From 23377f944a59b7022a2fc2c27493ba27ca45108d Mon Sep 17 00:00:00 2001
From: Andreas Kloeckner <inform@tiker.net>
Date: Mon, 19 Oct 2020 13:13:55 -0500
Subject: [PATCH 1/3] Drop paths-ignore from github CI config

---
 .github/workflows/ci.yml | 2 --
 1 file changed, 2 deletions(-)

diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml
index 914c5f7d..f6c48a87 100644
--- a/.github/workflows/ci.yml
+++ b/.github/workflows/ci.yml
@@ -4,8 +4,6 @@ on:
         branches:
         - master
     pull_request:
-        paths-ignore:
-        - 'doc/*.rst'
     schedule:
         - cron:  '17 3 * * 0'
 
-- 
GitLab


From 87d5a7ed33120125dece420dbf69b2391f663263 Mon Sep 17 00:00:00 2001
From: Andreas Kloeckner <inform@tiker.net>
Date: Mon, 19 Oct 2020 13:14:18 -0500
Subject: [PATCH 2/3] Add command_qeueue::get_hex_device_version

---
 src/wrap_cl.hpp | 35 ++++++++++++++++++++++++++++++++++-
 1 file changed, 34 insertions(+), 1 deletion(-)

diff --git a/src/wrap_cl.hpp b/src/wrap_cl.hpp
index 6224866a..ad5b7ab8 100644
--- a/src/wrap_cl.hpp
+++ b/src/wrap_cl.hpp
@@ -1225,7 +1225,7 @@ namespace pyopencl
         errno = 0;
         int match_count = sscanf(plat_version.c_str(), "OpenCL %d.%d ", &major_ver, &minor_ver);
         if (errno || match_count != 2)
-          throw error("Context._get_hex_version", CL_INVALID_VALUE,
+          throw error("Context._get_hex_platform_version", CL_INVALID_VALUE,
               "Platform version string did not have expected format");
 
         return major_ver << 12 | minor_ver << 4;
@@ -1576,6 +1576,39 @@ namespace pyopencl
       { PYOPENCL_CALL_GUARDED(clFlush, (m_queue)); }
       void finish()
       { PYOPENCL_CALL_GUARDED_THREADED(clFinish, (m_queue)); }
+
+      // not exposed to python
+      int get_hex_device_version() const
+      {
+        cl_device_id dev;
+
+        PYOPENCL_CALL_GUARDED(clGetCommandQueueInfo,
+            (m_queue, CL_QUEUE_DEVICE, sizeof(dev), &dev, nullptr));
+
+        std::string dev_version;
+        {
+          size_t param_value_size;
+          PYOPENCL_CALL_GUARDED(clGetDeviceInfo,
+              (dev, CL_DEVICE_VERSION, 0, 0, &param_value_size));
+
+          std::vector<char> param_value(param_value_size);
+          PYOPENCL_CALL_GUARDED(clGetDeviceInfo,
+              (dev, CL_DEVICE_VERSION, param_value_size,
+               param_value.empty( ) ? nullptr : &param_value.front(), &param_value_size));
+
+          dev_version =
+              param_value.empty( ) ? "" : std::string(&param_value.front(), param_value_size-1);
+        }
+
+        int major_ver, minor_ver;
+        errno = 0;
+        int match_count = sscanf(dev_version.c_str(), "OpenCL %d.%d ", &major_ver, &minor_ver);
+        if (errno || match_count != 2)
+          throw error("CommandQueue._get_hex_device_version", CL_INVALID_VALUE,
+              "Platform version string did not have expected format");
+
+        return major_ver << 12 | minor_ver << 4;
+      }
   };
 
   // }}}
-- 
GitLab


From f053edf6b83e5b0727dc9420bd5d7c61acf4de6e Mon Sep 17 00:00:00 2001
From: Andreas Kloeckner <inform@tiker.net>
Date: Mon, 19 Oct 2020 13:14:45 -0500
Subject: [PATCH 3/3] ImmediateAllocator: use clEnqueueMigrateMemObjects to
 force allocation on CL1.2+ devices

---
 src/wrap_mempool.cpp | 27 +++++++++++++++++++--------
 1 file changed, 19 insertions(+), 8 deletions(-)

diff --git a/src/wrap_mempool.cpp b/src/wrap_mempool.cpp
index 3f26a2f9..4ccb61f2 100644
--- a/src/wrap_mempool.cpp
+++ b/src/wrap_mempool.cpp
@@ -154,14 +154,25 @@ namespace
         // reported in a deferred manner, it has no way to react
         // (e.g. by freeing unused memory) because it is not part of
         // the call stack.)
-        unsigned zero = 0;
-        PYOPENCL_CALL_GUARDED(clEnqueueWriteBuffer, (
-              m_queue.data(),
-              ptr,
-              /* is blocking */ CL_FALSE,
-              0, std::min(s, sizeof(zero)), &zero,
-              0, NULL, NULL
-              ));
+        if (m_queue.get_hex_device_version() < 0x1020)
+        {
+          unsigned zero = 0;
+          PYOPENCL_CALL_GUARDED(clEnqueueWriteBuffer, (
+                m_queue.data(),
+                ptr,
+                /* is blocking */ CL_FALSE,
+                0, std::min(s, sizeof(zero)), &zero,
+                0, NULL, NULL
+                ));
+        }
+        else
+        {
+          PYOPENCL_CALL_GUARDED(clEnqueueMigrateMemObjects, (
+                m_queue.data(),
+                1, &ptr, CL_MIGRATE_MEM_OBJECT_CONTENT_UNDEFINED,
+                0, NULL, NULL
+                ));
+        }
 
         // No need to wait for completion here. clWaitForEvents (e.g.)
         // cannot return mem object allocation failures. This implies that
-- 
GitLab