From 6d4f79b71bb0876680e8713a4a68cfd5628673c5 Mon Sep 17 00:00:00 2001
From: "Zhang, Winston" <winston.zhang@intel.com>
Date: Wed, 1 Nov 2023 19:20:40 +0000
Subject: [PATCH] fix: hang during appendMemory

Fixed command buffer size estimation in appendMemoryCopy code path.

Related-To: NEO-8293

Signed-off-by: Zhang, Winston <winston.zhang@intel.com>
---
 .../source/cmdlist/cmdlist_hw_immediate.inl   |  3 ++-
 .../cmdlist/test_cmdlist_append_memory.cpp    | 22 +++++++++++++++++++
 2 files changed, 24 insertions(+), 1 deletion(-)

diff --git a/level_zero/core/source/cmdlist/cmdlist_hw_immediate.inl b/level_zero/core/source/cmdlist/cmdlist_hw_immediate.inl
index 2af2c25a92..359e36eda5 100644
--- a/level_zero/core/source/cmdlist/cmdlist_hw_immediate.inl
+++ b/level_zero/core/source/cmdlist/cmdlist_hw_immediate.inl
@@ -551,7 +551,8 @@ ze_result_t CommandListCoreFamilyImmediate<gfxCoreFamily>::appendMemoryCopy(
 
     auto estimatedSize = commonImmediateCommandSize;
     if (isCopyOnly()) {
-        auto nBlits = static_cast<size_t>(std::ceil(size / static_cast<double>(BlitterConstants::maxBlitWidth * BlitterConstants::maxBlitHeight)));
+        auto nBlits = size / (NEO::BlitCommandsHelper<GfxFamily>::getMaxBlitWidth(this->device->getNEODevice()->getRootDeviceEnvironment()) *
+                              NEO::BlitCommandsHelper<GfxFamily>::getMaxBlitHeight(this->device->getNEODevice()->getRootDeviceEnvironment(), true));
         auto sizePerBlit = sizeof(typename GfxFamily::XY_COPY_BLT) + NEO::BlitCommandsHelper<GfxFamily>::estimatePostBlitCommandSize(this->device->getNEODevice()->getRootDeviceEnvironment());
         estimatedSize += nBlits * sizePerBlit;
     }
diff --git a/level_zero/core/test/unit_tests/sources/cmdlist/test_cmdlist_append_memory.cpp b/level_zero/core/test/unit_tests/sources/cmdlist/test_cmdlist_append_memory.cpp
index 995be83fa7..9e1bc6ea45 100644
--- a/level_zero/core/test/unit_tests/sources/cmdlist/test_cmdlist_append_memory.cpp
+++ b/level_zero/core/test/unit_tests/sources/cmdlist/test_cmdlist_append_memory.cpp
@@ -607,6 +607,28 @@ HWTEST2_F(AppendMemoryCopy, givenSyncModeImmediateCommandListWhenAppendingMemory
     commandList->csr->getInternalAllocationStorage()->getTemporaryAllocations().freeAllGraphicsAllocations(device->getNEODevice());
 }
 
+HWTEST2_F(AppendMemoryCopy, givenImmediateCommandListWhenAppendingMemoryCopyThenNBlitsIsSuccessfullyCalculated, IsAtLeastSkl) {
+    Mock<CommandQueue> cmdQueue;
+    void *srcPtr = reinterpret_cast<void *>(0x1234);
+    void *dstPtr = reinterpret_cast<void *>(0x2345);
+
+    auto commandList = std::make_unique<WhiteBox<L0::CommandListCoreFamilyImmediate<gfxCoreFamily>>>();
+    ASSERT_NE(nullptr, commandList);
+    commandList->device = device;
+    commandList->cmdQImmediate = &cmdQueue;
+    commandList->cmdListType = CommandList::CommandListType::TYPE_IMMEDIATE;
+    commandList->csr = device->getNEODevice()->getDefaultEngine().commandStreamReceiver;
+    ze_result_t ret = commandList->initialize(device, NEO::EngineGroupType::Copy, 0u);
+    ASSERT_EQ(ZE_RESULT_SUCCESS, ret);
+
+    auto result = commandList->appendMemoryCopy(dstPtr, srcPtr, 8, nullptr, 0, nullptr, false, false);
+    EXPECT_TRUE(commandList->isCopyOnly());
+    ASSERT_EQ(ZE_RESULT_SUCCESS, result);
+
+    commandList->cmdQImmediate = nullptr;
+    commandList->csr->getInternalAllocationStorage()->getTemporaryAllocations().freeAllGraphicsAllocations(device->getNEODevice());
+}
+
 HWTEST2_F(AppendMemoryCopy, givenCommandListAndHostPointersWhenMemoryCopyCalledThenPipeControlWithDcFlushAdded, IsAtLeastSkl) {
     using PIPE_CONTROL = typename FamilyType::PIPE_CONTROL;