/25507
-
-```
-diff --git a/third-party/qthread/qthread-src/include/qthread/qthread.h b/third-party/qthread/qthread-src/include/qthread/qthread.h
-index 8269aa0949..bf022bc243 100644
---- a/third-party/qthread/qthread-src/include/qthread/qthread.h
-+++ b/third-party/qthread/qthread-src/include/qthread/qthread.h
-@@ -709,7 +709,7 @@ uint64_t qthread_syncvar_incrF(syncvar_t *restrict operand, uint64_t inc);
-
- Q_ENDCXX /* */
-
--#ifndef __cplusplus
-+#if !defined __cplusplus || defined CHPL_AVOID_CPP_CODE
-
- #define qthread_incr(ADDR, INCVAL) __sync_fetch_and_add(ADDR, INCVAL)
-
-```
-* Backport a few performance fixes from the development version of qthreads.
-  https://github.com/chapel-lang/chapel/pull/26138
-
-```
-diff --git a/include/qt_atomics.h b/include/qt_atomics.h
-index 2ac887ed..2312c954 100644
---- a/include/qt_atomics.h
-+++ b/include/qt_atomics.h
-@@ -21,8 +21,24 @@
- #define THREAD_FENCE_MEM_RELEASE THREAD_FENCE_MEM_RELEASE_IMPL
- #endif
-
-+#if QTHREAD_ASSEMBLY_ARCH == QTHREAD_IA32 ||                                   \
-+  QTHREAD_ASSEMBLY_ARCH == QTHREAD_AMD64
- #define SPINLOCK_BODY()                                                        \
--  do { MACHINE_FENCE; } while (0)
-+  do { __asm__ __volatile__("pause" ::: "memory"); } while (0)
-+#elif QTHREAD_ASSEMBLY_ARCH == QTHREAD_ARM ||                                  \
-+  QTHREAD_ASSEMBLY_ARCH == QTHREAD_ARMV8_A64
-+#define SPINLOCK_BODY()                                                        \
-+  do { __asm__ __volatile__("yield" ::: "memory"); } while (0)
-+#elif QTHREAD_ASSEMBLY_ARCH == QTHREAD_POWERPC64 ||                            \
-+  QTHREAD_ASSEMBLY_ARCH == QTHREAD_POWERPC32
-+// For whatever reason the 29 (mdoio) version of this instruction performed
-+// better in some back-of-the-envelope benchmarking so we're using that.
-+#define SPINLOCK_BODY()                                                        \
-+  do { __asm__ __volatile__("or 29,29,29" ::: "memory"); } while (0)
-+#else
-+#define SPINLOCK_BODY()                                                        \
-+  do { atomic_thread_fence(memory_order_acq_rel); } while (0)
-+#endif
-
- #define QTHREAD_FASTLOCK_SETUP()                                               \
-   do {                                                                         \
-```
-
-* Backport a few performance fixes from the development version of qthreads.
-  part of https://github.com/chapel-lang/chapel/pull/26328, the rest was
-  reverted in https://github.com/chapel-lang/chapel/pull/26468
-
-```
---- a/src/io.c
-+++ b/src/io.c
-@@ -74,7 +74,6 @@ static void qt_blocking_subsystem_internal_freemem(void) { /*{{{*/
- static void *qt_blocking_subsystem_proxy_thread(void *Q_UNUSED(arg)) { /*{{{*/
-   while (!atomic_load_explicit(&proxy_exit, memory_order_relaxed)) {
-     if (qt_process_blocking_call()) { break; }
--    MACHINE_FENCE;
-   }
-   atomic_fetch_sub_explicit(&io_worker_count, 1, memory_order_relaxed);
-   pthread_exit(NULL);
-```
diff --git a/third-party/qthread/qthread-src/.autogen-version b/third-party/qthread/qthread-src/.autogen-version