pocl v6.0 (#103)

* updated v6.0

* MNT: Re-rendered with conda-build 24.5.1, conda-smithy 3.36.2, and conda-forge-pinning 2024.06.18.06.09.12

* rm patch

* rm patch 2

* Revert "rm patch 2"

This reverts commit 5fb52e8.

* re-apply POLLRDHUP part of gh1392 patch

* more pocld patches

* adjust patch

* adjust patch 2

* copy right bc files for CUDA

* disable test_large_buf

---------

Co-authored-by: Matthias Diener <[email protected]>

.ci_support/osx_64_libhwloc1.yaml

@@ -25,4 +25,4 @@ macos_machine:
 target_platform:
 - osx-64
 zlib:
-- '1.2'
+- '1'

.ci_support/osx_64_libhwloc2.yaml

@@ -25,4 +25,4 @@ macos_machine:
 target_platform:
 - osx-64
 zlib:
-- '1.2'
+- '1'

.ci_support/osx_arm64_.yaml

@@ -25,4 +25,4 @@ macos_machine:
 target_platform:
 - osx-arm64
 zlib:
-- '1.2'
+- '1'

recipe/build.sh

@@ -139,7 +139,7 @@ if [[ "$CONDA_BUILD_CROSS_COMPILATION" != "1" ]]; then
 
   if [[ $target_platform == "linux-aarch64" ]]; then
     # These tests fail on aarch64
-    SKIP_TESTS="$SKIP_TESTS|test_printf_vectors|test_printf_vectors_ulong"
+    SKIP_TESTS="$SKIP_TESTS|test_printf_vectors|test_printf_vectors_ulong|test_large_buf"
   fi
 
   if [[ $target_platform == "linux-ppc64le" ]]; then

recipe/install-pkg.sh

@@ -5,7 +5,7 @@ mkdir -p $PREFIX/lib/pocl
 
 if [[ "$PKG_NAME" == "pocl-cuda" ]]; then
   mv libpocl-devices-cuda.so $PREFIX/lib/pocl/
-  mv kernel-nvptx64.bc $PREFIX/share/pocl/
+  mv kernel-nvptx64-*.bc $PREFIX/share/pocl/
   mv cuda $PREFIX/share/pocl/
 elif [[ "$PKG_NAME" == "pocl-cpu-minimal" ]]; then
   mv libpocl-devices-basic.so $PREFIX/lib/pocl/

recipe/meta.yaml

@@ -1,6 +1,6 @@
-{% set version = "5.0" %}
-{% set build_number = 6 %}
-{% set sha256 = "fd0bb6e50c2286278c11627b71177991519e1f7ab2576bd8d8742974db414549" %}
+{% set version = "6.0" %}
+{% set build_number = 0 %}
+{% set sha256 = "de9710223fc1855f833dbbf42ea2681e06aa8ec0464f0201104dc80a74dfd1f2" %}
 
 {% set llvm_version = "15.0.7" %}
 
@@ -14,10 +14,7 @@ source:
     patches:
       - patches/gh1392.patch
       - patches/alignof.diff  # [ppc64le]
-
-      # https://github.com/pocl/pocl/issues/1480, https://github.com/pocl/pocl/commit/ed97458ba737be9057f322594b7ff5701ad79b1f
-      # unnecessary after 5.0:
-      - patches/cache.diff
+      - patches/pocld.diff
 
 build:
   number: {{ build_number }}

recipe/patches/gh1392.patch

@@ -1,164 +1,3 @@
-From a7b707c80adbff86b7e66f5919875f8f40cc6af5 Mon Sep 17 00:00:00 2001
-From: Isuru Fernando <[email protected]>
-Date: Wed, 17 Jan 2024 00:42:55 -0600
-Subject: [PATCH 1/4] Fix ENABLE_LOADABLE_DRIVERS for remote
-
----
- include/pocl_cache.h           |  1 +
- lib/CL/devices/devices.c       | 12 ++++--------
- lib/CL/devices/remote/remote.c |  3 ++-
- lib/CL/pocl_cl.h               |  2 ++
- lib/CL/pocl_intfn.h            |  6 +++---
- 5 files changed, 12 insertions(+), 12 deletions(-)
-
-diff --git a/include/pocl_cache.h b/include/pocl_cache.h
-index c77f13e16d..41d0cdae54 100644
---- a/include/pocl_cache.h
-+++ b/include/pocl_cache.h
-@@ -64,6 +64,7 @@ int pocl_cache_write_spirv (char *spirv_path,
-                             const char *spirv_content,
-                             uint64_t file_size);
-
-+POCL_EXPORT
- int pocl_cache_write_generic_objfile (char *objfile_path,
-                                       const char *objfile_content,
-                                       uint64_t objfile_size);
-diff --git a/lib/CL/devices/devices.c b/lib/CL/devices/devices.c
-index 917486d3af..ce883d181c 100644
---- a/lib/CL/devices/devices.c
-+++ b/lib/CL/devices/devices.c
-@@ -100,9 +100,6 @@
-
- #ifdef BUILD_REMOTE_CLIENT
- #include "remote/remote.h"
--extern cl_int pocl_remote_setup_peer_mesh ();
--// from remote/remote.c
--extern const char *remote_device_name_ptr;
- #endif
-
- #define MAX_DEV_NAME_LEN 64
-@@ -660,12 +657,11 @@ pocl_init_devices ()
-
-           ++dev_index;
-         }
-+      if (pocl_device_ops[i].post_init != NULL)
-+        {
-+          pocl_device_ops[i].post_init(&pocl_device_ops[i]);
-+        }
-     }
--
--#ifdef BUILD_REMOTE_CLIENT
--  pocl_remote_setup_peer_mesh ();
--#endif
--
-   first_init_done = 1;
-   devices_active = 1;
- ERROR:
-diff --git a/lib/CL/devices/remote/remote.c b/lib/CL/devices/remote/remote.c
-index 42e9505cb3..c709b0d587 100644
---- a/lib/CL/devices/remote/remote.c
-+++ b/lib/CL/devices/remote/remote.c
-@@ -228,6 +228,7 @@ pocl_remote_init_device_ops (struct pocl_device_ops *ops)
-
-   ops->probe = pocl_remote_probe;
-   ops->init = pocl_remote_init;
-+  ops->post_init = pocl_remote_setup_peer_mesh;
-   // ops->uninit = pocl_remote_uninit;
-   // ops->reinit = pocl_remote_reinit;
-
-@@ -488,7 +489,7 @@ pocl_remote_init (unsigned j, cl_device_id device, const char *parameters)
- }
-
- cl_int
--pocl_remote_setup_peer_mesh ()
-+pocl_remote_setup_peer_mesh (struct pocl_device_ops *ops)
- {
-   return pocl_network_setup_peer_mesh ();
- }
-diff --git a/lib/CL/pocl_cl.h b/lib/CL/pocl_cl.h
-index d1e045acdd..9634f20900 100644
---- a/lib/CL/pocl_cl.h
-+++ b/lib/CL/pocl_cl.h
-@@ -434,6 +434,8 @@ struct pocl_device_ops {
-    *  parameters : optional environment with device-specific parameters
-    */
-   cl_int (*init) (unsigned j, cl_device_id device, const char *parameters);
-+  /* Device type initialization after all devices have been initialized */
-+  cl_int (*post_init) (struct pocl_device_ops *ops);
-   /* uninitializes the driver for a particular device. May free hardware resources. */
-   cl_int (*uninit) (unsigned j, cl_device_id device);
-   /* reinitializes the driver for a particular device. Called after uninit;
-diff --git a/lib/CL/pocl_intfn.h b/lib/CL/pocl_intfn.h
-index ba4f0c7d69..030d505be2 100644
---- a/lib/CL/pocl_intfn.h
-+++ b/lib/CL/pocl_intfn.h
-@@ -35,9 +35,9 @@ extern "C" {
- POdeclsym(clBuildProgram)
- POdeclsym(clLinkProgram)
- POdeclsym(clCompileProgram)
--POdeclsym(clCreateBuffer)
-+POdeclsymExport(clCreateBuffer)
- POdeclsym(clCreateCommandQueue)
--POdeclsym(clCreateContext)
-+POdeclsymExport(clCreateContext)
- POdeclsym(clCreateContextFromType)
- POdeclsym(clCreateImage2D) 
- POdeclsym(clCreateImage3D)
-@@ -93,7 +93,7 @@ POdeclsym(clGetKernelInfo)
- POdeclsym(clGetKernelArgInfo)
- POdeclsym(clGetKernelWorkGroupInfo)
- POdeclsym(clGetKernelSubGroupInfo)
--POdeclsym(clGetMemObjectInfo)
-+POdeclsymExport(clGetMemObjectInfo)
- POdeclsym(clGetPlatformIDs)
- POdeclsym(clGetPlatformInfo)
- POdeclsym(clGetProgramBuildInfo)
-
-From 6c32cd52953b822a580a8aab31f80062013af99c Mon Sep 17 00:00:00 2001
-From: Isuru Fernando <[email protected]>
-Date: Wed, 17 Jan 2024 00:43:53 -0600
-Subject: [PATCH 2/4] Fix remote for Apple
-
----
- CMakeLists.txt | 2 +-
- 1 file changed, 1 insertion(+), 1 deletion(-)
-
-diff --git a/CMakeLists.txt b/CMakeLists.txt
-index 27a950e3cf..dd3eea02c5 100644
---- a/CMakeLists.txt
-+++ b/CMakeLists.txt
-@@ -1079,7 +1079,7 @@ else()
-   set(TESTS_USE_ICD 0)
- endif()
-
--if (APPLE AND NOT ENABLE_ICD AND VISIBILITY_HIDDEN)
-+if (APPLE AND VISIBILITY_HIDDEN)
-   add_compile_definitions("-DCL_API_CALL=__attribute__ ((visibility (\"default\")))")
- endif()
-
-
-From c053673eeb1a9ac3d17f397169602bf5c482fca4 Mon Sep 17 00:00:00 2001
-From: Isuru Fernando <[email protected]>
-Date: Wed, 17 Jan 2024 00:43:09 -0600
-Subject: [PATCH 3/4] Fix ENABLE_LOADABLE_DRIVERS for cuda
-
----
- lib/llvmopencl/LLVMUtils.h | 1 +
- 1 file changed, 1 insertion(+)
-
-diff --git a/lib/llvmopencl/LLVMUtils.h b/lib/llvmopencl/LLVMUtils.h
-index 014105c292..341c2987ed 100644
---- a/lib/llvmopencl/LLVMUtils.h
-+++ b/lib/llvmopencl/LLVMUtils.h
-@@ -68,6 +68,7 @@ void eraseFunctionAndCallers(llvm::Function *Function);
-
- bool isAutomaticLocal(llvm::Function *F, llvm::GlobalVariable &Var);
-
-+POCL_EXPORT
- bool isGVarUsedByFunction(llvm::GlobalVariable *GVar, llvm::Function *F);
-
- // Checks if the given argument of Func is a local buffer.
-
 From aa5595f8713bb5b5068b439b9a51d586d9fc4979 Mon Sep 17 00:00:00 2001
 From: Isuru Fernando <[email protected]>
 Date: Wed, 17 Jan 2024 00:51:18 -0600
@@ -168,13 +7,13 @@ Subject: [PATCH 4/4] Fix typo
  pocld/pocld.cc | 2 +-
  1 file changed, 1 insertion(+), 1 deletion(-)
 
-diff --git a/pocld/pocld.cc b/pocld/pocld.cc
+diff --git a/pocld/daemon.cc b/pocld/daemon.cc
 index bd674cd148..efb81aa099 100644
---- a/pocld/pocld.cc
-+++ b/pocld/pocld.cc
-@@ -70,7 +70,7 @@
- #endif
- 
+--- a/pocld/daemon.cc
++++ b/pocld/daemon.cc
+@@ -50,7 +50,7 @@
+ #include "daemon.hh"
+
  #ifndef POLLRDHUP
 -#define PULLRDHUP 0
 +#define POLLRDHUP 0

recipe/patches/pocld.diff

@@ -0,0 +1,51 @@
+diff --git a/pocld/daemon.cc b/pocld/daemon.cc
+index bdf711cf8..f8aa50a1c 100644
+--- a/pocld/daemon.cc
++++ b/pocld/daemon.cc
+@@ -308,18 +308,25 @@ int PoclDaemon::launch(std::string ListenAddress, struct ServerPorts &Ports,
+   addrinfo *ai = ResolvedAddress;
+   NumListenFds = 0;
+   for (addrinfo *ai = ResolvedAddress; ai; ai = ai->ai_next) {
++#ifdef ENABLE_VSOCK
+     if (ai->ai_family != AF_INET && ai->ai_family != AF_INET6 &&
+         ai->ai_family != AF_VSOCK)
+       continue;
++#else
++    if (ai->ai_family != AF_INET && ai->ai_family != AF_INET6)
++      continue;
++#endif
+     struct sockaddr *base_addr = ai->ai_addr;
+     int base_addrlen = ai->ai_addrlen;
+     std::string addr_string = describe_sockaddr(base_addr, base_addrlen);
++#ifdef ENABLE_VSOCK
+     if (UseVsock && ai->ai_family != AF_VSOCK) {
+       POCL_MSG_ERR("Using vsock requires using the correct address "
+                    "vsock:<cid>, instead of %s\n",
+                    addr_string.c_str());
+       break;
+     }
++#endif
+     int listen_command_fd = 0;
+     int listen_stream_fd = 0;
+     struct sockaddr_storage server_addr_command, server_addr_stream;
+diff --git a/pocld/shared_cl_context.cc b/pocld/shared_cl_context.cc
+index 6bc358bbe..b1678af03 100644
+--- a/pocld/shared_cl_context.cc
++++ b/pocld/shared_cl_context.cc
+@@ -23,6 +23,7 @@
+    IN THE SOFTWARE.
+ */
+
++#include <unistd.h>
+ #include <cassert>
+ #include <cstdio>
+ #include <filesystem>
+@@ -630,7 +631,7 @@ SharedCLContext::SharedCLContext(cl::Platform *p, unsigned pid,
+     if (MaxMemAllocSize < MaxSVMAllocSize)
+       MaxSVMAllocSize = MaxMemAllocSize;
+
+-    MaxTotalAllocatableSVM = std::min(MaxTotalAllocatableSVM,
++    MaxTotalAllocatableSVM = std::min(MaxTotalAllocatableSVM, (size_t)
+                                       Dev.getInfo<CL_DEVICE_GLOBAL_MEM_SIZE>());
+   }
+