Memory test case description

The goal of this test case is to check communications between the Memory xbtest hardware IP (xbtIP) and memories available:

• On the card: Typically, DDRs.

• In the FPGA: For example, HBM.

Note

In the MMIO test case, xbtest application software (xbtSW) controls data transfers between host and PCIe BAR registers in the FPGA.

In the DMA test case, xbtSW controls data transfers between host and memories available on the card and in the FPGA.

In the Memory test case, xbtSW commands the Memory xbtIP to transfer data between the xbtIP and their associated memories located on the card, in the FPGA, or on the host.

The Memory test case includes the following features:

• All memories are tested in parallel.

• All memories of a type (for example DDR, HBM, or HOST) are tested with the same sequence (see Memory xbtest hardware IP types).

• Data integrity is checked using PRBS31 generator/checker within the xbtIP.

• Write and read bandwidths and latencies are measured.

• Mode of data transfer is configurable. Available modes are only/alternate/simultaneous write/read.

• Write/read rates or bandwidth requests, burst size, maximum number of outstanding transactions, block size and start address offset of the Memory xbtIP data transfers are configurable separately.

• All transfers are performed using linear addressing.

• Multiple AXI ID threads can be supported by the Memory xbtIP depending on the card.

Test parameters

The mandatory test configuration parameters are listed below. For more information, see Memory Test JSON Members.

• duration: Specifies the test duration in seconds.

• mode: Describes the data transfer mode: alternate_wr_rd, only_wr, only_rd or simultaneous_wr_rd.

Important

By default, xbtest loads a valid PRBS31 data sequence in the memory before any only_rd or simultaneous_wr_rd test. This can be disabled (disable_prewrite) and the presence of valid PRBS data in the memory can be managed manually within your test_sequence, defining at least one only_wr test prior to the first only_rd or simultaneous_wr_rd test.

The following optional parameters may also be specified:

• wr_rate: Write data transfer rate.

• rd_rate: Read data transfer rate.

• wr_bandwidth: Write data transfer bandwidth request.

• rd_bandwidth: Read data transfer bandwidth request.

• wr_burst_size: Write burst size.

• rd_burst_size: Read burst size.

• wr_outstanding: Maximum number of outstanding write transactions.

• rd_outstanding: Maximum number of outstanding read transactions.

• wr_block_size: Write block size.

• rd_block_size: Read block size.

• wr_start_addr: Write start address offset.

• rd_start_addr: Read start address offset.

Important

For some memory types, the Memory xbtIP has been intentionally designed to exceed the power capacity of the card. Your server/workstation may reboot or xbtest may be interrupted if you try to use a high write or read data transfer rates. Nominal write and read data transfer rates are specified in Card definition JSON file.

Main test steps

For each test configuration, the following steps are repeated:

1. The test is run for at least the defined duration. The entire range of the memory is always checked, meaning that, if needed, the test duration is extended.

2. Every second, xbtest application software (xbtSW) requests status and measurements from the xbtest hardware IP (xbtIP):

   • Read/write bandwidths and latencies.

   • Data integrity status.

3. After the test completes, xbtest application software (xbtSW) displays the measured average read/write bandwidth and latency. Bandwidths and latencies are also checked against thresholds under some conditions (see Bandwidth and latency check conditions).

Measurements

The Memory xbtest hardware IP (xbtIP) computes latency for each write and read burst. The bandwidth is computed every second by xbtest application software (xbtSW) after requesting measurements from the xbtIP.

Warning

Some cards include safety feature such as clock throttling when the power or temperature exceed pre-defined limits (refer to the documentation for your card). If the clock throttling has been activated while doing a memory test, the Memory xbtIP measurements are impacted as they depend on expected clock frequency and the Memory test case might fail.

• xbtest can detects clock throttling is on only when continuous clock has been connected to the xbtIP.

• If the memory test fails while the Power test case is running, a warning (MEM_051) is reported to alert on potential clock throttling.

Bandwidth

The average write/read bandwidth is computed as write/read burst size multiplied by number of write/read burst per second. The measured average bandwidth can be checked against thresholds when enabled.

Latency

The following sections describes the start and end point for latency measurement. The Memory test case reports average burst latency. The measured average latency can be checked against thresholds when enabled.

Latency

Latency	Description	Illustration
Write	The write burst latency is measured for a given burst using the following: Start point: WVALID, WLAST and WREADY are asserted. End point: BVALID and BREADY are asserted.	The following figure represents the write latency measurement: Write burst latency
Read	The read burst latency is measured for a given burst using the following: Start point: ARVALID and ARREADY are asserted. End point: First time RVALID and RREADY are asserted.	The following figure represents the read latency measurement. Read burst latency

Bandwidth and latency check conditions

The average read/write bandwidth and latency are checked against thresholds only when the following parameters are set to their nominal values:

• duration is greater than 20 seconds.

• wr_rate, rd_rate, wr_burst_size, rd_burst_size, wr_outstanding and rd_outstanding equal nominal values specified in Card definition.

• wr_block_size, rd_block_size, wr_start_addr and rd_start_addr are set such as the full memory size is check.

These nominal values are used by default. If any of them is overwritten, the check of the bandwidth and latency are disabled and a warning (MEM_050) is reported. This can be overruled by setting check_bw or check_latency if it is still required.

Memory xbtest hardware IP types

Different Alveo™ cards support various memory types of different sizes (for example, multiple 16 GB of DDR, 8 GB of HBM, 2 GB of PL-DDR, and/or 4 GB of PS-DDR) which are automatically detected by xbtest during the verify test case. Each of these memories is tested with either of these different types of Memory xbtest hardware IP (xbtIP):

• Single-channel: One or more xbtIP with 1 channel (1 AXI interface). Each xbtIP targets a different memory entirely. All Memory test cases for each single-channel Memory xbtIP run in parallel. For example, DDR, PL_DDR or HOST are single-channel memory types.

• multi-channel : Only one xbtIP with up to 32 channels targeting different areas of the same memory. All Memory test cases for each channel of the multi-channel Memory xbtIP run in parallel. For example, HBM, PS_DDR are multi-channel memory types.

Note

For example, as the HBM stack can be split into pseudo channels (PCs) (see AXI High Bandwidth Controller LogiCORE IP Product Guide (PG276)), a multi-channel Memory xbtIP is used. Each channel of the Memory xbtIP can be connected either one or more PCs.

For each memory types, the size of AXI W/RDATA buses can be different. The maximum data size is typically used (512 bits) but it can be lower for example for PS_DDR memory type (128 bits).

Note

Different Alveo™ cards might have different memory types. To identify the names of available memory types on the card selected with card BDF <BDF>, use the following command (see Command line options):

$ xbtest -d <BDF> -g memory

More information on the configuration is also available in xbtest.log, with message ID ITF_058 for each memory type and message ID ITF_100 for each xbtIP.

Memory xbtest hardware IP types

Type	Description	Illustration
32-channel HBM	One 32-channel Memory xbtIP, with each channel targets a different HBM PC (different memory tags).	32-channel HBM Memory xbtest hardware IP type
16-channel HBM	One 16-channel Memory xbtIP, with each channel targeting a different series of two HBM PCs (different memory tags).	16-channel HBM Memory xbtest hardware IP type
DDR	Four single-channel Memory xbtIP, with each xbtIP targeting a different DDR.	DDR Memory xbtest hardware IP type
PS_DDR	One 4-channel Memory xbtIP, with each channel targeting a different area of the same memory (same memory tag).	PS_DDR Memory xbtest hardware IP type

Write and read ranges/blocks

Data is transferred from/to the memory via multiple AXI bursts, which address linearly the entire range of the memory. By default, the entire range of the memory is tested. When the Memory xbtest hardware IP (xbtIP) is:

• Single-channel: The range is simply the size of the memory.

• multi-channel: The range is based on the quantity of memories connected to each channel of the Memory xbtIP. For example, with:

  • 32-channel HBM: Each channel of the Memory xbtIP has a range of 256 MB (size of one HBM Pseudo Channel).

  • 16-channel HBM: Each channel of the Memory xbtIP has a range of 512 MB (size of two HBM Pseudo Channel).

Throughout the documentation, the terminology of block is also used to refer to the range under test.

The wr_block_size and rd_block_size can be overwritten.

A block of data is fully:

• Written to the memory, when the last burst of data is acknowledged (BVALID) to the Memory xbtIP.

• Read from the memory, when the last data (RLAST) of the last burst is received by the Memory xbtIP.

For more information about AXI protocol, refer to Vivado Design Suite: AXI reference guide (UG1037).

Note

When a block of data is fully read/written, the Memory xbtIP repeats the transfer of data during the entire duration, but the memory range is always entirely transferred. The duration of the test is automatically increased accordingly, meaning that the reading/writing of a block is never interrupted.

Test modes

The Memory xbtest hardware IP (xbtIP) can be configured to write and/or read data to/from the targeted memory in four different modes set with parameter mode.

Test modes

Test mode	Description	Illustration
only_wr	The memory range (block) is fully written over and over during the entire duration of the test.	The following figure represents the blocks transferred during an only_wr test. only_wr test mode operations
only_rd	The memory range (block) is fully read over and over during the entire duration of the test. A single preliminary write (with known PRBS31 data) of the memory is performed prior starting the reading.	The following figure represents the blocks transferred during an only_rd test. only_rd test mode operations
simultaneous_wr_rd	The first half of the memory range is fully written and the second half of the memory range is read simultaneously. A single preliminary write (with known PRBS31 data) of the second half of the memory is performed prior starting the reading. The following figure represents the blocks transferred during a simultaneous_wr_rd test. In this mode, a block represents half of the memory range.	The following figure represents the blocks transferred during a simultaneous_wr_rd test. simultaneous_wr_rd test mode operations
alternate_wr_rd	The full range of the memory is written, then fully read, over and over during the entire duration of the test.	The following figure represents the blocks transferred during an alternate_wr_rd test. alternate_wr_rd test mode operations

AXI bursts

The Memory xbtest hardware IP (xbtIP) communicates with the memory via one or more AXI interfaces (channels). The size of the AXI burst is typically 4 kB. For more information about AXI protocol, refer to Vivado Design Suite: AXI reference guide (UG1037).

AXI data size vs. AXI beat

AXI data size (in bits)	AXI beat quantity	AXI burst size (in kB)
512	64	4
256	128	4
128	128	2

The wr_burst_size and rd_burst_size can be overwritten.

Maximum number of outstanding transactions

The Memory xbtest hardware IP (xbtIP) can be configured to limit the number of outstanding write and/or read transaction to/from the targeted memory.

The maximum number of outstanding transaction impact the Bandwidth and Latency measurements.

The wr_outstanding and rd_outstanding can be overwritten.

Maximum number of outstanding writes

An outstanding write transaction is defined as:

• Start point: when AWVALID and AWREADY are asserted.

• End point: when BVALID and BREADY are asserted.

Maximum Number of Outstanding Writes

Maximum number of outstanding writes	Illustration
1	The following figure represents the AXI bursts when the number of outstanding writes is limited to 1. Outstanding writes limited to 1.
2	The following figure represents the AXI bursts when the maximum number of outstanding writes is set to 2. Outstanding writes limited to 2
0 (not limited)	The following figure represents the AXI bursts when the number of outstanding writes is not limited. Outstanding writes not limited

Maximum number of outstanding reads

An outstanding read transaction is defined as:

• Start point: when ARVALID and ARREADY are asserted.

• End point: when RVALID, RLAST and RREADY are asserted.

Maximum number of outstanding reads

Maximum number of outstanding reads	Illustration
1	The following figure represents the AXI bursts when the number of outstanding reads is limited to 1. Outstanding reads limited to 1.
2	The following figure represents the AXI bursts when the maximum number of outstanding reads is set to 2. Outstanding reads limited to 2.
0 (not limited)	The following figure represents the AXI bursts when the number of outstanding reads is not limited. Outstanding reads not limited

AXI ID threads

The Memory xbtest hardware IP (xbtIP) can be configured to support multiple AXI ID threads (see AXI reference guide (UG1037)). The maximum number of AXI ID threads supported by the Memory xbtIP is reported by the host application in messages ITF_058.

By default, the Memory xbtIP will transfer data to/from the memory using the maximum number of ID thread. When the parameter single_axi_thread is set, the Memory xbtIP will only use a constant ID for all AXI transactions.

When multiple AXI ID threads are enabled, the Memory xbtIP will generate each consecutive AXI burst request with a different, rotating AXI ID. For example, if the number of AXI ID threads is 4, then the AXI ID of the consecutive requests will be: 0, 1, 2, 3, 0, 1, 2, 3, 0, etc.

Quality of Service (QoS) - Rate control

Depending on the memory controller built time settings, you may notice that write and read bandwidths are not even (or balanced) during simultaneous_wr_rd test. Although it may not change the total bandwidth available, you may want to change the way write and read bandwidths are shared. This can be achieved by using rate control, wr/rd_rate (see test_sequence).

By controlling individually wr_rate and rd_rate, you’ll be able to slow down (or accelerate) the quantity of write or read requests created by the Memory xbtest hardware IP (xbtIP). This rate control is still subject to Maximum number of outstanding transactions. So, you may need to increase or simply disable them to achieve the expected results.

All default settings are defined in Card definition. Changing any default settings automatically disables the check of the results against thresholds (when applicable), see Bandwidth and latency check conditions. If it’s not the desired behaviour, you’ll need to enable manually these checks in your test JSON file by using check_bw or check_latency.

Memory Test JSON Members

Target memories on the card

Following is an example of Memory test cases targeting all memories of type DDR and HBM available on the card. Note that all memories are tested in parallel.

"memory" : {
  "DDR": {
    "global_config": {
      "test_sequence": [
        { "duration": 30, "mode": "simultaneous_wr_rd" },
        { "duration": 30, "mode": "only_wr" },
        { "duration": 30, "mode": "only_rd" },
        { "duration": 30, "mode": "alternate_wr_rd" }
      ]
    }
  },
  "HBM": {
    "global_config": {
      "test_sequence": [
        { "duration": 30, "mode": "simultaneous_wr_rd" },
        { "duration": 30, "mode": "only_wr" },
        { "duration": 30, "mode": "only_rd" },
        { "duration": 30, "mode": "alternate_wr_rd" }
      ]
    }
  }
}

Target memory on the host

Following is an example of Memory test case targeting all memories of type HOST:

"memory" : {
  "HOST": {
    "global_config": {
      "test_sequence": [
        { "duration": 30, "mode": "simultaneous_wr_rd" },
        { "duration": 30, "mode": "only_wr" },
        { "duration": 30, "mode": "only_rd" },
        { "duration": 30, "mode": "alternate_wr_rd" }
      ]
    }
  }
}

Important

The Memory test case targeting memories available on the host should not be run in parallel with following test cases:

• Memory.

• Power.

• GT MAC.

• GTF/GTYP/GTM PRBS.

Single-channel override

For single-channel memory types, some test JSON members can be overwritten for each tag targeted by the Memory xbtest hardware IP (xbtIP) using the test JSON member memory_tag_config.

The following example shows how to run a Memory test case only for one (identified by memory tag DDR[1]) of the memories of the single-channel memory type named DDR.

"memory" : {
  "DDR": {
    "global_config": {
      "test_sequence": [
        { "duration": 30, "mode": "simultaneous_wr_rd" }
      ],
      "disable_memory": true
    },
    "memory_tag_config": {
      "DDR[1]": {
        "disable_memory": false
      }
    }
  }
}

Multi-channel override

For multi-channel memory types, some test JSON members can be overwritten for each channel of the Memory xbtest hardware IP (xbtIP) using the test JSON member memory_channel_config.

The following example shows how to run a Memory test case only for one (channel 12) channel of the multi-channel memory type named HBM.

"memory" : {
  "HBM": {
    "global_config": {
      "test_sequence": [
        { "duration": 30, "mode": "simultaneous_wr_rd" }
      ],
      "disable_memory": true
    },
    "memory_channel_config": {
      "12": {
        "disable_memory": false
      }
    }
  }
}

Definition

In the Memory test case, valid test JSON members depend on the configuration of the Memory xbtest hardware IP (xbtIP) (see Memory xbtest hardware IP types), which are automatically detected by xbtest application software (xbtSW) during the verify test case. The following table shows all members available for this test case. More details are provided for each member in the subsequent sections:

Memory test case members

Member	Mandatory / Optional	Description
test_sequence	Mandatory	Describes the sequence of tests to perform.
check_bw	Optional	Enable the check of the bandwidths. This check is by default for memories available: On the card: Enabled. On the host: Disabled.
hi_thresh_alt_wr_bw hi_thresh_alt_rd_bw hi_thresh_only_wr_bw hi_thresh_only_rd_bw hi_thresh_simul_wr_bw hi_thresh_simul_rd_bw	Optional	Overwrite high threshold of write/read bandwidth (MB/s).
lo_thresh_alt_wr_bw lo_thresh_alt_rd_bw lo_thresh_only_wr_bw lo_thresh_only_rd_bw lo_thresh_simul_wr_bw lo_thresh_simul_rd_bw	Optional	Overwrite low threshold of the write/read bandwidth (MB/s).
check_latency	Optional	Enable the check of the latencies. This check is by default for memories available: On the card: Enabled. On the host: Disabled.
hi_thresh_alt_wr_lat hi_thresh_alt_rd_lat hi_thresh_only_wr_lat hi_thresh_only_rd_lat hi_thresh_simul_wr_lat hi_thresh_simul_rd_lat	Optional	Overwrite high threshold of write/read latency (ns).
lo_thresh_alt_wr_lat lo_thresh_alt_rd_lat lo_thresh_only_wr_lat lo_thresh_only_rd_lat lo_thresh_simul_wr_lat lo_thresh_simul_rd_lat	Optional	Overwrite low threshold of the write/read latency (ns).
disable_prewrite	Optional	Disable write of valid PRBS31 data in the memory before any only_rd or simultaneous_wr_rd test.
disable_memory	Optional	Disable memory test case.
single_axi_thread	Optional	Disable usage of multiple AXI ID threads.
pattern	Optional	Define the data pattern transferred to/from the memory.
check_data_integrity	Optional	Enable the check of data integrity.

test_sequence

Mandatory. Describes the sequence of tests to perform. Tests are performed serially, and a failure in one test does not stop the sequence (the next test will be launched). There is no limitation to the length of the test sequence.

This field contains a list of tests, each test being defined by an object of key–value parameters pairs: [ {}, {}, {} ].

The following table defines the parameters supported in the Memory test sequence:

Memory test sequence parameters

Member	Mandatory / Optional	Description
duration	Mandatory	The duration of the test in seconds; Range [1, 232-1].
mode	Mandatory	Test mode; Possible value: alternate_wr_rd, only_wr, only_rd and simultaneous_wr_rd.
wr/rd_rate 1	Optional	Write/read rate in percent; Overwrites nominal rate specified in Card definition JSON file. Range: [1, 100];
wr/rd_bandwidth 1	Optional	Write/read bandwidth request in MBps; Based on AXI data size and AXI clock frequency, the host application will set the wr/rd_rate corresponding to requested bandwidth.
wr/rd_burst_size	Optional	Write/read burst size. Overwrites nominal burst size specified in Card definition JSON file, or defaults to its maximum when not specified. Range: Minimum: Two times the AXI data size. Maximum: 4KB when the AXI data size is 512 bits, and otherwise 128 times the AXI data size.
wr/rd_outstanding	Optional	Maximum number of outstanding write/read transaction; Overwrites nominal value specified in Card definition JSON file. Range: [0, 255]; A value of 0 means the number of outstanding writes/reads is not limited.
wr/rd_block_size	Optional	Optional write/read block size. Default value is such as the full range of memory is tested. Range depends on mode, memory size, wr_start_addr and rd_start_addr.
wr/rd_start_addr	Optional	Optional write/read start address offset. Default value is such as the full range of memory is tested. Range depends on mode, memory size, wr_block_size and rd_block_size.

Note

1(1,2)

wr/rd_rate and wr/rd_bandwidth parameters cannot be combined.

For example:

• Single test:

  • "test_sequence": [ { "duration": 60, "mode": "alternate_wr_rd" } ]
    

  • "test_sequence": [ { "duration": 60, "mode": "only_rd", "rd_rate": 25 } ]
    

  • "test_sequence": [ { "duration": 60, "mode": "only_wr", "wr_bandwidth": 6500 } ]
    

  • "test_sequence": [ { "duration": 60, "mode": "simultaneous_wr_rd", "wr_rate": 20, "rd_rate": 25, "wr_outstanding": 0, "rd_outstanding": 0 } ]
    

  • "test_sequence": [ { "duration": 60, "mode": "only_rd", "rd_burst_size": 2048 } ]
    

  • "test_sequence": [ { "duration": 60, "mode": "simultaneous_wr_rd", "wr_start_addr": 0, "wr_block_size": 128, "rd_start_addr": 128, "rd_block_size": 128 } ]
    

• Multiple tests:

  • "test_sequence": [
      { "duration": 60, "mode": "alternate_wr_rd" },
      { "duration": 60, "mode": "only_wr" },
      { "duration": 60, "mode": "only_rd" }
    ]
    

The write and read memory areas, defined by wr_start_addr, rd_start_addr, wr_block_size and rd_block_size:

• Must be the same when mode is alternate_wr_rd, such as:

  • Valid:

    "test_sequence": [ { "duration": 60, "mode": "alternate_wr_rd", "wr_start_addr": 0, "wr_block_size": 256, "rd_start_addr":   0, "rd_block_size": 256 } ]
    

  • Invalid:

    "test_sequence": [ { "duration": 60, "mode": "alternate_wr_rd", "wr_start_addr": 0, "wr_block_size": 256, "rd_start_addr": 128, "rd_block_size": 256 } ]
    

• Cannot overlap when mode is simultaneous_wr_rd, such as:

  • Valid:

    "test_sequence": [ { "duration": 60, "mode": "simultaneous_wr_rd", "wr_start_addr": 0, "wr_block_size": 256, "rd_start_addr": 512, "rd_block_size": 256 } ]
    

  • Invalid:

    "test_sequence": [ { "duration": 60, "mode": "simultaneous_wr_rd", "wr_start_addr": 0, "wr_block_size": 256, "rd_start_addr": 128, "rd_block_size": 256 } ]
    

Override test_sequence

The test_sequence can be overwritten for each channel of the multi-channel Memory xbtest hardware IP (xbtIP) or for each memory tag targeted by the single-channel Memory xbtIP.

• The number of tests in all test_sequence parameters specified for a memory type must be the same.

• The duration of a test must be the same in all test_sequence.

The following example shows how to run a Memory test case of the multi-channel memory type named HBM where mode is set to:

• only_rd, then only_wr for channel 1.

• only_wr, then only_rd for all other channels.

"memory" : {
  "HBM": {
    "global_config": {
      "test_sequence": [
        { "duration": 15, "mode": "only_wr" },
        { "duration": 25, "mode": "only_rd" }
      ]
    },
    "memory_channel_config": {
      "1": {
        "test_sequence": [
          { "duration": 15, "mode": "only_rd" },
          { "duration": 25, "mode": "only_wr" }
        ]
      }
    }
  }
}

The following example shows how to run a Memory test case of the multi-channel memory type named HBM where mode is set to:

• only_wr only for channel DDR[2].

• only_wr, then only_rd for all other DDR memory banks.

"memory" : {
  "DDR": {
    "global_config": {
      "test_sequence": [
        { "duration": 1,  "mode": "only_wr" },
        { "duration": 15, "mode": "only_rd" }
      ]
    },
    "memory_tag_config": {
      "DDR[2]": {
        "test_sequence": [
          { "duration": 1,  "mode": "only_wr" },
          { "duration": 15, "mode": "only_wr" }
        ]
      }
    }
  }
}

check_bw

Optional; Type : boolean; Possible values: true or false; Default : for memories available:

• On the card: true.

• On the host: false.

By setting this member to false, no average bandwidth measurement will be compared against defined thresholds. When set to true, average bandwidth measurements will be checked even if the test parameters are not nominal (see Bandwidth and latency check conditions).

Default bandwidth limits are defined in Card definition and are displayed at the beginning of the tests.

The bandwidth limits can be overwritten using the following parameters:

• hi_thresh_alt_wr_bw, hi_thresh_alt_rd_bw, hi_thresh_only_wr_bw, hi_thresh_only_rd_bw, hi_thresh_simul_wr_bw, hi_thresh_simul_rd_bw.

• lo_thresh_alt_wr_bw, lo_thresh_alt_rd_bw, lo_thresh_only_wr_bw, lo_thresh_only_rd_bw, lo_thresh_simul_wr_bw, lo_thresh_simul_rd_bw.

hi_thresh_alt_wr_bw, hi_thresh_alt_rd_bw, hi_thresh_only_wr_bw, hi_thresh_only_rd_bw, hi_thresh_simul_wr_bw, hi_thresh_simul_rd_bw

Optional; Type : integer; Range : [1, 2³²-1]; Default: specified in the Card definition.

Overwrite high threshold of the average write/read bandwidth (MB/s) specified in Card definition.

After all bandwidth measurements made during the test duration are complete, if the measured average bandwidth is greater than this threshold, the test fails.

The following table gives the write and read bandwidth high thresholds based on mode:

Bandwidth high thresholds

Test mode	Write BW threshold	Read BW threshold
alternate_wr_rd	hi_thresh_alt_wr_bw	hi_thresh_alt_rd_bw
only_wr	hi_thresh_only_wr_bw	n/a
only_rd	n/a	hi_thresh_only_rd_bw
simultaneous_wr_rd	hi_thresh_simul_wr_bw	hi_thresh_simul_rd_bw

lo_thresh_alt_wr_bw, lo_thresh_alt_rd_bw, lo_thresh_only_wr_bw, lo_thresh_only_rd_bw, lo_thresh_simul_wr_bw, lo_thresh_simul_rd_bw

Optional; Type : integer; Range : [1, 2³²-1]; Default: specified in the Card definition.

Overwrite low threshold of the average write/read bandwidth (MB/s) specified in Card definition.

After all bandwidth measurements made during the test duration are complete, if the measured average bandwidth is lower than this threshold, the test fails.

Low threshold must be lower than high threshold. The following table gives the write and read bandwidth low thresholds based on mode:

Bandwidth low thresholds

Test mode	Write BW threshold	Read BW threshold
alternate_wr_rd	lo_thresh_alt_wr_bw	lo_thresh_alt_rd_bw
only_wr	lo_thresh_only_wr_bw	n/a
only_rd	n/a	lo_thresh_only_rd_bw
simultaneous_wr_rd	lo_thresh_simul_wr_bw	lo_thresh_simul_rd_bw

check_latency

Optional; Type : boolean; Possible values: true or false; Default : for memories available:

• On the card: true.

• On the host: false.

By setting this member to false, no latency measurement (read or write) is checked against pass/failed criteria. When set to true, average latency measurements will be checked even if the test parameters are not nominal (see Bandwidth and latency check conditions).

Default latency limits are defined in Card definition and are displayed at the beginning of the tests.

The latency limits can be overwritten using the following parameters:

• hi_thresh_alt_wr_lat, hi_thresh_alt_rd_lat, hi_thresh_only_wr_lat, hi_thresh_only_rd_lat, hi_thresh_simul_wr_lat, hi_thresh_simul_rd_lat.

• lo_thresh_alt_wr_lat, lo_thresh_alt_rd_lat, lo_thresh_only_wr_lat, lo_thresh_only_rd_lat, lo_thresh_simul_wr_lat, lo_thresh_simul_rd_lat.

hi_thresh_alt_wr_lat, hi_thresh_alt_rd_lat, hi_thresh_only_wr_lat, hi_thresh_only_rd_lat, hi_thresh_simul_wr_lat, hi_thresh_simul_rd_lat

Optional; Type : integer; Range : [1, 2³²-1]; Default: specified in the Card definition.

Overwrite high threshold of the average write/read latency (ns) specified in Card definition.

After all latency measurements made during the test duration are complete, if the measured average latency is greater than this threshold, the test fails.

The following table gives the write and read latency high thresholds based on mode:

Latency high thresholds

Test mode	Write latency threshold	Read latency threshold
alternate_wr_rd	hi_thresh_alt_wr_lat	hi_thresh_alt_rd_lat
only_wr	hi_thresh_only_wr_lat	n/a
only_rd	n/a	hi_thresh_only_rd_lat
simultaneous_wr_rd	hi_thresh_simul_wr_lat	hi_thresh_simul_rd_lat

lo_thresh_alt_wr_lat, lo_thresh_alt_rd_lat, lo_thresh_only_wr_lat, lo_thresh_only_rd_lat, lo_thresh_simul_wr_lat, lo_thresh_simul_rd_lat

Optional; Type : integer; Range : [1, 2³²-1]; Default: specified in the Card definition.

Overwrite low threshold of the average write/read latency (ns) specified in Card definition.

After all latency measurements made during the test duration are complete, if the measured average latency is lower than this threshold, the test fails.

Low threshold must be lower than high threshold.

The following table gives the write and read latency high thresholds based on mode:

Latency low thresholds

Test mode	Write latency threshold	Read latency threshold
alternate_wr_rd	lo_thresh_alt_wr_lat	lo_thresh_alt_rd_lat
only_wr	lo_thresh_only_wr_lat	n/a
only_rd	n/a	lo_thresh_only_rd_lat
simultaneous_wr_rd	lo_thresh_simul_wr_lat	lo_thresh_simul_rd_lat

disable_prewrite

Optional; Type : boolean; Possible values: true or false; Default : false.

• When set to false, xbtest loads a valid PRBS31 data sequence in the memory before any only_rd or simultaneous_wr_rd test.

• When set to true, this is disabled. The presence of valid PRBS data in the memory can be managed manually within your test_sequence, defining at least one only_wr test prior to the first only_rd or simultaneous_wr_rd test.

disable_memory

Optional; Type : boolean; Possible values: true or false; Default : false.

By default, for all Memory xbtest hardware IP (xbtIP), all channels are enabled and the associated memory is checked.

single_axi_thread

Optional; Type : boolean; Possible values: true or false; Default : false.

By default, for all Memory xbtest hardware IP (xbtIP), the maximum number of ID threads supported is used to transfer data from/to the memory.

• When set to true, the Memory xbtIP use the same constant ID (set to 0) for all AXI transactions.

This parameter is supported only if the memory type supports multiple AXI ID threads.

pattern

Optional; Type: Object.

The data pattern transferred to/from the memory is defined using the parameters control and patterns defined below.

The following example shows how to run a 30-second write only test for memory type named DDR with pattern alternating between 0xFFFFFFFF and 0x00000000.

"memory" : {
  "DDR": {
    "global_config": {
      "test_sequence": [
        { "duration": 30,  "mode": "only_wr" }
      ],
      "pattern": {
        "control": "alternate",
        "patterns": ["0xFFFFFFFF", "0x00000000"]
      }
    }
  }
}

control

Optional; Type : string; Possible values: prbs, fixed or alternate; Default : prbs.

This parameter allows to change the type of data pattern transferred to/from the memory:

• When set to prbs, the pattern is a PRBS31. Data integrity is checked.

• When set to fixed, the data is constant. Data integrity is not checked.

• When set to alternate, the data alternates between two values for each AXI beat. Data integrity is not checked.

patterns

Optional; Type : List of string; Default : ["0xCAFECAFE", "0xF00DF00D"].

This parameter allows to change the values of data pattern transferred to/from the memory for:

• fixed mode: only one value must be provided.

• alternate mode: two values must be provided.

The pattern values in the list represent 32-bit hexadecimal values and must be provided with following format:

• Start wit 0x.

• Followed by 1 to 8 hexadecimal digits.

For example, provided value 0xAB corresponds to 32-bit value 0x000000AB.

The write data of the AXI beat is set by repeating the 32-bit value. For example, when a pattern value is set to 0x12345678, then WDATA is set to 0x1234567812345678...12345678.

check_data_integrity

Optional; Type : boolean; Possible values: true or false; Default : true.

By setting this member to false, the host application will not check that the data read from the memory by the memory xbtest hardware IP (xbtIP) matches the data written to the memory. When set to true, the data integrity is checked.

The data integrity is:

• Not checked when pattern control is set to fixed or alternate.

• Not applicable when the test sequence mode is set to only_wr.

Output files

All memory measurements are stored in output CSV files which are generated in xbtest logging directory. The values are stored in CSV type format with one column for each information type.

Important

If the command line option -L is used while calling xbtest application software (xbtSW), no output file is generated.

In the Memory test case, different CSV files are used to store all Memory test case results:

memory test case output files

Output file	Description	Naming convention
Results CSV output files	Configuration and results for each test of the test_sequence. These files are named with _detail.csv suffix.	Files are named with a prefix the following convention: Single-channel: memory_<tag>. Multi-channel: memory_<tag>_ch_<channel>. Where: <tag> is the formatted tag of the memory targeted by the Memory xbtIP (lower case letters, numbers, square brackets or underscores are used). <channel> is the index of the Memory xbtIP channel. In addition, for multi-channel memory types, another prefix is used: memory_<type>_combined, where <type> is the name of the memory type. These files contain the combined results for all channels of the Memory xbtIP: Bandwidth: Sum of the BW of each channel. Latency: Average of the latency of each channel.
Detail CSV output files	All intermediate bandwidth and latency results computed every second. These files are named with _detail.csv suffix.
Multi-channel Summary CSV output files	Contains the configuration and the results for each channels of the Memory xbtIP grouped to allow easier comparison of results between each channel. These files are generated for multi-channel memory types only.	Files are named memory_<type>_mc_summary.csv where <type> is the name of memory type.
Power CSV output files	Contains the configuration and all power measurements.	Files are named: Single-channel: memory_<type>_<tag>_power.csv. Multi-channel: memory_<type>_power.csv. Where: <type> is the name of memory type. <tag> is the formatted tag of the memory targeted by the Memory xbtIP (lower case letters, numbers, square brackets or underscores are used).

For example, depending on the memory type supported by xbtest for targeted Alveo Versal Example Design (AVED), the generated output CSV files are:

• One single-channel memory type named DDR with 2 Memory xbtest hardware IP (xbtIP) (one connected to DDR[0] and the other one to DDR[1]):

  • memory_ddr[0]_detail.csv.

  • memory_ddr[0]_result.csv.

  • memory_ddr[1]_detail.csv.

  • memory_ddr[1]_result.csv.

  • memory_DDR_ddr[0]_power.csv.

  • memory_DDR_ddr[1]_power.csv.

• One multi-channel memory type named HBM with a 16-channel Memory xbtest hardware IP (xbtIP) (channel 0 connected to HBM[0:1], channel 1 to HBM[2:3], … and channel 15 to HBM[30:31] ):

  • memory_hbm[0_1]_ch_0_detail.csv.

  • memory_hbm[0_1]_ch_0_result.csv.

  • memory_hbm[2_3]_ch_1_detail.csv.

  • memory_hbm[2_3]_ch_1_result.csv.

  • …

  • memory_hbm[30_31]_ch_15_detail.csv.

  • memory_hbm[30_31]_ch_15_result.csv.

  • memory_HBM_combined_detail.csv.

  • memory_HBM_combined_result.csv.

  • memory_HBM_mc_summary.csv.

  • memory_HBM_power.csv.

Results CSV output files

For each test of the test_sequence, a new row containing the test configuration and results computed is present in this file. All columns present in the file are defined as:

• Test: Index of current test within the test_sequence. Index of first test is 1.

• duration (s): Duration of current test within the test_sequence.

• test mode: mode of current test within the test_sequence.

• data integrity: Overall data integrity result: set to KO if data integrity error was detected, otherwise set to OK when no error detected. Data integrity status can also be Not Checked or n/a.

• average total write+read BW (MBps): Sum of average write BW (MBps) and average read BW (MBps).

• write configuration: This group contains the following columns. These columns are set to n/a when mode is only_rd.

  • write rate (%): Rate used for writing the memory.

  • write bandwidth (MBps): Requested bandwidth used for writing the memory.

  • write start address offset (MB): Starting address of the memory to be written.

  • write end address offset (MB): Ending address of the memory to be written.

  • write burst size (Bytes): Size of each bursts to be written in the memory.

  • write block size (MB): Size of the block to be written in the memory.

  • maximum number of outstanding writes: Limit of outstanding write transactions.

  • AXI write data size (Bytes): Size of an AXI beat in a burst.

  • number of write transfers per burst: Quantity of AXI beats per burst for write direction.

  • total number of write bursts per block: Quantity of bursts per block for write direction.

  • total number of write transfers per block: Quantity of AXI beats to write the memory based on write block size (MB) and AXI write data size (Bytes).

• write results: This group contains the following columns. These columns are set to n/a when mode is only_rd.

  • average write BW (MBps): Average write BW (MB/s).

  • average number of write bursts per second: Average quantity of bursts written in the memory every second.

  • average write burst latency (ns): Average latency for writing a burst.

• read configuration: Same columns as write configuration group but for read direction. These columns are set to n/a when mode is only_wr.

• read results: Same columns as write results group but for read direction. These columns are set to n/a when mode is only_wr.

Note

When you Override test_sequence, then for multi-channel memory types in the output CSV file with prefix memory_<type>_combined (where <type> is the name of the memory type):

• The configuration parameters are set to OVR. The parameters used for each channel are available in their own output CSV file.

• The combined BW and latency measurements depend on which channels are writing and/or reading from/to the memory.

Detail CSV output files

This file contains all intermediate bandwidths computed every second, based on information retrieved from the Memory xbtest hardware IP (xbtIP). There is one line of result for every second of each test of the test_sequence. All columns present in the file are defined as:

• Global time (s): Global elapsed time since xbtest application software (xbtSW) execution started.

• Test: Index of current test within the test_sequence. Index of first test is 1.

• test mode: mode of current test within the test_sequence.

• Measurement ID: Measurement identifier. ID of first measurement is 0.

• data integrity results: This group contains the following columns. These columns are set to n/a when mode is only_wr. Data integrity status can also be Not Checked.

  • live data integrity: Live data integrity result: set to KO when data integrity error is detected, otherwise set to OK when no error detected.

  • data integrity: Overall data integrity result: set to KO as soon as data integrity error is detected, otherwise set to OK when no error detected.

• write+read BW: This group contains the following columns.

  • live total write+read BW (MBps): Sum of live write BW (MBps) and live read BW (MBps).

  • average total write+read BW (MBps): Sum of average write BW (MBps) and average read BW (MBps).

• write results: This group contains the following groups. These columns are set to n/a when mode is only_rd.

  • write BW: This group contains the following columns.

    • live write BW (MBps): Current live measurement of write BW measurement.

    • average write BW (MBps): Current average value of write BW measurements.

  • number of write bursts: This group contains the following columns.

    • live number of write bursts per second: Current live number of write bursts per second performed by the Memory xbtIP.

    • average number of write bursts per second: Current average number of write bursts per second performed by the Memory xbtIP.

  • write burst latency: This group contains the following columns.

    • live average write burst latency (ns): Current live measurement of average write burst latency.

    • average write burst latency (ns): Current average value of average write burst latency measurements.

• read results: Same columns as write results group but for read direction. These columns are set to n/a when mode is only_wr.

• Timestamp: Timestamp read from Memory xbtIP. Counter incrementing by 1 every second.

Note

The live values are for internal use only of xbtest application software (xbtSW).

Note

When you Override test_sequence, then for multi-channel memory types in the output CSV file with prefix memory_<type>_combined (where <type> is the name of the memory type):

• The configuration parameters are set to OVR. The parameters used for each channel are available in their own output CSV file.

• The combined BW and latency measurements depend on which channels are writing and/or reading from/to the memory.

Multi-channel Summary CSV output files

For each test of the test_sequence, a new row containing the test configuration and results computed is present in this file. All columns present in the file are defined as:

• Test: Index of current test within the test_sequence. Index of first test is 1.

• duration (s): Duration of current test within the test_sequence.

• test mode: mode of current test within the test_sequence.

• write configuration: This group contains the following columns. These columns are set to n/a when mode is only_rd.

  • write rate (%): Rate used for writing the memory.

  • write bandwidth (MBps): Requested bandwidth used for writing the memory.

  • write start address offset (MB): Starting address of the memory to be written.

  • write burst size (Bytes): Size of each bursts to be written in the memory.

  • write block size (MB): Size of the block to be written in the memory.

  • maximum number of outstanding writes: Limit of outstanding write transactions.

• read configuration: Same columns as write configuration group but for read direction. These columns are set to n/a when mode is only_wr.

• Combined results: This group contains the following columns. These columns for write and for read results are set to n/a when mode is only_rd and only_wr respectively.

  • Combined average write BW (MBps): Average write BW (MB/s).

  • Combined average read BW (MBps): Average read BW (MB/s).

  • Combined average write burst latency (ns): Average latency for writing a burst.

  • Combined average read burst latency (ns): Average latency for reading a burst.

  • Combined data integrity: Overall data integrity result: set to KO as soon as data integrity error is detected, otherwise set to OK when no error detected. Data integrity status can also be Not Checked or n/a.

  • Combined average total write+read BW (MBps): Sum of average write BW (MBps) and average read BW (MBps).

Then, one column is present per channel <i>, for each of the following groups:

• Ch <i> average write BW (MBps): Average write BW (MB/s)

• Ch <i> average read BW (MBps): Average read BW (MB/s).

• Ch <i> average write burst latency (ns): Average latency for writing a burst.

• Ch <i> average read burst latency (ns): Average latency for reading a burst.

Note

When you Override test_sequence, then in this output CSV file:

• The configuration parameters are set to OVR. The parameters used for each channel are available in their own output CSV file.

• The combined BW and latency measurements depend on which channels are writing and/or reading from/to the memory.

Power CSV output files

In this file, the test configuration and all power measurements are stored. A new line is written in this file every time the Memory xbtest hardware IP (xbtIP) status are available. At a minimum, the following values are recorded:

• Global time (s): Global elapsed time since xbtest application software (xbtSW) execution started.

• Test: Index of current test within the test_sequence. Index of first test is 1.

• test mode: Mode of current test within the test_sequence.

• Measurement ID: Measurement identifier. ID of first measurement is 0.

• write rate (%): Rate used for writing the memory.

• read rate (%): Rate used for reading the memory.

• Measurement valid: Set to OK if xbtest application software (xbtSW) was able to successfully gets power and temperature measurements via the AVED Management Interface (AMI) APIs, otherwise set to KO.

• Thermal measurements: Group of one or more columns recording measurements and status for each thermal sensor source monitored by xbtest.

• Electrical measurements: Group of one or more columns recording detailed measurements and status for each electrical sensor source monitored by xbtest.

See Sensor source definition for more information on the sensor sources monitored by xbtest.