From fe97612800d2bf9afa2c72ccb32f03ea7bf5cee2 Mon Sep 17 00:00:00 2001
From: Yong Zhao <Yong.Zhao@amd.com>
Date: Fri, 10 Jan 2020 15:26:35 -0500
Subject: [PATCH] kfdtest: Add basic tests for XGMI SDMA queues

After XGMI SDMA queues were separated from regular SDMA queues, they
were not covered in the current tests. Add tests for them now.

Change-Id: I036e3ca5d583ab7f022a9dc6cda3ef867f4773a0
Signed-off-by: Yong Zhao <Yong.Zhao@amd.com>
---
 tests/kfdtest/scripts/kfdtest.exclude |  1 +
 tests/kfdtest/src/KFDQMTest.cpp       | 81 ++++++++++++++++++++++++++-
 2 files changed, 81 insertions(+), 1 deletion(-)

diff --git a/tests/kfdtest/scripts/kfdtest.exclude b/tests/kfdtest/scripts/kfdtest.exclude
index 92fdc31bbe..a19b08dda7 100644
--- a/tests/kfdtest/scripts/kfdtest.exclude
+++ b/tests/kfdtest/scripts/kfdtest.exclude
@@ -18,6 +18,7 @@ FILTER[core_sws]=\
 "KFDQMTest.SubmitNopSdmaQueue:"\
 "KFDQMTest.SubmitPacketSdmaQueue:"\
 "KFDQMTest.AllSdmaQueues:"\
+"KFDQMTest.AllXgmiSdmaQueues:"\
 "KFDQMTest.AllQueues:"\
 "KFDEventTest.SignalEvent"
 
diff --git a/tests/kfdtest/src/KFDQMTest.cpp b/tests/kfdtest/src/KFDQMTest.cpp
index 0d8a43886f..6a0d60db46 100644
--- a/tests/kfdtest/src/KFDQMTest.cpp
+++ b/tests/kfdtest/src/KFDQMTest.cpp
@@ -28,7 +28,7 @@
 #include "PM4Queue.hpp"
 #include "PM4Packet.hpp"
 #include "SDMAPacket.hpp"
-#include "SDMAQueue.hpp"
+#include "XgmiOptimizedSDMAQueue.hpp"
 #include "AqlQueue.hpp"
 #include <algorithm>
 
@@ -212,6 +212,9 @@ TEST_F(KFDQMTest, AllSdmaQueues) {
 
     const unsigned int numSdmaQueues = m_numSdmaEngines * m_numSdmaQueuesPerEngine;
 
+    LOG() << "Regular SDMA engines number: " << m_numSdmaEngines
+            << " SDMA queues per engine: " << m_numSdmaQueuesPerEngine << std::endl;
+
     HsaMemoryBuffer destBuf(bufSize << 1 , defaultGPUNode, false);
     HsaMemoryBuffer srcBuf(bufSize, defaultGPUNode, false);
     destBuf.Fill(0xFF);
@@ -243,6 +246,53 @@ TEST_F(KFDQMTest, AllSdmaQueues) {
     TEST_END
 }
 
+TEST_F(KFDQMTest, AllXgmiSdmaQueues) {
+    TEST_START(TESTPROFILE_RUNALL)
+
+    int defaultGPUNode = m_NodeInfo.HsaDefaultGPUNode();
+    int bufSize = PAGE_SIZE;
+    int j;
+    ASSERT_GE(defaultGPUNode, 0) << "failed to get default GPU Node";
+
+    const unsigned int numXgmiSdmaQueues =
+            m_numSdmaXgmiEngines * m_numSdmaQueuesPerEngine;
+
+    LOG() << "XGMI SDMA engines number: " << m_numSdmaXgmiEngines
+            << " SDMA queues per engine: " << m_numSdmaQueuesPerEngine << std::endl;
+
+    HsaMemoryBuffer destBuf(bufSize << 1 , defaultGPUNode, false);
+    HsaMemoryBuffer srcBuf(bufSize, defaultGPUNode, false);
+    destBuf.Fill(0xFF);
+
+    std::vector<XgmiOptimizedSDMAQueue> xgmiSdmaQueues(numXgmiSdmaQueues);
+
+    for (j = 0; j < numXgmiSdmaQueues; ++j)
+        ASSERT_SUCCESS(xgmiSdmaQueues[j].Create(defaultGPUNode));
+
+    for (j = 0; j < numXgmiSdmaQueues; ++j) {
+        destBuf.Fill(0x0);
+        srcBuf.Fill(j + 0xa0);
+        xgmiSdmaQueues[j].PlaceAndSubmitPacket(
+            SDMACopyDataPacket(xgmiSdmaQueues[j].GetFamilyId(),
+                    destBuf.As<unsigned int*>(), srcBuf.As<unsigned int*>(), bufSize));
+        xgmiSdmaQueues[j].PlaceAndSubmitPacket(
+            SDMAWriteDataPacket(xgmiSdmaQueues[j].GetFamilyId(),
+                    destBuf.As<unsigned int*>() + bufSize/4, 0x02020202));
+
+        xgmiSdmaQueues[j].Wait4PacketConsumption();
+
+        EXPECT_TRUE(WaitOnValue(destBuf.As<unsigned int*>() + bufSize/4, 0x02020202));
+
+        EXPECT_SUCCESS(memcmp(
+            destBuf.As<unsigned int*>(), srcBuf.As<unsigned int*>(), bufSize));
+    }
+
+    for (j = 0; j < numXgmiSdmaQueues; ++j)
+        EXPECT_SUCCESS(xgmiSdmaQueues[j].Destroy());
+
+    TEST_END
+}
+
 TEST_F(KFDQMTest, AllQueues) {
     TEST_START(TESTPROFILE_RUNALL)
 
@@ -254,6 +304,8 @@ TEST_F(KFDQMTest, AllQueues) {
 
     const unsigned int numCpQueues = m_numCpQueues;
     const unsigned int numSdmaQueues = m_numSdmaEngines * m_numSdmaQueuesPerEngine;
+    const unsigned int numXgmiSdmaQueues =
+            m_numSdmaXgmiEngines * m_numSdmaQueuesPerEngine;
 
     HsaMemoryBuffer destBufCp(PAGE_SIZE, defaultGPUNode, false);
     destBufCp.Fill(0xFF);
@@ -264,6 +316,7 @@ TEST_F(KFDQMTest, AllQueues) {
 
     std::vector<PM4Queue> cpQueues(numCpQueues);
     std::vector<SDMAQueue> sdmaQueues(numSdmaQueues);
+    std::vector<XgmiOptimizedSDMAQueue> xgmiSdmaQueues(numXgmiSdmaQueues);
 
     for (i = 0; i < numCpQueues; ++i)
         ASSERT_SUCCESS(cpQueues[i].Create(defaultGPUNode)) << " QueueId=" << i;
@@ -271,6 +324,10 @@ TEST_F(KFDQMTest, AllQueues) {
     for (j = 0; j < numSdmaQueues; ++j)
         ASSERT_SUCCESS(sdmaQueues[j].Create(defaultGPUNode));
 
+    for (j = 0; j < numXgmiSdmaQueues; ++j)
+        ASSERT_SUCCESS(xgmiSdmaQueues[j].Create(defaultGPUNode));
+
+
     for (i = 0; i < numCpQueues; ++i) {
         cpQueues[i].PlaceAndSubmitPacket(PM4WriteDataPacket(destBufCp.As<unsigned int*>()+i*2, i, i));
 
@@ -295,12 +352,34 @@ TEST_F(KFDQMTest, AllQueues) {
             destBuf.As<unsigned int*>(), srcBuf.As<unsigned int*>(), bufSize));
     }
 
+    for (j = 0; j < numXgmiSdmaQueues; ++j) {
+        destBuf.Fill(0x0);
+        srcBuf.Fill(j + 0xa0);
+        xgmiSdmaQueues[j].PlaceAndSubmitPacket(
+            SDMACopyDataPacket(xgmiSdmaQueues[j].GetFamilyId(),
+                    destBuf.As<unsigned int*>(), srcBuf.As<unsigned int*>(), bufSize));
+        xgmiSdmaQueues[j].PlaceAndSubmitPacket(
+            SDMAWriteDataPacket(xgmiSdmaQueues[j].GetFamilyId(),
+                    destBuf.As<unsigned int*>() + bufSize/4, 0x02020202));
+
+        xgmiSdmaQueues[j].Wait4PacketConsumption();
+
+        EXPECT_TRUE(WaitOnValue(destBuf.As<unsigned int*>() + bufSize/4, 0x02020202));
+
+        EXPECT_SUCCESS(memcmp(
+            destBuf.As<unsigned int*>(), srcBuf.As<unsigned int*>(), bufSize));
+    }
+
+
     for (i = 0; i < numCpQueues; ++i)
        EXPECT_SUCCESS(cpQueues[i].Destroy());
 
     for (j = 0; j < numSdmaQueues; ++j)
         EXPECT_SUCCESS(sdmaQueues[j].Destroy());
 
+    for (j = 0; j < numXgmiSdmaQueues; ++j)
+        EXPECT_SUCCESS(xgmiSdmaQueues[j].Destroy());
+
     TEST_END
 }