Programmable Vision Accelerator (PVA)#
NVIDIA DriveOS provides support for the Programmable Vision Accelerator (PVA) engine. The PVA engine is an advanced VLIW SIMD digital signal processor, which is optimized to implement image processing and computer vision algorithms. PVA provides excellent performance with extremely low power consumption. PVA can be used asynchronously and concurrently with the CPU, GPU, and other accelerators on the DRIVE platform as part of a heterogeneous compute pipeline.
Enabling VPU Authentication#
VPU Application authentication can be enabled by adding below property in Device Tree for production builds. BSP image needs to be built after adding this property in the Device Tree file.
Device Tree File (THOR)
hardware/nvidia/t264/nv-public/tegra264.dtsi
Property
nvidia,vpu-auth = <0x1>;
Example
pva0@818c000000 {
compatible = "nvidia,tegra264-pva", "simple-bus";
reg = <0x81 0x8c000000 0x0 0x900000>,
<0x0 0x10700000 0x0 0x100000>;
interrupts = <GIC_SPI 0x195 IRQ_TYPE_LEVEL_HIGH>,
<GIC_SPI 0x196 IRQ_TYPE_LEVEL_HIGH>,
<GIC_SPI 0x197 IRQ_TYPE_LEVEL_HIGH>,
<GIC_SPI 0x198 IRQ_TYPE_LEVEL_HIGH>,
<GIC_SPI 0x199 IRQ_TYPE_LEVEL_HIGH>,
<GIC_SPI 0x19a IRQ_TYPE_LEVEL_HIGH>,
<GIC_SPI 0x19b IRQ_TYPE_LEVEL_HIGH>,
<GIC_SPI 0x19c IRQ_TYPE_LEVEL_HIGH>,
<GIC_SPI 0x19d IRQ_TYPE_LEVEL_HIGH>;
resets = <&bpmp TEGRA264_RESET_PVA0_ALL>;
reset-names = "pva0";
clocks = <&bpmp TEGRA264_CLK_PVA0_CPU_AXI>,
<&bpmp TEGRA264_CLK_NAFLL_PVA0_VPS>,
<&bpmp TEGRA264_CLK_PVA0_VPS>;
clock-names = "axi", "vps0", "vps1";
power-domains = <&bpmp TEGRA264_POWER_DOMAIN_PVA0>;
iommus = <&smmu4_mmu TEGRA_SID_PVA>;
nvidia,vpu-auth = <0x1>;
dma-coherent;
status = "disabled";
};
If you are planning to build PVA applications or to use PVA applications built in to DriveWorks, update the PVA allowlist located at the following IFS path:
<rootfs>/lib/firmware/pva_auth_allowlist
NVIDIA DriveWorks is the primary API for the PVA engine. Refer to DriveWorks API documentation for supported algorithms and usage.
With additional software, it is also possible to program the PVA directly with custom algorithms. Contact NVIDIA for more information.