Add and update samples for cuda 11.0 support

2026-10-23 21:28:17 +08:00 · 2020-05-18 22:22:06 +05:30
parent 6be514679b
commit 4f6e02970b
410 changed files with 21920 additions and 3864 deletions
--- a/Samples/EGLStream_CUDA_Interop/Makefile
+++ b/Samples/EGLStream_CUDA_Interop/Makefile
@@ -72,9 +72,9 @@ endif
 # architecture
 HOST_ARCH   := $(shell uname -m)
 TARGET_ARCH ?= $(HOST_ARCH)
-ifneq (,$(filter $(TARGET_ARCH),x86_64 aarch64 ppc64le armv7l))
+ifneq (,$(filter $(TARGET_ARCH),x86_64 aarch64 sbsa ppc64le armv7l))
    ifneq ($(TARGET_ARCH),$(HOST_ARCH))
-        ifneq (,$(filter $(TARGET_ARCH),x86_64 aarch64 ppc64le))
+        ifneq (,$(filter $(TARGET_ARCH),x86_64 aarch64 sbsa ppc64le))
            TARGET_SIZE := 64
        else ifneq (,$(filter $(TARGET_ARCH),armv7l))
            TARGET_SIZE := 32
@@ -85,8 +85,17 @@ ifneq (,$(filter $(TARGET_ARCH),x86_64 aarch64 ppc64le armv7l))
 else
    $(error ERROR - unsupported value $(TARGET_ARCH) for TARGET_ARCH!)
 endif
+
+# sbsa and aarch64 systems look similar. Need to differentiate them at host level for now.
+ifeq ($(HOST_ARCH),aarch64)
+    ifeq ($(CUDA_PATH)/targets/sbsa-linux,$(shell ls -1d $(CUDA_PATH)/targets/sbsa-linux))
+        HOST_ARCH := sbsa
+        TARGET_ARCH := sbsa
+    endif
+endif
+
 ifneq ($(TARGET_ARCH),$(HOST_ARCH))
-    ifeq (,$(filter $(HOST_ARCH)-$(TARGET_ARCH),aarch64-armv7l x86_64-armv7l x86_64-aarch64 x86_64-ppc64le))
+    ifeq (,$(filter $(HOST_ARCH)-$(TARGET_ARCH),aarch64-armv7l x86_64-armv7l x86_64-aarch64 x86_64-sbsa x86_64-ppc64le))
        $(error ERROR - cross compiling from $(HOST_ARCH) to $(TARGET_ARCH) is not supported!)
    endif
 endif
@@ -141,6 +150,8 @@ else ifneq ($(TARGET_ARCH),$(HOST_ARCH))
        else ifeq ($(TARGET_OS), android)
            HOST_COMPILER ?= aarch64-linux-android-clang++
        endif
+    else ifeq ($(TARGET_ARCH),sbsa)
+        HOST_COMPILER ?= aarch64-linux-gnu-g++
    else ifeq ($(TARGET_ARCH),ppc64le)
        HOST_COMPILER ?= powerpc64le-linux-gnu-g++
    endif
@@ -185,19 +196,27 @@ ifneq ($(TARGET_ARCH),$(HOST_ARCH))
                CCFLAGS += --sysroot=$(TARGET_FS)
            endif
            LDFLAGS += --sysroot=$(TARGET_FS)
-            LDFLAGS += -rpath-link=$(TARGET_FS)/lib -L $(TARGET_FS)/lib
-            LDFLAGS += -rpath-link=$(TARGET_FS)/usr/lib -L $(TARGET_FS)/usr/lib
-            LDFLAGS += -rpath-link=$(TARGET_FS)/usr/lib/aarch64-linux-gnu -L $(TARGET_FS)/usr/lib/aarch64-linux-gnu
+            LDFLAGS += -rpath-link=$(TARGET_FS)/lib -L$(TARGET_FS)/lib
+            LDFLAGS += -rpath-link=$(TARGET_FS)/lib/aarch64-linux-gnu -L$(TARGET_FS)/lib/aarch64-linux-gnu
+            LDFLAGS += -rpath-link=$(TARGET_FS)/usr/lib -L$(TARGET_FS)/usr/lib
+            LDFLAGS += -rpath-link=$(TARGET_FS)/usr/lib/aarch64-linux-gnu -L$(TARGET_FS)/usr/lib/aarch64-linux-gnu
            LDFLAGS += --unresolved-symbols=ignore-in-shared-libs
-            CCFLAGS += -isystem=$(TARGET_FS)/usr/include
-            CCFLAGS += -isystem=$(TARGET_FS)/usr/include/aarch64-linux-gnu
+            CCFLAGS += -isystem=$(TARGET_FS)/usr/include  -I$(TARGET_FS)/usr/include
+            CCFLAGS += -isystem=$(TARGET_FS)/usr/include/aarch64-linux-gnu -I$(TARGET_FS)/usr/include/aarch64-linux-gnu
+        endif
+    endif
+    ifeq ($(TARGET_ARCH)-$(TARGET_OS),aarch64-qnx)
+        CCFLAGS += -DWIN_INTERFACE_CUSTOM -I/usr/include/aarch64-qnx-gnu
+        LDFLAGS += -lsocket
+        LDFLAGS += -rpath=/usr/lib/aarch64-qnx-gnu -L/usr/lib/aarch64-qnx-gnu
+        ifneq ($(TARGET_FS),)
+            LDFLAGS += -rpath=$(TARGET_FS)/usr/lib -L $(TARGET_FS)/usr/lib
+            LDFLAGS += -rpath=$(TARGET_FS)/usr/libnvidia -L $(TARGET_FS)/usr/libnvidia
+        endif
+        ifdef TARGET_OVERRIDE # cuda toolkit targets override
+            NVCCFLAGS += -target-dir $(TARGET_OVERRIDE)
        endif
    endif
-endif
-
-ifeq ($(TARGET_OS),qnx)
-    CCFLAGS += -DWIN_INTERFACE_CUSTOM
-    LDFLAGS += -lsocket
 endif

 # Install directory of different arch
@@ -206,6 +225,8 @@ ifeq ($(TARGET_ARCH)-$(TARGET_OS),armv7l-linux)
    CUDA_INSTALL_TARGET_DIR = targets/armv7-linux-gnueabihf/
 else ifeq ($(TARGET_ARCH)-$(TARGET_OS),aarch64-linux)
    CUDA_INSTALL_TARGET_DIR = targets/aarch64-linux/
+else ifeq ($(TARGET_ARCH)-$(TARGET_OS),sbsa-linux)
+    CUDA_INSTALL_TARGET_DIR = targets/sbsa-linux/
 else ifeq ($(TARGET_ARCH)-$(TARGET_OS),armv7l-android)
    CUDA_INSTALL_TARGET_DIR = targets/armv7-linux-androideabi/
 else ifeq ($(TARGET_ARCH)-$(TARGET_OS),aarch64-android)
@@ -289,6 +310,10 @@ else
    CUDA_SEARCH_PATH ?= $(CUDA_PATH)/targets/aarch64-linux/lib/stubs
  endif

+  ifeq ($(TARGET_ARCH)-$(TARGET_OS),sbsa-linux)
+    CUDA_SEARCH_PATH ?= $(CUDA_PATH)/targets/sbsa-linux/lib/stubs
+  endif
+
  ifeq ($(TARGET_ARCH)-$(TARGET_OS),armv7l-android)
    CUDA_SEARCH_PATH ?= $(CUDA_PATH)/targets/armv7-linux-androideabi/lib/stubs
  endif
@@ -303,12 +328,19 @@ else

  ifeq ($(TARGET_ARCH)-$(TARGET_OS),aarch64-qnx)
    CUDA_SEARCH_PATH ?= $(CUDA_PATH)/targets/aarch64-qnx/lib/stubs
+    ifdef TARGET_OVERRIDE
+        CUDA_SEARCH_PATH := $(CUDA_PATH)/targets/$(TARGET_OVERRIDE)/lib/stubs
+    endif
  endif

  ifeq ($(TARGET_ARCH),ppc64le)
    CUDA_SEARCH_PATH ?= $(CUDA_PATH)/targets/ppc64le-linux/lib/stubs
  endif

+  ifeq ($(HOST_ARCH),ppc64le)
+    CUDA_SEARCH_PATH += $(CUDA_PATH)/lib64/stubs
+  endif
+
  CUDALIB ?= $(shell find -L $(CUDA_SEARCH_PATH) -maxdepth 1 -name libcuda.so 2> /dev/null)
  ifeq ("$(CUDALIB)","")
    $(info >>> WARNING - libcuda.so not found, CUDA Driver is not installed.  Please re-install the driver. <<<)
--- a/Samples/EGLStream_CUDA_Interop/NsightEclipse.xml
+++ b/Samples/EGLStream_CUDA_Interop/NsightEclipse.xml
@@ -49,7 +49,6 @@
    <scope>1:CUDA Basic Topics</scope>
    <scope>2:Graphics Interop</scope>
  </scopes>
-  <sm-arch>sm30</sm-arch>
  <sm-arch>sm35</sm-arch>
  <sm-arch>sm37</sm-arch>
  <sm-arch>sm50</sm-arch>
@@ -59,6 +58,7 @@
  <sm-arch>sm70</sm-arch>
  <sm-arch>sm72</sm-arch>
  <sm-arch>sm75</sm-arch>
+  <sm-arch>sm80</sm-arch>
  <supported_envs>
    <env>
      <arch>x86_64</arch>
--- a/Samples/EGLStream_CUDA_Interop/README.md
+++ b/Samples/EGLStream_CUDA_Interop/README.md
@@ -10,7 +10,7 @@ EGLStreams Interop

 ## Supported SM Architectures

-[SM 3.0 ](https://developer.nvidia.com/cuda-gpus)  [SM 3.5 ](https://developer.nvidia.com/cuda-gpus)  [SM 3.7 ](https://developer.nvidia.com/cuda-gpus)  [SM 5.0 ](https://developer.nvidia.com/cuda-gpus)  [SM 5.2 ](https://developer.nvidia.com/cuda-gpus)  [SM 6.0 ](https://developer.nvidia.com/cuda-gpus)  [SM 6.1 ](https://developer.nvidia.com/cuda-gpus)  [SM 7.0 ](https://developer.nvidia.com/cuda-gpus)  [SM 7.2 ](https://developer.nvidia.com/cuda-gpus)  [SM 7.5 ](https://developer.nvidia.com/cuda-gpus)
+[SM 3.5 ](https://developer.nvidia.com/cuda-gpus)  [SM 3.7 ](https://developer.nvidia.com/cuda-gpus)  [SM 5.0 ](https://developer.nvidia.com/cuda-gpus)  [SM 5.2 ](https://developer.nvidia.com/cuda-gpus)  [SM 6.0 ](https://developer.nvidia.com/cuda-gpus)  [SM 6.1 ](https://developer.nvidia.com/cuda-gpus)  [SM 7.0 ](https://developer.nvidia.com/cuda-gpus)  [SM 7.2 ](https://developer.nvidia.com/cuda-gpus)  [SM 7.5 ](https://developer.nvidia.com/cuda-gpus)  [SM 8.0 ](https://developer.nvidia.com/cuda-gpus)

 ## Supported OSes

@@ -30,7 +30,7 @@ cuDeviceGet, cuDeviceGetAttribute, cuDeviceComputeCapability, cuDeviceGetCount,

 ## Prerequisites

-Download and install the [CUDA Toolkit 10.2](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
+Download and install the [CUDA Toolkit 11.0](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
 Make sure the dependencies mentioned in [Dependencies]() section above are installed.

 ## Build and Run
--- a/Samples/EGLStream_CUDA_Interop/cuda_producer.cpp
+++ b/Samples/EGLStream_CUDA_Interop/cuda_producer.cpp
@@ -245,12 +245,30 @@ CUresult cudaProducerTest(test_cuda_producer_s *cudaProducer, char *file) {
  cudaEgl.eglColorFormat = eglColorFormat;
  cudaEgl.cuFormat = CU_AD_FORMAT_UNSIGNED_INT8;

+  static int numFramesPresented = 0;
+  // If there is a frame presented before we check if consumer
+  // is done with it using cuEGLStreamProducerReturnFrame.
+  while (numFramesPresented) {
+    CUeglFrame returnedCudaEgl;
+    cuStatus = cuEGLStreamProducerReturnFrame(&cudaProducer->cudaConn,
+                                              &returnedCudaEgl, NULL);
+    if (cuStatus == CUDA_ERROR_LAUNCH_TIMEOUT) {
+      continue;
+    } else if (cuStatus != CUDA_SUCCESS) {
+      printf("cuda Producer return frame FAILED with custatus= %d\n", cuStatus);
+      return cuStatus;
+    } else {
+      numFramesPresented--;
+    }
+  }
+
  cuStatus =
      cuEGLStreamProducerPresentFrame(&cudaProducer->cudaConn, cudaEgl, NULL);
  if (cuStatus != CUDA_SUCCESS) {
    printf("cuda Producer present frame FAILED with custatus= %d\n", cuStatus);
    goto done;
  }
+  numFramesPresented++;

 done:
  if (file_p) {
@@ -281,6 +299,13 @@ CUresult cudaDeviceCreateProducer(test_cuda_producer_s *cudaProducer,
      "%d.%d\n\n",
      device, deviceName, major, minor);

+  if (major < 6) {
+    printf(
+        "EGLStreams_CUDA_Interop requires SM 6.0 or higher arch GPU.  "
+        "Exiting...\n");
+    exit(2);  // EXIT_WAIVED
+  }
+
  if (CUDA_SUCCESS !=
      (status = cuCtxCreate(&cudaProducer->context, 0, device))) {
    printf("failed to create CUDA context\n");
--- a/Samples/EGLStream_CUDA_Interop/findegl.mk
+++ b/Samples/EGLStream_CUDA_Interop/findegl.mk
@@ -133,10 +133,14 @@ ifeq ("$(TARGET_OS)","linux")
 else
 endif

+ifeq ("$(TARGET_OS)","qnx")
+    HOST_CCFLAGS := -V5.4.0,gcc_ntoaarch64le
+endif
+
 # Attempt to compile a minimal EGL application and run to check if EGL_SUPPORT_REUSE_NV is supported in the EGL headers available.
 ifneq ($(SAMPLE_ENABLED), 0)
      $(shell printf "#include <EGL/egl.h>\n#include <EGL/eglext.h>\nint main() {\n#ifdef EGL_SUPPORT_REUSE_NV \n #error \"Compatible EGL header found\" \n  return 0;\n#endif \n return 1;\n}"  > test.c; )
-      EGL_DEFINES := $(shell $(HOST_COMPILER) $(CCFLAGS) $(EXTRA_CCFLAGS) -lEGL test.c -c 2>&1 | grep -ic "Compatible EGL header found";)
+      EGL_DEFINES := $(shell $(HOST_COMPILER) $(HOST_CCFLAGS) $(CCFLAGS) $(EXTRA_CCFLAGS) -lEGL test.c -c 2>&1 | grep -ic "Compatible EGL header found";)
      SHOULD_WAIVE := 0
      ifeq ($(EGL_DEFINES),0)
        SHOULD_WAIVE := 1