From 46445a92128540549a0be5dfca532b5cd6692284 Mon Sep 17 00:00:00 2001
From: zhongsunjian1 <zhongsunjian1@huawei.com>
Date: Mon, 26 May 2025 15:21:41 +0800
Subject: [PATCH] ut_test

---
 .../mix/attention/test_flash_attention.cpp    | 91 -------------------
 1 file changed, 91 deletions(-)

diff --git a/tests/unittest/kernels/mix/attention/test_flash_attention.cpp b/tests/unittest/kernels/mix/attention/test_flash_attention.cpp
index 0774ddd9..a72d57bb 100644
--- a/tests/unittest/kernels/mix/attention/test_flash_attention.cpp
+++ b/tests/unittest/kernels/mix/attention/test_flash_attention.cpp
@@ -1630,95 +1630,4 @@ TEST(TestFlashAttentionNd, FlashAttentionTestKVPtr_relayAttention_bigbatchkvhead
     SVector<std::string> inDataFiles = Mki::Test::GetInDataFiles(atbopsHomeDir, subDir);
     Status status = opTest.RunWithDataFileKVPtr(opDesc, inTensorDesc, inDataFiles);
     ASSERT_EQ(status.Ok(), false);
-}
-
-TEST(TestFlashAttentionNd, FlashAttentionTestKVPtr_relayAttention_bigbatchkvheadcase5_batch60_kvhead8)
-{
-    // should generate data first
-    CHECK_DEVICE_VERSION_ASCEND910B();
-    const char *atbopsHome = std::getenv("ASDOPS_HOME_PATH");
-    ASSERT_NE(atbopsHome, nullptr);
-    std::string atbopsHomeDir(atbopsHome);
-    std::string subDir = "attention";
-    bool isRelay = true;
-    std::vector<int> shareIdx;
-    for(int i = 0; i < 100; i++) {
-        if(i % 3 == 0) {
-            shareIdx.push_back(0);
-        } else if (i % 2 == 0){
-            shareIdx.push_back(1);
-        } else if (i % 5 == 0){
-            shareIdx.push_back(-1);
-        } else {
-            shareIdx.push_back(2);
-        }
-    }
-    uint32_t shareLength = 10;
-    std::vector<int> shareLen;
-    for(int i = 0; i < shareLength; i++){
-        shareLen.push_back(1024);
-    }
-    std::ofstream shareIdxBin(atbopsHomeDir + "/../../../tests/unittest/kernels/mix/" + subDir + "/data/share_idx.bin", std::ios::binary);
-    shareIdxBin.write((const char *)shareIdx.data(), shareIdx.size() * sizeof(int));
-    shareIdxBin.close();
-    std::ofstream shareLenBin(atbopsHomeDir + "/../../../tests/unittest/kernels/mix/" + subDir + "/data/share_len.bin", std::ios::binary);
-    shareLenBin.write((const char *)shareLen.data(), shareLen.size() * sizeof(int));
-    shareLenBin.close();
-    std::string dataGen = "python3 " + atbopsHomeDir + "/../../../tests/unittest/kernels/mix/" + 
-                          subDir + "/data/unpad_FA_data_gen_renew.py 60 2048 8 1 32 128 2048 0 0 5 1 0 1";
-    int ret = system(dataGen.c_str());
-    ASSERT_EQ(WEXITSTATUS(ret), 0);
-    Mki::Test::MkiOpTest opTest;
-    std::vector<uint32_t> scalars; // {batch, embd, qHead, kvHead, maxSeq}
-    bool batchDynamic = false;
-    PrepareScalarParams(scalars, batchDynamic, opTest, atbopsHomeDir, subDir);
-    uint32_t batch = scalars[0];
-    uint32_t edim = scalars[1];
-    uint32_t qHead = scalars[2];
-    uint32_t kvHead = scalars[3];
-    uint32_t maxSeq = scalars[4];  
-    std::vector<int32_t> q_seqlen(batch, 0);
-    std::vector<int32_t> kv_seqlen(batch, 0);
-    std::vector<int32_t> batch_state(batch, 0);
-    PrepareVectors(q_seqlen, kv_seqlen, batch_state, opTest, atbopsHomeDir, subDir);
-    uint64_t qnTokens = 0;
-    for (auto &it : q_seqlen) {
-        qnTokens += it;
-    }
-    std::vector<uint16_t> expect(qnTokens * edim * qHead, 0);
-    std::vector<uint32_t> expect_high(qnTokens * edim * qHead, 0);
-    PrepareExpect(expect, expect_high, opTest, atbopsHomeDir, subDir);
-    std::string dtype = "fp16";
-    opTest.Golden(std::bind(FlashAttentionNdGolden, expect.data(), expect_high.data(), dtype, std::placeholders::_1));
-    
-    TensorDesc kvTensorDesc = {TENSOR_DTYPE_FLOAT16, TENSOR_FORMAT_ND, {maxSeq, kvHead * edim}};
-    std::vector<std::string> kvFiles;
-    opTest.shareIdx = shareIdx;
-    opTest.shareLen = shareLen;
-    opTest.isRelay = isRelay;
-    PrepareKvFiles(batch, kvFiles, opTest, atbopsHomeDir, subDir, isRelay, shareLength);
-    opTest.PrepareKVcacheBatchwiseTensors(batch, kvTensorDesc, kvFiles, shareLength);
-    OpParam::UnpadFlashAttention opParam;
-    opParam.type = OpParam::UnpadFlashAttention::RELAY_ATTENTION_DECODER_ND;
-    opParam.qSeqLen = q_seqlen;
-    opParam.kvSeqLen = kv_seqlen;
-    opParam.headSize = qHead;
-    opParam.kvHead = kvHead;
-    opParam.tor = 1;
-    opParam.kShareTensorList = opTest.kSepCacheBatchWise;
-    opParam.vShareTensorList = opTest.vSepCacheBatchWise;
-    opParam.kTensorList = opTest.kCacheBatchWise;
-    opParam.vTensorList = opTest.vCacheBatchWise;
-    opParam.kvShareMap = shareIdx;
-    opParam.kvShareLen = shareLen;
-    opParam.maskType = OpParam::UnpadFlashAttention::MASK_TYPE_NONE;
-    Mki::Test::UtOpDesc opDesc = {"UnpadFlashAttentionOperation", opParam};
-    
-    SVector<TensorDesc> inTensorDesc = {
-        {TENSOR_DTYPE_FLOAT16, TENSOR_FORMAT_ND, {qnTokens, qHead * edim}},       // q
-        {TENSOR_DTYPE_FLOAT16, TENSOR_FORMAT_ND, {}},                       // mask
-    };
-    SVector<std::string> inDataFiles = Mki::Test::GetInDataFiles(atbopsHomeDir, subDir);
-    Status status = opTest.RunWithDataFileKVPtr(opDesc, inTensorDesc, inDataFiles);
-    ASSERT_EQ(status.Ok(), true);
 }
\ No newline at end of file
-- 
Gitee