Add and update samples for cuda 11.0 support

2026-10-23 13:18:16 +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/vulkanImageCUDA/Makefile
+++ b/Samples/vulkanImageCUDA/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)
@@ -268,12 +289,18 @@ include ./findvulkan.mk

 # Vulkan specific libraries
 ifeq ($(TARGET_OS),linux)
- LIBRARIES += -L $(VULKAN_SDK_PATH)/lib
+ ifneq ($(TARGET_ARCH),$(HOST_ARCH))
+  LIBRARIES += -L$(VULKAN_SDK_LIB) -lvulkan
+  LIBRARIES += -lglfw
+  INCLUDES  += -I$(VULKAN_HEADER)
+ else
+ LIBRARIES += -L$(VULKAN_SDK_LIB)
 LIBRARIES += `pkg-config --static --libs glfw3` -lvulkan
- INCLUDES  += `pkg-config --static --cflags glfw3` -I$(VULKAN_SDK_PATH)/include
+ INCLUDES  += `pkg-config --static --cflags glfw3` -I$(VULKAN_HEADER)
+ endif
 endif

-#Detect if installed version of GCC supports C++11
+#Detect if installed version of GCC supports required C++11
 ifeq ($(TARGET_OS),linux)
    empty :=
    space := $(empty) $(empty)
@@ -295,16 +322,16 @@ ifeq ($(TARGET_OS),linux)
    ifeq ($(IS_MIN_VERSION), 1)
        $(info >>> GCC Version is greater or equal to 4.7.0 <<<)
    else
-        $(info >>> Waiving build. Minimum GCC version required for C++11 is 4.7.0 <<<)
+        $(info >>> Waiving build. Minimum GCC version required is 4.7.0<<<)
        SAMPLE_ENABLED := 0
    endif
 endif

 # Gencode arguments
 ifeq ($(TARGET_ARCH),$(filter $(TARGET_ARCH),armv7l aarch64))
-SMS ?= 30 35 37 50 52 60 61 70 72 75
+SMS ?= 35 37 50 52 60 61 70 72 75 80
 else
-SMS ?= 30 35 37 50 52 60 61 70 75
+SMS ?= 35 37 50 52 60 61 70 75 80
 endif

 ifeq ($(SMS),)
--- a/Samples/vulkanImageCUDA/NsightEclipse.xml
+++ b/Samples/vulkanImageCUDA/NsightEclipse.xml
@@ -47,7 +47,6 @@
    <scope>1:CUDA Advanced Topics</scope>
    <scope>1:CUDA Vulkan Interop</scope>
  </scopes>
-  <sm-arch>sm30</sm-arch>
  <sm-arch>sm35</sm-arch>
  <sm-arch>sm37</sm-arch>
  <sm-arch>sm50</sm-arch>
@@ -57,6 +56,7 @@
  <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/vulkanImageCUDA/README.md
+++ b/Samples/vulkanImageCUDA/README.md
@@ -10,7 +10,7 @@ Graphics Interop, CUDA Vulkan Interop, Data Parallel Algorithms

 ## 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 @@ cudaImportExternalMemory, cudaExternalMemoryGetMappedMipmappedArray, cudaImportE

 ## 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/vulkanImageCUDA/findvulkan.mk
+++ b/Samples/vulkanImageCUDA/findvulkan.mk
@@ -51,7 +51,7 @@ ifeq ("$(TARGET_OS)","linux")
 endif

 ifeq ("$(TARGET_OS)","linux")
-    # Each set of Linux Distros have different paths for where to find their GLM/GLFW3 libraries reside
+    # Each set of Linux Distros have different paths for where to find libraries
    UBUNTU = $(shell echo $(DISTRO) | grep -i ubuntu      >/dev/null 2>&1; echo $$?)
    FEDORA = $(shell echo $(DISTRO) | grep -i fedora      >/dev/null 2>&1; echo $$?)
    RHEL   = $(shell echo $(DISTRO) | grep -i 'red\|rhel' >/dev/null 2>&1; echo $$?)
@@ -107,16 +107,17 @@ ifeq ("$(TARGET_OS)","linux")
  VULKAN_SDK_PATH ?= ${VULKAN_SDK}

  ifeq ("$(VULKAN_SDK_PATH)","")
-      $(info >>> WARNING - Vulkan SDK not found, please install Vulkan SDK <<<)
-      SAMPLE_ENABLED := 0
+      VULKAN_SDK_PATH := $(DFLT_PATH)
  endif

-  VULKAN_SDK_LIB  := $(shell find -L $(VULKAN_SDK_PATH)/lib -name libvulkan.so    -print 2>/dev/null)
+  VULKAN_SDK_LIB  := $(shell find -L $(VULKAN_SDK_PATH) -name libvulkan.so    -print 2>/dev/null)
  X11LIB          := $(shell find -L $(GLPATH) $(DFLT_PATH) -name libX11.so    -print 2>/dev/null)

  ifeq ("$(VULKAN_SDK_LIB)","")
-      $(info >>> WARNING - libvulkan.so not found, please install libvulkan.so <<<)
+      $(info >>> WARNING - libvulkan.so not found, please install Vulkan SDK and pass VULKAN_SDK_PATH=<PATH_TO_VULKAN_SDK> <<<)
      SAMPLE_ENABLED := 0
+  else
+      VULKAN_SDK_LIB := $(shell echo $(VULKAN_SDK_LIB) | sed "s/ .*//" | sed "s/\/libvulkan.so//" )
  endif

  ifeq ("$(X11LIB)","")
@@ -132,11 +133,13 @@ ifeq ("$(TARGET_OS)","linux")
      HEADER_SEARCH_PATH += /usr/aarch64-linux-gnu/include
  endif

-  VULKANHEADER  := $(shell find -L $(VULKAN_SDK_PATH)/include -name vulkan.h -print 2>/dev/null)
+  VULKAN_HEADER  := $(shell find -L $(VULKAN_SDK_PATH) $(HEADER_SEARCH_PATH) -name vulkan.h -print 2>/dev/null)

-  ifeq ("$(VULKANHEADER)","")
+  ifeq ("$(VULKAN_HEADER)","")
      $(info >>> WARNING - vulkan.h not found, please install vulkan.h <<<)
      SAMPLE_ENABLED := 0
+  else
+      VULKAN_HEADER := $(shell echo $(VULKAN_HEADER) | sed "s/ .*//" | sed "s/\/vulkan\/vulkan.h//" )
  endif
 else
 endif
--- a/Samples/vulkanImageCUDA/vulkanImageCUDA_vs2015.vcxproj
+++ b/Samples/vulkanImageCUDA/vulkanImageCUDA_vs2015.vcxproj
@@ -62,7 +62,7 @@
      <OutputFile>$(OutDir)/vulkanImageCUDA.exe</OutputFile>
    </Link>
    <CudaCompile>
-      <CodeGeneration>compute_30,sm_30;compute_35,sm_35;compute_37,sm_37;compute_50,sm_50;compute_52,sm_52;compute_60,sm_60;compute_61,sm_61;compute_70,sm_70;compute_75,sm_75;</CodeGeneration>
+      <CodeGeneration>compute_35,sm_35;compute_37,sm_37;compute_50,sm_50;compute_52,sm_52;compute_60,sm_60;compute_61,sm_61;compute_70,sm_70;compute_75,sm_75;compute_80,sm_80;</CodeGeneration>
      <AdditionalOptions>-Xcompiler "/wd 4819" %(AdditionalOptions)</AdditionalOptions>
      <Include>./;../../Common</Include>
      <Defines>WIN32</Defines>
--- a/Samples/vulkanImageCUDA/vulkanImageCUDA_vs2017.vcxproj
+++ b/Samples/vulkanImageCUDA/vulkanImageCUDA_vs2017.vcxproj
@@ -67,7 +67,7 @@
      <OutputFile>$(OutDir)/vulkanImageCUDA.exe</OutputFile>
    </Link>
    <CudaCompile>
-      <CodeGeneration>compute_30,sm_30;compute_35,sm_35;compute_37,sm_37;compute_50,sm_50;compute_52,sm_52;compute_60,sm_60;compute_61,sm_61;compute_70,sm_70;compute_75,sm_75;</CodeGeneration>
+      <CodeGeneration>compute_35,sm_35;compute_37,sm_37;compute_50,sm_50;compute_52,sm_52;compute_60,sm_60;compute_61,sm_61;compute_70,sm_70;compute_75,sm_75;compute_80,sm_80;</CodeGeneration>
      <AdditionalOptions>-Xcompiler "/wd 4819" %(AdditionalOptions)</AdditionalOptions>
      <Include>./;../../Common</Include>
      <Defines>WIN32</Defines>
--- a/Samples/vulkanImageCUDA/vulkanImageCUDA_vs2019.vcxproj
+++ b/Samples/vulkanImageCUDA/vulkanImageCUDA_vs2019.vcxproj
@@ -63,7 +63,7 @@
      <OutputFile>$(OutDir)/vulkanImageCUDA.exe</OutputFile>
    </Link>
    <CudaCompile>
-      <CodeGeneration>compute_30,sm_30;compute_35,sm_35;compute_37,sm_37;compute_50,sm_50;compute_52,sm_52;compute_60,sm_60;compute_61,sm_61;compute_70,sm_70;compute_75,sm_75;</CodeGeneration>
+      <CodeGeneration>compute_35,sm_35;compute_37,sm_37;compute_50,sm_50;compute_52,sm_52;compute_60,sm_60;compute_61,sm_61;compute_70,sm_70;compute_75,sm_75;compute_80,sm_80;</CodeGeneration>
      <AdditionalOptions>-Xcompiler "/wd 4819" %(AdditionalOptions)</AdditionalOptions>
      <Include>./;../../Common</Include>
      <Defines>WIN32</Defines>