Multi Source Translation Content

cancel
Showing results for 
Show  only  | Search instead for 
Did you mean: 

Multi Source Translation Content

Discussions

Sort by:
SPI Communication on "s32k31XEVB-Q100 Evaluation Board" Hello, I have a s32k31XEVB-Q100 Evaluation Board and try to implement a SPI communication between the S32 and the FS23. My Pin-Configuration: PTB0 -> lpspi0_pcs0 PTC8 -> lpxpi0_sck PTC9 ->lpxpi0_sin PTB1 -> lpxpi0_sout In Peripherals I added Lpspi /* SPI */ /* Time to transfer all frame data */ #define BCC_MSG_SIZE 4U #define RX_BUFFER_SIZE 4U #define TIMEOUT ((uint32)1000000UL) #define NUMBER_OF_BYTES (10) uint16 numberOfBytes = BCC_MSG_SIZE; static void handleSPI(void) { static int counterSPI = 0; __attribute__(( aligned(32) )) uint8_t rxBuffer[BCC_MSG_SIZE] = {0x00};// = {0,0,0,0,0,0,0,0,0,0,0,0}; /* Buffer for receiving. */ __attribute__(( aligned(32) )) uint8_t txBuffer[BCC_MSG_SIZE] = {0x00}; /* Transmission buffer. */ /* M_DEV_PROG_ID * */ txBuffer[0] = 0x01 << 1; // General status flags ( Register address / Read/Write) txBuffer[1] = 0x00; // Read Data txBuffer[2] = 0x00; // Read Data txBuffer[3] = 0x00; // CRC if (counterSPI++ >= 5000000) { counterSPI = 0; lpspi_status = Lpspi_Ip_SyncTransmit(&MASTER_EXTERNAL_DEVICE, &txBuffer[0], &rxBuffer[0], numberOfBytes, 100000); } } int main(void) { /* Write your code here */ Clock_Ip_StatusType clockStatus; clockStatus = Clock_Ip_Init(&Clock_Ip_aClockConfig[0]); if (CLOCK_IP_SUCCESS != clockStatus) { exit_code = -1; } /* Initial Pins */ Siul2_Port_Ip_Init(NUM_OF_CONFIGURED_PINS_PortContainer_0_BOARD_InitPeripherals, g_pin_mux_InitConfigArr_PortContainer_0_BOARD_InitPeripherals); /* Initial ISR */ IntCtrl_Ip_Init(&IntCtrlConfig_0); IntCtrl_Ip_EnableIrq(FlexCAN0_0_IRQn); /* already enabled in IntCtrl_Ip - Interrupt Controller */ //IntCtrl_Ip_EnableIrq(FlexCAN0_1_IRQn); //IntCtrl_Ip_EnableIrq(FlexCAN0_2_IRQn); //IntCtrl_Ip_EnableIrq(FlexCAN0_2_IRQn); FlexCAN_Ip_Init(INST_FLEXCAN_0, &FlexCAN_State0, &FlexCAN_Config0); FlexCAN_Ip_SetRxIndividualMask(INST_FLEXCAN_0,RX_MB_IDX,0xFFFFFFFF); FlexCAN_Ip_ConfigRxMb(INST_FLEXCAN_0, RX_MB_IDX, &rx_info, Rx_MSG_ID); FlexCAN_Ip_SetStartMode(INST_FLEXCAN_0); //FlexCAN_Ip_Receive(INST_FLEXCAN_0,RX_MB_IDX,&rxData,false); lpspi_status = Lpspi_Ip_Init(&Lpspi_Ip_PhyUnitConfig_SpiPhyUnit_0_Instance_0); /* update Transfer Mode */ //Lpspi_Ip_UpdateTransferMode(MASTER_EXTERNAL_DEVICE.Instance, LPSPI_IP_INTERRUPT); for(;;) { handleSPI(); handleCAN(); handleLED(); if(exit_code != 0) { break; } } return exit_code; } Now I connected my Logic-Analyzer to J39 on the Board. The result of this code is: Now I have some questions: 1) Do I send the values in the right order? 2) Do I have to send the CRC when I just want to read some Data? 3) Which register of the FS23 on this board returns fixed data? Or data which I could know to verify the answer? 4) in the CLK I see sometimes that the clk is longer than normal. Do you have some hints to get the SPI-Communication working for me? Re: SPI Communication on "s32k31XEVB-Q100 Evaluation Board" Hi John Please see comment below: Here I only support FS23 not the S32K so I will reply to you from the view point of FS23. Now I have some questions: 1) Do I send the values in the right order? [gw]Please see answer2 for more detail format about read and write operations. Please also read AN12149 page16--5 MCU communication for getting more info about READ/WRITE/CRC 2) Do I have to send the CRC when I just want to read some Data? [gw]Yes,The SPI communication frame format is as follows: CRC is present by default and cannot be disabled. During read and write operations, the MSB is transmitted first and the LSB last. When SPI is idle, i.e., when the CSB pin is high, MISO is in a high-impedance state. In the AN14129 Page11 page16 introduce the INIT CRC and normal CRC calculation methods. They are different. 3) Which register of the FS23 on this board returns fixed data? Or data which I could know to verify the answer? [gw] M_DEV_PROG_ID  address is  0x01 4) in the CLK I see sometimes that the clk is longer than normal. [gw]You test waveform when CS change to low the CLCK is high but in our timing diagram it's low.
View full article
Request for archived S32K3 FreeRTOS 2.0.0 package for MR-CANHUBK344 Hello, I am using the NXP MR-CANHUBK344 evaluation board and I need the archived FreeRTOS package required by the original NXP example project. Required file: SW32K3_FreeRTOS_10_4_6_UOS_2_0_0_DS_updatesite_D2204.zip My environment: Board: MR-CANHUBK344 MCU: S32K344 S32 Design Studio: 3.4 S32K3 Development Package: 3.4.3 S32K3 RTD AUTOSAR 4.4: 2.0.0 I have already downloaded the required S32DS, S32K3 Development Package and RTD versions, but FreeRTOS 2.0.0 is no longer available in the download portal. Could you please provide access to this archived FreeRTOS 2.0.0 package or enable it for my NXP account? Thank you. Re: Request for archived S32K3 FreeRTOS 2.0.0 package for MR-CANHUBK344 Hi @lxndrkurus  You cannot see the product because it is archived. It means there can be bugs in it, that will not be fixed. We usually recommend that customers use the newer version. However, if you insist on testing the older version, then please wait while I contact the internal team to have them add the software to your account. Best Regards, Robin Re: Request for archived S32K3 FreeRTOS 2.0.0 package for MR-CANHUBK344 @lxndrkurus  Go to: Automotive SW - S32K3 Standard Software -> Automotive SW - S32K3 - FreeRTOS ->  Previous: SW32K3_FreeRTOS_10.4.6_2.0.0_D2204 Re: Request for archived S32K3 FreeRTOS 2.0.0 package for MR-CANHUBK344 Hi @lxndrkurus  It should have been added to your account by now. Please note that the product you are requesting—SW32K3_FreeRTOS_10.4.6_2.0.0_D2204—has been archived, which means that any existing bugs may not be fixed. Please download it by click S32K3 Reference Software -> Automotive SW - S32K3 - FreeRTOS -> Previous -> 2.0.0 SW32K3_FreeRTOS_10.4.6_2.0.0_D2204 Best Regards, Robin
View full article
PF09:KITPF09FRDMPGMとKL25Zなしで、別のI²C MCUボード(例:STM3 こんにちは、NXPチームの皆様、 PF09の場合、KITPF09FRDMPGMとKL25Zを使わず、STM32ボードのようなI²Cインターフェースを持つ別のMCUボードを使ってOTPエミュレーションやプログラミングを行うことは可能でしょうか? これに関する規定や文書的なサポートはありますか?可能であれば、それについて説明している文書を教えてください。 ありがとう、 マヘシュ i.MX用PMIC Re: PF09: Can OTP be programmed without KITPF09FRDMPGM and KL25Z, using another I²C MCU board (e.g. こんにちは、マヘシュタンカラ そのようなOTPプログラムはサポートしていませんが申し訳ありません。
View full article
S32K312 ICU SAIC DMA Hi NXP Team: 关于K3的ICU模块,有一个问题如下: 配置一个ICU输入通道测量PWM,SAIC模式,采用DMA搬运数据。首先是SAIC DMA模式RTD代码里会强制改为双边沿触发,然后在双边沿触发的情况下,如果输入的PWM信号线中途断开后再重新接上,当接上信号那一瞬间如果PWM波形处于高电平时刻,则下一个触发的边沿为下降沿,导致后续测量的占空比均是反的,例如PWM波形是20%占空比,ICU的测量结果就是80% 一个完整的PWM波形为: 上升沿——>高电平——>下降沿——>低电平 双边沿触发模式下有可能从下降沿开始采样,将前一个波形的低电平区和后一个波形的高电平区重新组成为一个波形。 请问是否有遇到过类似问题以及有什么好的解决方案吗? 谢谢! Re: S32K312 ICU SAIC DMA 我将按照上述步骤进行调试。非常感谢您的回复! Re: S32K312 ICU SAIC DMA 您好, 您所描述的行为可能发生在 SAIC 与双边沿触发一起使用,并且软件仅根据捕获的时间戳顺序确定 PWM 高/低时间时。输入信号中断后,下一个捕获的边沿可能是下降沿,这可能导致测量极性反转。 但是,eMIOS SAIC 提供了一个边缘捕获功能,特别适用于这种情况。启用 IcuSaicEdgeCapturing 后,AS2 位 31 包含与捕获的时间戳关联的输入信号电平。即使使用双边沿触发,软件也能区分上升沿和下降沿,从而在信号重新连接后恢复正确的 PWM 边沿序列。 请考虑启用此功能,并在处理 DMA 采样时使用捕获的信号电平信息。应检查具体的 DMA 数据处理,以确保 AS2 信号电平位得到传输和保留。 或者,如果符合应用/通道配置,可以考虑使用专用的 eMIOS IPW/IPWM 测量模式,因为这些模式是专门为脉冲宽度/PWM 测量而设计的。 BR,彼得
View full article
RFIDDiscover hello i need to use RFIDdiscover Lite with pegoda CLRD730, can you provide me with link to download it. also i found that the full version of the RFIDdiscover needs a NDA agreement, so can any one tell me whats the steps i need to do (the process of the NDA) so i can complete the NDA and have access to the tool. thanks RF
View full article
FS26 standby Hi, NXP expert, I have encountered a problem related to FS26: Testing revealed that the FS26 cannot enter standby mode. Single-step execution continues, but the FS26 fails to enter standby mode. A reset occurs whenever the Sbc_fs26_SetOperatingMode() interface is run at full speed. The hardware wake signal has been removed, and no other signals are connected to the reset pin. Small Engine Driver Re: FS26 standby Hi Chenxu From which state did you switch to standby state? Could you please provide the command you issued? Re: FS26 standby Hi, thank you for your help, the problem is solved. Testing revealed that when the wake-up time is less than 5ms, reading the M_WIO_FLG register returns 0, indicating that the wake-up source cannot be read. Could you please help answer this question? Thank you. Re: FS26 standby HI CHENXU Please submit a new ticket for the new topic! https://support.nxp.com/s/?language=en_US
View full article
Scarthgap 和 Wrynose 的 I.MX8 NPU 性能差异 您好,NXP, 我已经对 NXP 提供的 `mobilenet_v1_1.0_224_quant.tflite` 进行了基准测试。使用 VX NPU 代理的模型。下面的结果是在 Scarthgap 6.6.23和 Wrynose 6.18.37 上重复基准测试运行的稳态“推理(平均)”计时结果。 Wrynose算法的平均推理时间比Scarthgap算法的结果高出约27%。 这种差异是预期之内的吗?我应该检查哪些 NPU、委托或基准测试设置? 基准测试命令为: ```bash ./benchmark_model--graph=mobilenet_v1_1.0_224_quant.tflite --external_delegate_path=/usr/lib/libvx_delegate.so ``` 此致敬礼, 加布里埃尔·戈斯登 Re: I.MX8 NPU performance difference between Scarthgap and Wrynose 你好, 我在我们的 i.MX 8M Plus EVK 上也重现了类似的性能差异。使用相同的 MobileNet 模型和测试配置,我在 Scarthgap 上测得的时间为约 2.96 毫秒,而在 Wrynose 上测得的时间为 3.55 毫秒,推理时间大约增加了 20% 。 我的测试证实,两个版本中都成功委托了整个模型,并且相关的 NPU/ML 时钟频率配置为相同的速率。我还通过性能分析确认,性能差异仍然存在于 VX Delegate 执行路径中。 该现象可复现,现有结果表明,差异与软件执行堆栈有关,而非CPU回退或较低的NPU标称时钟频率。
View full article
Interface hiding issue in guiguider 2.0.1 In GuiGuider 2.0.1, sometimes adding the "hidden" flag to the properties on the right side of an interface can cause other interfaces to also become invisible, and this cannot be fixed without restarting the software. 回复: 界面隐藏问题guiguider2.0.1 Hi @ZYQ  Thanks for your information. I can reproduce this issue. And i have reported it to our GUIGUIDER team. BR Harry
View full article
Multicore Project setup of an additional MPU region I've been working on making a project that is essentially a combination of 2 SDK examples. I've been using the RT1170-EVKB SDK, version 26.09.00. The idea of the project is to take a multicore example and an LVGL example to have the M7 core run the graphics then communicate with the M4 core running the program. To do this I have taken the SDK examples: freertos_message_buffers (both primary and secondary) as my base project. I then slowly add in the graphics by copying and pasting the code and driver files from the lvgl_guider_cm7 project. I keep lvgl_guider_cm7 mcux_config.h file the same so that it does use PXP and VGLite. In order to do this, framebuffers need to placed in non-cache. The shared memory used in the SDK example also needs to be placed in non-cache. When this happens, I get artifacts on the display (little lines through objects that shouldn't be there). I can fix this by removing the __USE_SHMEM preprocessor, obviously this causes the multicore communication to fail. I believe it is because of the BOARD_EarlyConfigMPU function. Inside the function there is a section that goes like: #elif defined(__MCUXPRESSO) #if defined(__USE_SHMEM) extern uint32_t __base_rpmsg_sh_mem; extern uint32_t __top_rpmsg_sh_mem; uint32_t nonCacheStart = (uint32_t)(&__base_rpmsg_sh_mem); uint32_t size = (uint32_t)(&__top_rpmsg_sh_mem) - nonCacheStart; #else extern uint32_t __base_NCACHE_REGION; extern uint32_t __top_NCACHE_REGION; uint32_t nonCacheStart = (uint32_t)(&__base_NCACHE_REGION); uint32_t size = (uint32_t)(&__top_NCACHE_REGION) - nonCacheStart; #endif In this section it defines non-cache for either the shared memory or regular non-cache based on the __USE_SHMEM preprocessor. Further down in the function it declares the MPU region like so: volatile uint32_t i = 0; ... while ((size >> i) > 0x1U) { i++; } if (i != 0) { /* The MPU region size should be 2^N, 5<=N<=32, region base should be multiples of size. */ BOARD_EARLY_ASSERT(!(nonCacheStart % size)); BOARD_EARLY_ASSERT(size == (uint32_t)(1 << i)); BOARD_EARLY_ASSERT(i >= 5); /* Region 10 setting: Memory with Normal type, not shareable, non-cacheable */ MPU->RBAR = ARM_MPU_RBAR(BOARD_MPU_REGION_NONCACHE, nonCacheStart); MPU->RASR = ARM_MPU_RASR(0, ARM_MPU_AP_FULL, 1, 0, 0, 0, 0, i - 1); } Now I want to be able to define both regions in the MPU to have both LVGL working without artifacts and the rpmsg message buffers be able to communicate. My memory regions can be found below: What I tried to do was modify the code above to perform both MPU region definitions, like so: #elif defined(__MCUXPRESSO) extern uint32_t __base_NCACHE_REGION; extern uint32_t __top_NCACHE_REGION; uint32_t nonCacheStart = (uint32_t)(&__base_NCACHE_REGION); uint32_t nonCacheSize = (uint32_t)(&__top_NCACHE_REGION) - nonCacheStart; #if defined(__USE_SHMEM) extern uint32_t __base_rpmsg_sh_mem; extern uint32_t __top_rpmsg_sh_mem; uint32_t rpmsgShmemStart = (uint32_t)(&__base_rpmsg_sh_mem); uint32_t rpmsgShmemSize = (uint32_t)(&__top_rpmsg_sh_mem) - rpmsgShmemStart; #endif This way the regular non-cache region is always defined and the shared memory region is only defined if used. I modified the MPU region definition like so: volatile uint32_t i = 0; ... while ((nonCacheSize >> i) > 0x1U) { i++; } if (i != 0) { /* The MPU region size should be 2^N, 5<=N<=32, region base should be multiples of size. */ BOARD_EARLY_ASSERT(!(nonCacheStart % nonCacheSize)); BOARD_EARLY_ASSERT(nonCacheSize == (uint32_t)(1 << i)); BOARD_EARLY_ASSERT(i >= 5); /* Region 10 setting: Memory with Normal type, not shareable, non-cacheable */ MPU->RBAR = ARM_MPU_RBAR(BOARD_MPU_REGION_NONCACHE, nonCacheStart); MPU->RASR = ARM_MPU_RASR(0, ARM_MPU_AP_FULL, 1, 0, 0, 0, 0, i - 1); } #if (__USE_SHMEM) i = 0; while ((rpmsgShmemSize >> i) > 0x1U) { i++; } if (i != 0) { /* The MPU region size should be 2^N, 5<=N<=32, region base should be multiples of size. */ BOARD_EARLY_ASSERT(!(rpmsgShmemStart % rpmsgShmemSize)); BOARD_EARLY_ASSERT(rpmsgShmemSize == (uint32_t)(1 << i)); BOARD_EARLY_ASSERT(i >= 5); /* Region 16 setting: Memory with Normal type, not shareable, non-cacheable */ MPU->RBAR = ARM_MPU_RBAR(BOARD_MPU_REGION_SHMEM, rpmsgShmemStart); MPU->RASR = ARM_MPU_RASR(0, ARM_MPU_AP_FULL, 1, 0, 0, 0, 0, i - 1); } #endif In doing this I created a macro for the additional MPU region inside board.h like so: #define BOARD_MPU_REGION_SHMEM (16) /* rpmsg shared memory */ Performing the modifications allowed LVGL to draw without artifacts but the multicore communication still fails. I put a breakpoint in the M4 thread running the multicore communication. Sometimes it'll trigger at the start, notifying that it received a message, but the message data is incorrect so it will not execute the code associated with the message's data. After a couple triggers of the breakpoint it will never trigger again. Is there anything I'm missing to be able to get this project running, the MPU region defines, etc...? Re: Multicore Project setup of an additional MPU region The changes were only applied to the primary core, as the primary core runs BOARD_EarlyConfigMPU and the secondary core runs BOARD_ConfigMPU, in the secondary core's function it already has the behaviour that I changed the primary core to have. Below are snippets from the what the secondary core runs on hardware init. #elif __CORTEX_M == 4 void BOARD_ConfigMPU(void) { #if defined(__CC_ARM) || defined(__ARMCC_VERSION) ... #elif defined(__MCUXPRESSO) extern uint32_t __base_NCACHE_REGION; extern uint32_t __top_NCACHE_REGION; uint32_t nonCacheStart = (uint32_t)(&__base_NCACHE_REGION); uint32_t nonCacheSize = (uint32_t)(&__top_NCACHE_REGION) - nonCacheStart; #elif defined(__ICCARM__) || defined(__GNUC__) ... #endif #if defined(__USE_SHMEM) #if defined(__CC_ARM) || defined(__ARMCC_VERSION) ... #elif defined(__MCUXPRESSO) extern uint32_t __base_rpmsg_sh_mem; extern uint32_t __top_rpmsg_sh_mem; uint32_t rpmsgShmemStart = (uint32_t)(&__base_rpmsg_sh_mem); uint32_t rpmsgShmemSize = (uint32_t)(&__top_rpmsg_sh_mem) - rpmsgShmemStart; #elif defined(__ICCARM__) || defined(__GNUC__) ... #endif #endif And then later in the function: uint32_t i = 0; /* Only config non-cacheable region on system bus */ assert(nonCacheStart >= 0x20000000); ... while ((nonCacheSize >> i) > 0x1U) { i++; } if (i != 0) { /* The MPU region size should be 2^N, 5<=N<=32, region base should be multiples of size. */ assert(!(nonCacheStart % nonCacheSize)); assert(nonCacheSize == (uint32_t)(1 << i)); assert(i >= 5); /* Region 0 setting: Memory with device type, not shareable, non-cacheable */ MPU->RBAR = ARM_MPU_RBAR(BOARD_MPU_REGION_NONCACHE, nonCacheStart); MPU->RASR = ARM_MPU_RASR(0, ARM_MPU_AP_FULL, 2, 0, 0, 0, 0, i - 1); } #if defined(__USE_SHMEM) i = 0; while ((rpmsgShmemSize >> i) > 0x1U) { i++; } if (i != 0) { /* The MPU region size should be 2^N, 5<=N<=32, region base should be multiples of size. */ assert(!(rpmsgShmemStart % rpmsgShmemSize)); assert(rpmsgShmemSize == (uint32_t)(1 << i)); assert(i >= 5); /* Region 1 setting: Memory with device type, not shareable, non-cacheable */ MPU->RBAR = ARM_MPU_RBAR(BOARD_MPU_REGION_RPMSG_SHMEM, rpmsgShmemStart); MPU->RASR = ARM_MPU_RASR(0, ARM_MPU_AP_FULL, 2, 0, 0, 0, 0, i - 1); } #endif Both projects have __USE_SHMEM defined in the preprocessor. Below is the memory config for the secondary core: Am I missing something else, or perhaps are my modifications wrong? Do I need to add something extra to the "Manage Linker Script" section of the primary project's settings? When I get the chance I'll upload a .zip of the projects that run on the EVK-B board so that you may see/replicate the issue. Re: Multicore Project setup of an additional MPU region Hi @CMoran, Your understanding makes sense, but it wasn't clear if that changed you mention were applied to both CM7 and CM4 cores. Each core has a definition of board.h, where they use their respective configuration of the MPU to setup during the initial hardware initialization. If the changes you mention were only made for the primary core, it's likely that the cores don't treat the memory regions equally, resulting in these unexpected behaviors. Please verify that the changes are applied to both cores to ensure they coincide on MPU setup, as well as ensuring that __USE_SHMEM was setup for both project's preprocessor definitions. Re: Multicore Project setup of an additional MPU region Here's the .zip with the m7 and m4 projects within. I've left it to be before my modifications, as the modifications didn't work anyway. In this example if you open a terminal and push the button in the center, both the m7 and m4 cores will print a message. The settings icon moving around the screen has a bunch of artifacts in it as described. Removing the __USE_SHMEM preprocessor from the m7 project will fix the artifacts in the settings icon, but the multicore communication will fail. Adding the modifications I've applied above will also fix the settings icon, but the multicore communication will fail. Screen used in this demo is the RK055HDMIPI4MA0. Re: Multicore Project setup of an additional MPU region Hi @CMoran, Got it. Going back on your first reply though: You mentioned that you created a new definition for a memory region: #define BOARD_MPU_REGION_SHMEM (16) /* rpmsg shared memory */ But the definition of ARM_MPU_RBAR() uses MPU_RBAR_REGION_Msk, which is defined as 15: #define MPU_RBAR_REGION_Msk (0xFUL /*<< MPU_RBAR_REGION_Pos*/)  The CM7 defines the number of regions as 16, numbered 0-15. Therefore, you are actually setting region 0 by doing so. Re: Multicore Project setup of an additional MPU region Unfortunately before creating the new memory region definition: #define BOARD_MPU_REGION_SHMEM (16) /* rpmsg shared memory */ I was just reusing the predefined memory region for non-cache: #define BOARD_MPU_REGION_NONCACHE (10) /* Noncacheable space in application */ And I had the same result. I've changed my new memory region definition to be 10 now to match the non-cache. With no luck on the cores being able to communicate. I've checked the start location and the size of the rpmsg shared memory on both cores with breakpoints, both have the correct start (0x20340000) and size (0x10000). I've changed the 4th parameter of the ARM_MPU_RASR() function on both cores to 1 to indicate that the region is "shareable". No luck there still. When you or a member of the team gets a chance, I posted my project's code in another reply. I'm running out of ideas of what to try. Re: Multicore Project setup of an additional MPU region Hi again @CMoran, After building your project and analyzing the memory regions on the resulting .ld files, I see that the regions where the LVGL framebuffers and the multicore message buffers are located at separate memory regions in a way that can't be merged as they are non-contiguous physical memories: NCACHE_REGION at 0x81000000 (16 MB, in SDRAM) for the LVGL framebuffers rpmsg_sh_mem at 0x20340000 (64 KB, in OCRAM) for the multicore message buffers That said, MPU region 7 currently covers 0x20300000–0x2037FFFF (512 KB) as cacheable Normal memory, a range that includes the rpmsg_sh_mem region, which could mean that the shared memory ends up being cacheable, resulting in the inter-core communication errors that you are seeing. Reducing the size of Region 7 to ensure rpmsg_sh_mem falls through to region 3, which is already non-cacheable might be a straightforward fix to ensure the proper functionality of the message buffers, without MPU driver level changes on the existing drivers. Re: Multicore Project setup of an additional MPU region Hi @EdwinHz  I've taken a second to test your suggestion. I changed the MPU Region 7 (OCRAM2) to be 256 KB instead of 512 KB like so: /* Region 7 setting: Memory with Normal type, not shareable, outer/inner write back */ MPU->RBAR = ARM_MPU_RBAR(BOARD_MPU_REGION_OCRAM_B, 0x20300000U); MPU->RASR = ARM_MPU_RASR(0, ARM_MPU_AP_FULL, 0, 0, 1, 1, 0, ARM_MPU_REGION_SIZE_256KB); After I removed the __USE_SHMEM preprocessor as the rpmsg_sh_mem will fall through to MPU Region 3, which is already defined as non-cacheable. The demo I created is now working as intended.
View full article
S32K312 ICU SAIC DMA Hi NXP Team: Regarding the ICU module of K3, I have the following question: Configure an ICU input channel to measure PWM in SAIC mode, using DMA for data transfer. First, the SAIC DMA mode RTD code will be forced to use double-edge triggering. Then, in double-edge triggering mode, if the input PWM signal line is interrupted and then reconnected, if the PWM waveform is high at the moment of reconnection, the next trigger edge will be a falling edge, causing the duty cycle of subsequent measurements to be inverted. For example, if the PWM waveform is 20 %占空比,ICU的测量结果就是80% A complete PWM waveform is as follows: Rising edge -> High level -> Falling edge -> Low level In dual-edge triggered mode, it is possible to start sampling from the falling edge, and reassemble the low-level area of the previous waveform and the high-level area of the next waveform into a single waveform. Have you encountered a similar problem before, and do you have any good solutions? Thanks! Re: S32K312 ICU SAIC DMA Hi, The behavior you describe can occur when SAIC is used with both-edge triggering and the software determines the PWM high/low time only from the order of captured timestamps. After the input signal is interrupted, the next captured edge can be a falling edge, which can cause the measurement polarity to become reversed. However, eMIOS SAIC provides an edge-capturing feature specifically useful for this case. With IcuSaicEdgeCapturing enabled, AS2 bit 31 contains the input signal level associated with the captured timestamp. This allows software to distinguish rising and falling edges even when both-edge triggering is used and therefore to restore the correct PWM edge sequence after the signal is reconnected. Please consider enabling this feature and using the captured signal-level information when processing the DMA samples. The exact DMA data handling should be checked to ensure that the AS2 signal-level bit is transferred and preserved. Alternatively, the dedicated eMIOS IPW/IPWM measurement mode can be considered if it fits the application/channel configuration, since these modes are intended specifically for pulse-width/PWM measurement. BR, Petr Re: S32K312 ICU SAIC DMA I will try debugging according to the above steps. Thank you  very much for your reply!
View full article
HSE选项 你好, 出于某些出口目的,我们希望移除 HSE(加密部分)模块。我在你们的列表中没有找到任何不带 HSE 模块的参考资料,你们是否确认没有不带 HSE(加密部件)模块的 S32G2 或 S32G3 参考资料? B.R. Re: HSE option 你好, @COLLIN_O 感谢分享。 从最新推出的 S32G2/3 产品来看,HSE 包含在所有列出的型号中。 BR 陈银
View full article
A53 core frequency in S32G3 BSP Hi Community and NXP export! In BSP 48 manual, there is information about A53 core clock like this root@s32g399aevb3:/sys/devices/system/cpu/cpufreq/policy0/# cat scaling_available_frequencies 48148 54166 61904 72222 86666 104000 136842 200000 371428 1300000 My Question is, why the clock lower down from 1.3G to 371M directly, while no 866/650 available?Is there some special consideration? Best Regards! Guangyu Re: A53 core frequency in S32G3 BSP Hello, @Guangyu_D  Thanks for the post. 1. From the available document, I do not see there is special consideration regarding to it. 2. The current results is caused by the original frequency-sampling strategy introduced in the initial implementation, It selects evenly spaced divider values, rather than evenly spaced frequencies, between the limits derived from the allowed frequency range. This naturally produces sparse frequency points at the high end and denser points at the low end. BR Chenyin
View full article
RFIDDiscover こんにちは Pegoda CLRD730でRFIDdiscover Liteを使う必要があるのですが、ダウンロード用のリンクを教えてもらえますか? また、RFIDdiscoverのフルバージョンにはNDA契約が必要だと分かりました。NDAを完成させてツールにアクセスできるように、どのような手続き(NDA)をすればよいのか、どなたか教えてもらえますか? よろしくお願いします。 RF
View full article
S32G3 BSPにおけるA53コア周波数 コミュニティの皆様、そしてNXPのエクスポート機能について、こんにちは! BSP 48のマニュアルには、A53コアクロックに関する情報が次のように記載されています。 root@s32g399aevb3:/sys/devices/system/cpu/cpufreq/policy0/# cat scaling_available_frequencies 48148 54166 61904 72222 86666 104000 136842 200000 371428 1300000 私の質問は、なぜ866/650が利用できないのに、クロック周波数が1.3Gから371Mに直接下げられているのかということです。何か特別な理由があるのでしょうか? よろしくお願いいたします! 光宇 Re: A53 core frequency in S32G3 BSP こんにちは、 @Guangyu_D 投稿ありがとうございます。 1. 入手可能な文書からは、それに関する特別な考慮事項は見当たりません。 2. 現在の結果は、初期実装で導入された元の周波数サンプリング戦略に起因しており、許容周波数範囲から導出された制限内で、均等間隔の周波数ではなく、均等間隔の分割値を選択しています。これは当然ながら、高周波数領域では周波数点がまばらになり、低周波数領域では周波数点が密になるという結果をもたらす。 BR チェイン
View full article
多核项目设置额外的 MPU 区域 我一直在开发一个项目,该项目本质上是两个 SDK 示例的结合体。我一直在使用 RT1170-EVKB SDK,版本为 26.09.00。该项目的想法是采用多核示例和 LVGL 示例,让 M7 核心运行图形,然后与运行程序的 M4 核心进行通信。为此,我采用了 SDK 示例: 我的基础项目是 freertos_message_buffers(包括主缓冲区和辅助缓冲区)。然后我通过从 lvgl_guider_cm7 项目中复制和粘贴代码和驱动程序文件,慢慢地添加图形。 我保留了 lvgl_guider_cm7 mcux_config.h文件内容相同,以便它能够使用 PXP 和 VGLite。为了实现这一点,需要将帧缓冲区放置在非缓存位置。SDK 示例中使用的共享内存也需要放置在非缓存中。发生这种情况时,屏幕上会出现瑕疵(不应该出现在那里的物体上出现细线)。 我可以通过移除 __USE_SHMEM 预处理器来解决这个问题,但显然这会导致多核通信失败。我认为这是由于 BOARD_EarlyConfigMPU 函数引起的。函数内部有一个部分的内容如下: #elif defined(__MCUXPRESSO) #if defined(__USE_SHMEM) extern uint32_t __base_rpmsg_sh_mem; extern uint32_t __top_rpmsg_sh_mem; uint32_t nonCacheStart = (uint32_t)(&__base_rpmsg_sh_mem); uint32_t size = (uint32_t)(&__top_rpmsg_sh_mem) - nonCacheStart; #else extern uint32_t __base_NCACHE_REGION; extern uint32_t __top_NCACHE_REGION; uint32_t nonCacheStart = (uint32_t)(&__base_NCACHE_REGION); uint32_t size = (uint32_t)(&__top_NCACHE_REGION) - nonCacheStart; #endif 本节根据 __USE_SHMEM 预处理器定义了共享内存或常规非缓存的非缓存。函数后面部分对 MPU 区域进行了如下声明: volatile uint32_t i = 0; ... while ((size >> i) > 0x1U) { i++; } if (i != 0) { /* The MPU region size should be 2^N, 5<=N<=32, region base should be multiples of size. */ BOARD_EARLY_ASSERT(!(nonCacheStart % size)); BOARD_EARLY_ASSERT(size == (uint32_t)(1 << i)); BOARD_EARLY_ASSERT(i >= 5); /* Region 10 setting: Memory with Normal type, not shareable, non-cacheable */ MPU->RBAR = ARM_MPU_RBAR(BOARD_MPU_REGION_NONCACHE, nonCacheStart); MPU->RASR = ARM_MPU_RASR(0, ARM_MPU_AP_FULL, 1, 0, 0, 0, 0, i - 1); } 现在我希望能够在 MPU 中定义这两个区域,以便 LVGL 能够正常工作而不会出现瑕疵,并且 rpmsg 消息缓冲区能够进行通信。我的记忆区域如下: 我尝试修改上面的代码,使其能够同时定义两个 MPU 区域,如下所示: #elif defined(__MCUXPRESSO) extern uint32_t __base_NCACHE_REGION; extern uint32_t __top_NCACHE_REGION; uint32_t nonCacheStart = (uint32_t)(&__base_NCACHE_REGION); uint32_t nonCacheSize = (uint32_t)(&__top_NCACHE_REGION) - nonCacheStart; #if defined(__USE_SHMEM) extern uint32_t __base_rpmsg_sh_mem; extern uint32_t __top_rpmsg_sh_mem; uint32_t rpmsgShmemStart = (uint32_t)(&__base_rpmsg_sh_mem); uint32_t rpmsgShmemSize = (uint32_t)(&__top_rpmsg_sh_mem) - rpmsgShmemStart; #endif 这样一来,常规的非缓存区域始终会被定义,而共享内存区域仅在需要时才会被定义。我按如下方式修改了 MPU 区域定义: volatile uint32_t i = 0; ... while ((nonCacheSize >> i) > 0x1U) { i++; } if (i != 0) { /* The MPU region size should be 2^N, 5<=N<=32, region base should be multiples of size. */ BOARD_EARLY_ASSERT(!(nonCacheStart % nonCacheSize)); BOARD_EARLY_ASSERT(nonCacheSize == (uint32_t)(1 << i)); BOARD_EARLY_ASSERT(i >= 5); /* Region 10 setting: Memory with Normal type, not shareable, non-cacheable */ MPU->RBAR = ARM_MPU_RBAR(BOARD_MPU_REGION_NONCACHE, nonCacheStart); MPU->RASR = ARM_MPU_RASR(0, ARM_MPU_AP_FULL, 1, 0, 0, 0, 0, i - 1); } #if (__USE_SHMEM) i = 0; while ((rpmsgShmemSize >> i) > 0x1U) { i++; } if (i != 0) { /* The MPU region size should be 2^N, 5<=N<=32, region base should be multiples of size. */ BOARD_EARLY_ASSERT(!(rpmsgShmemStart % rpmsgShmemSize)); BOARD_EARLY_ASSERT(rpmsgShmemSize == (uint32_t)(1 << i)); BOARD_EARLY_ASSERT(i >= 5); /* Region 16 setting: Memory with Normal type, not shareable, non-cacheable */ MPU->RBAR = ARM_MPU_RBAR(BOARD_MPU_REGION_SHMEM, rpmsgShmemStart); MPU->RASR = ARM_MPU_RASR(0, ARM_MPU_AP_FULL, 1, 0, 0, 0, 0, i - 1); } #endif 为此,我在 board.h 中为额外的 MPU 区域创建了一个宏。就像这样: #define BOARD_MPU_REGION_SHMEM (16) /* rpmsg shared memory */ 进行这些修改后,LVGL 可以无瑕疵地绘制图形,但多核通信仍然失败。我在运行多核通信的 M4 线程中设置了一个断点。有时它会在启动时触发信号,通知它已收到消息,但消息数据不正确,因此不会执行与消息数据关联的代码。断点触发信号几次后,将不再触发。 我是否遗漏了什么才能让这个项目运行起来,例如 MPU 区域定义等等……? Re: Multicore Project setup of an additional MPU region 这些更改仅应用于主核心,因为主核心运行 BOARD_EarlyConfigMPU,而辅助核心运行 BOARD_ConfigMPU,辅助核心的函数已经具有我更改主核心的行为。以下是辅助核心在硬件初始化时运行的部分代码片段。 #elif __CORTEX_M == 4 void BOARD_ConfigMPU(void) { #if defined(__CC_ARM) || defined(__ARMCC_VERSION) ... #elif defined(__MCUXPRESSO) extern uint32_t __base_NCACHE_REGION; extern uint32_t __top_NCACHE_REGION; uint32_t nonCacheStart = (uint32_t)(&__base_NCACHE_REGION); uint32_t nonCacheSize = (uint32_t)(&__top_NCACHE_REGION) - nonCacheStart; #elif defined(__ICCARM__) || defined(__GNUC__) ... #endif #if defined(__USE_SHMEM) #if defined(__CC_ARM) || defined(__ARMCC_VERSION) ... #elif defined(__MCUXPRESSO) extern uint32_t __base_rpmsg_sh_mem; extern uint32_t __top_rpmsg_sh_mem; uint32_t rpmsgShmemStart = (uint32_t)(&__base_rpmsg_sh_mem); uint32_t rpmsgShmemSize = (uint32_t)(&__top_rpmsg_sh_mem) - rpmsgShmemStart; #elif defined(__ICCARM__) || defined(__GNUC__) ... #endif #endif 然后在函数的后面部分: uint32_t i = 0; /* Only config non-cacheable region on system bus */ assert(nonCacheStart >= 0x20000000); ... while ((nonCacheSize >> i) > 0x1U) { i++; } if (i != 0) { /* The MPU region size should be 2^N, 5<=N<=32, region base should be multiples of size. */ assert(!(nonCacheStart % nonCacheSize)); assert(nonCacheSize == (uint32_t)(1 << i)); assert(i >= 5); /* Region 0 setting: Memory with device type, not shareable, non-cacheable */ MPU->RBAR = ARM_MPU_RBAR(BOARD_MPU_REGION_NONCACHE, nonCacheStart); MPU->RASR = ARM_MPU_RASR(0, ARM_MPU_AP_FULL, 2, 0, 0, 0, 0, i - 1); } #if defined(__USE_SHMEM) i = 0; while ((rpmsgShmemSize >> i) > 0x1U) { i++; } if (i != 0) { /* The MPU region size should be 2^N, 5<=N<=32, region base should be multiples of size. */ assert(!(rpmsgShmemStart % rpmsgShmemSize)); assert(rpmsgShmemSize == (uint32_t)(1 << i)); assert(i >= 5); /* Region 1 setting: Memory with device type, not shareable, non-cacheable */ MPU->RBAR = ARM_MPU_RBAR(BOARD_MPU_REGION_RPMSG_SHMEM, rpmsgShmemStart); MPU->RASR = ARM_MPU_RASR(0, ARM_MPU_AP_FULL, 2, 0, 0, 0, 0, i - 1); } #endif 这两个项目的预处理器中都定义了 __USE_SHMEM。以下是辅助核心的内存配置: 我是否遗漏了什么,或者我的修改是否错误?我是否需要在主项目设置的“管理链接器脚本”部分添加一些额外的内容?有机会的话,我会上传一个包含 EVK-B 板上运行的项目的 .zip 文件,以便您可以查看/重现该问题。 Re: Multicore Project setup of an additional MPU region 嗨@CMoran , 你的理解很有道理,但是你提到的变化是否同时适用于 CM7 和 CM4 内核,这一点还不清楚。每个核心都有 board.h 的定义,它们使用各自的 MPU 配置在初始硬件初始化期间进行设置。如果您提到的更改仅针对主核心进行,则核心对内存区域的处理方式可能不一致,从而导致这些意外行为。 请验证更改是否已应用于两个核心,以确保 MPU 设置一致,并确保已为两个项目的预处理器定义设置 __USE_SHMEM。 Re: Multicore Project setup of an additional MPU region 这是 .zip 文件。其中包含 m7 和 m4 项目。我保留了修改前的样子,因为修改后也没起作用。在这个例子中,如果你打开一个终端并按下中间的按钮,m7 和 m4 内核都会打印一条消息。如前所述,屏幕上移动的设置图标包含许多瑕疵。 从 m7 项目中移除 __USE_SHMEM 预处理器将修复设置图标中的瑕疵,但多核通信将会失败。 添加我上面应用的修改后,设置图标也会修复,但多核通信将会失败。 本演示中使用的屏幕是 RK055HDMIPI4MA0。 Re: Multicore Project setup of an additional MPU region 遗憾的是,在创建新的内存区域定义之前: #define BOARD_MPU_REGION_SHMEM (16) /* rpmsg shared memory */ 我只是重复使用了预定义的非缓存内存区域: #define BOARD_MPU_REGION_NONCACHE (10) /* Noncacheable space in application */ 我的结果也一样。我已经将新的内存区域定义更改为10 ,以匹配非缓存区域。核心之间始终无法通信。 我用断点检查了两个核心上 rpmsg 共享内存的起始位置和大小,它们的起始位置 (0x20340000) 和大小 (0x10000) 都正确。我已将两个核心上的 ARM_MPU_RASR() 函数的第 4 个参数更改为1 ,以表明该区域是“可共享的”。还是没找到办法。当您或团队成员有空时,我会在另一个回复中发布我的项目代码。我已经想不出还能尝试什么了。 Re: Multicore Project setup of an additional MPU region 嗨@CMoran , 知道了。不过,回到你最初的回复: 您提到您创建了一个新的内存区域定义: #define BOARD_MPU_REGION_SHMEM (16) /* rpmsg shared memory */ 但是 ARM_MPU_RBAR() 的定义使用了 MPU_RBAR_REGION_Msk,其定义为 15: #define MPU_RBAR_REGION_Msk (0xFUL /*<< MPU_RBAR_REGION_Pos*/) CM7 将区域数量定义为 16 个,编号为 0-15。 因此,这样做实际上是将区域 0 设置到该区域。 Re: Multicore Project setup of an additional MPU region @CMoran ,你好! 项目构建完成后,请分析生成的 .ld 文件中的内存区域。从文件中可以看出,LVGL 帧缓冲区和多核消息缓冲区位于不同的内存区域,由于它们是不连续的物理内存,因此无法合并: LVGL 帧缓冲区的 NCACHE_REGION 位于 0x81000000(16 MB,位于 同步动态随机存取存储器(SDRAM) 中) rpmsg_sh_mem 位于 0x20340000(64 KB,OCRAM),用于多核消息缓冲区 也就是说,MPU 区域 7 目前覆盖 0x20300000–0x2037FFFF (512 KB) 作为可缓存的普通内存,该范围包括 rpmsg_sh_mem 区域,这意味着共享内存最终可能是可缓存的,从而导致您看到的内核间通信错误。减小 Region 7 的大小,以确保 rpmsg_sh_mem 降级到 Region 3(该区域本身不可缓存),这可能是一个简单的修复方法,可以确保消息缓冲区的正常功能,而无需对现有驱动程序进行 MPU 驱动程序级别的更改。 Re: Multicore Project setup of an additional MPU region 嗨@EdwinHz 我花了一点时间测试了一下你的建议。我将 MPU 区域 7 (OCRAM2) 的值从 512 KB 改为 256 KB,如下所示: /* Region 7 setting: Memory with Normal type, not shareable, outer/inner write back */ MPU->RBAR = ARM_MPU_RBAR(BOARD_MPU_REGION_OCRAM_B, 0x20300000U); MPU->RASR = ARM_MPU_RASR(0, ARM_MPU_AP_FULL, 0, 0, 1, 1, 0, ARM_MPU_REGION_SIZE_256KB); 移除 __USE_SHMEM 预处理器后,rpmsg_sh_mem 将降级到 MPU 区域 3,而该区域已被定义为不可缓存。我创建的演示程序现在已经按预期运行了。
View full article
guiguider 2.0.1におけるインターフェース非表示の問題 GuiGuider 2.0.1では、インターフェースの右側のプロパティに「hidden」フラグを追加すると、他のインターフェースも非表示になってしまうことがあり、ソフトウェアを再起動しないと修正できません。 回复: 界面隐藏问题guiguider2.0.1 こんにちは、 @ZYQさん 情報ありがとうございます。 この問題を再現できます。 そして、その件はGUIGUIDERチームに報告しました。 BR ハリー
View full article
FS26 standby Hi,NXP专家,我遇到一个FS26相关的问题: 测试发现FS26无法进入standby模式,单步运行的时候可以运行下去,但是FS26无法进入standby。只要全速运行Sbc_fs26_SetOperatingMode()接口的时候会出现复位。硬件上的Wake信号已经去掉了,且reset引脚上没有接其他信号。 Small Engine Driver Re: FS26 standby Hi Chenxu 您是从哪个状态切换到standby状态? 如何发的命令能给出吗? Re: FS26 standby Hi,谢谢你的帮助,这个问题解决了。 测试发现当唤醒时间小于5ms的话,读M_WIO_FLG寄存器出现0,读不到唤醒源的情况。可以帮忙回答这个问题吗,谢谢 Re: FS26 standby HI CHENXU 新的topic请重新提交新的ticket吧! https://support.nxp.com/s/?language=en_US
View full article
S32K312 ICU SAIC DMA NXPチームの皆様へ: K3のICUモジュールに関して、以下の質問があります。 ICU入力チャネルを、データ転送にDMAを使用してSAICモードでPWMを測定するように構成します。まず、SAIC DMAモードRTDコードはダブルエッジトリガを使用するように強制されます。次に、ダブルエッジトリガモードでは、入力PWM信号線が遮断されてから再接続された場合、再接続時にPWM波形がハイレベルであれば、次のトリガエッジは立ち下がりエッジとなり、後続の測定のデューティサイクルが反転します。たとえば、PWM波形が20 %占空比,ICU的测量结果就是80%の場合 完全なPWM波形は以下のとおりです。 上昇エッジ→高レベル→下降エッジ→低レベル デュアルエッジトリガーモードでは、立ち下がりエッジからサンプリングを開始し、前の波形の低レベル領域と次の波形の高レベル領域を再構成して単一の波形を作成することが可能です。 以前に似たような問題に遭遇したことはありますか?何か良い解決策をご存知でしたら教えてください。 ありがとう! Re: S32K312 ICU SAIC DMA こんにちは、 あなたが述べた動作は、SAICを両端トリガーで使用し、ソフトウェアがキャプチャされたタイムスタンプの順序からPWMのハイ/ロータイムを判定した場合に起こり得ます。入力信号が途切れた後、次に捕獲されるエッジはフォーリングエッジとなり、測定極性が逆になることがあります。 しかし、eMIOS SAICはこのCASEに特化したエッジキャプチャ機能を提供しています。IcuSaicEdgeCapturingが有効になっている場合、AS2ビット31には、キャプチャされたタイムスタンプに関連付けられた入力信号レベルが含まれます。これにより、ソフトウェアは両辺のトリガーが使われても昇降エッジを識別でき、信号再接続後に正しいPWMエッジシーケンスを復元できます。 この機能を有効にし、DMAサンプルをプロセッシングする際に取得した信号レベル情報を使用することを検討してください。AS2信号レベルのビットが転送され、保持されていることを確認するために、DMAデータの正確な処理方法をチェックする必要があります。 あるいは、専用のeMIOS IPW/IPWM測定モードがアプリケーション/チャネル構成に適合する場合も検討できます。これらのモードはパルス幅/PWM測定専用を意図しています。 BR、ペトル Re: S32K312 ICU SAIC DMA 上記の手順に従ってデバッグを試みます。ご返信いただき、誠にありがとうございます!
View full article
MC33774A(400μA)におけるIC間電流不均衡の解決策 こんにちは、チームのみなさん。 MC33774Aのデータシートには、IC間の電流消費の変動は最大400uAに達すると記載されています。そんな大きな不均衡の理由を教えてもらえますか? この問題を解決するための方法はありますか?実際、私たちはこのCMUを、分散型高電圧BMSシステムのスレーブ機器として使用しています。この不均衡は私たちに悪影響を及ぼす可能性があります。 可能であれば、IC間の電流不均衡に関する正規分布曲線や関連データも共有していただけますか? ありがとうございます Re: Solution for IC to IC current imbalance for MC33774A (400uA) こんにちは、 Hwd_kitさん 400 µAという値は、規定された動作条件下における2つのMC33774Aデバイス間の最大電流差であり、すべてのデバイスペアにおける典型的な不一致値ではありません。主に、内部バイアス回路、基準電圧回路、レギュレータ、ADC、およびデジタル回路におけるプロセス変動と温度変動を対象としています。 すべてのCMUでSYS_SUPPLY_CFG.CURMATCHが有効になっていることを確認してください。この機能は、デバイス間の電流差を指定されたΔIBAT(FP)値に制限します。 分散型高電圧BMSの場合、電流整合機能を有効にし、システム電力バジェットおよびバッテリー不均衡解析において、指定された400μAの不整合を考慮してください。 残念ながら、データシートには正規分布曲線や大規模サンプルに基づく統計データは記載されていません。 よろしくお願いします!
View full article
I.MX8 NPUのScarthgapとWrynoseの性能差 こんにちは、NXPさん。 NXPが提供する`mobilenet_v1_1.0_224_quant.tflite`のベンチマークテストを実施しました。VX NPUデリゲートを用いたモデル。以下の結果は、Scarthgap 6.6.23および Wrynose 6.18.37 で繰り返しベンチマークを実行した際の、定常状態の `Inference (avg)` 実行時間です。 Wrynoseの平均推論時間は、Scarthgapの結果よりも約27%長い。 この差は想定内ですか?確認すべきNPU、デリゲート、またはベンチマークの設定はありますか? ベンチマークコマンドは以下のとおりです。 ```bash ./benchmark_model--graph=mobilenet_v1_1.0_224_quant.tflite --external_delegate_path=/usr/lib/libvx_delegate.so 「`」 敬具 ガブリエル・ゴスデン Re: I.MX8 NPU performance difference between Scarthgap and Wrynose こんにちは、 i.MX の8M Plus EVKでも同様の性能差を再現できました。同じMobileNetモデルとテスト構成を用いて、Scarthgapで約2.96ms、Wrynoseで3.55msと約20%の推論時間増加を測定しました。 私のテストでは、両方のリリースで完全なモデルが正常に委任されており、関連するNPU/MLクロック周波数も同じレートで設定されていることが確認されました。プロファイリングによって、パフォーマンスの差はVXデリゲートの実行パス内に留まっていることも確認しました。 この挙動は再現可能であり、利用可能な結果はCPUフォールバックや低いNPUクロック周波数ではなく、ソフトウェア実行スタックに関連していることが示されています。
View full article