数组大小增加时，CUDA内核不起作用

Question

我正在尝试与Cuda进行矩阵乘法。这是我的代码：

SELECT
    PVXME.MT,
    PVXME.MT_VERSION,
    PVXME.START_DATE_LOCAL,
    PVXMEDS.CREATION_DATE_LOCAL,
    PVXMIHS.USER_NAME,
    PVXMEDE.PAT_ID,
    PVXMEDE.STRING_VALUE,
    PVXMEDE.NUM_VALUE AS MEDE_NUM_VALUE,
    RVXMIIF2.II,
    RVXMIIF2.NUM_VALUE AS MIIF_NUM_VALUE
FROM 
    PVXMEDE
RIGHT OUTER JOIN 
    PVXMEDG P1 ON P1.ME = PVXMEDE.ME    -- use table alias P1 here
RIGHT OUTER JOIN 
    PVXMEDG P2 ON P2.SEQUENCE = PVXMEDE.SEQUENCE  -- use table alias P2 here
               AND P2.SOURCE_TP = PVXMEDE.SOURCE_TP 
               AND P2.SOURCE_ID = PVXMEDE.SOURCE_ID 
               AND P2.SOURCE_VERSION = PVXMEDE.SOURCE_VERSION
RIGHT OUTER JOIN 
    PVXMEDS ON PVXMEDS.ME = P2.ME 
            AND PVXMEDS.SEQUENCE = P2.SEQUENCE
RIGHT OUTER JOIN 
    PVXME ON PVXME.ME = PVXMEDS.ME AND PVXME.MT = ?????? -- ** COMPLETE THIS! **
RIGHT OUTER JOIN 
    PVXMI ON PVXMI.MI = PVXME.MI
LEFT OUTER JOIN 
    PVXMIHS ON PVXMIHS.MI = PVXME.MI
LEFT OUTER JOIN 
    PVXMIID ON PVXME.ME = PVXMIID2.ME
LEFT OUTER JOIN 
    RVXMIII ON PVXMIID2.MI = RVXMIIF2.MI
            AND PVXMIID2.ID = RVXMIIF2.ID 
            AND PVXMIID2.ID_SEQUENCE = RVXMIIF2.ID_SEQUENCE 
            AND PVXMIID2.ME = RVXMIIF2.ME
WHERE
    (PVXMIHS.USER_NAME <> 'sipat'
    AND PVXMIHS.WHAT = 'MethodPrepare'
    AND PVXME.CX_STRING_4  =  '20190117-7h40m'
    AND PVXME.MT  LIKE  'MK-0431%Tab CA%'
    AND PVXMEDE.PAT_ID  NOT LIKE  '%Spectrum')
    AND (RVXMIIF2.II  =  'LeverageLimit'OR RVXMIIF2.II  =  'XresidualLimit')

当ARRAY_SIZE为256、512或1024时，一切正常且超快，但当ARRAY_SIZE为1280时，CUDA函数中没有任何反应。当我进行错误检查（我希望我做对了）时，我得到以下信息：

  

GPUassert：无效的配置参数

但是问题出在ARRAY_SIZE上。