Add HIP-RT support for rendering on AMD GPUs

.gitignore

@@ -2,6 +2,7 @@
 .#*
 #*#
 src/build
+src/ext/hiprtSdk/hiprt
 .DS_Store
 .ipynb_checkpoints/
 build/

0001-MSVC-HIP.patch

@@ -0,0 +1,24 @@
+From 56933471af36147e1032fbbc7912ca0088797b78 Mon Sep 17 00:00:00 2001
+From: Aaryaman Vasishta <[email protected]>
+Date: Sun, 10 Dec 2023 18:40:22 +0900
+Subject: [PATCH] Fix BitScanreverse by including intrin.h for MSVC
+
+---
+ common/compiler_msc.h | 1 +
+ 1 file changed, 1 insertion(+)
+
+diff --git a/common/compiler_msc.h b/common/compiler_msc.h
+index 18cfa12..2eafd55 100644
+--- a/common/compiler_msc.h
++++ b/common/compiler_msc.h
+@@ -4,6 +4,7 @@
+
+ #include <stdint.h>
+ #include <stdlib.h> /* for _byteswap_*() */
++#include <intrin.h>
+
+ #define LIBEXPORT	__declspec(dllexport)
+
+-- 
+2.33.0.windows.2
+

LICENSE.txt

@@ -1,3 +1,7 @@
+All of the modifications made in commit ID 14582eebabcf1ca33323e6b6d6757a49d2306dbf are covered by the MIT license (below).
+The rest would be covered by the existing Apache license.
+
+---------------------------------------------------------
 
                                  Apache License
                            Version 2.0, January 2004
@@ -200,3 +204,28 @@
    WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
    See the License for the specific language governing permissions and
    limitations under the License.
+
+
+---------------------------------------------------------
+
+The MIT License (MIT)
+
+Copyright (c) 2022-2024 Advanced Micro Devices, Inc.
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in
+all copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
+THE SOFTWARE.

README.md

@@ -228,3 +228,33 @@ is straightforward:
 ```bash
 $ imgtool denoise-optix noisy.exr --outfile denoised.exr
 ```
+
+Instructions to build the HIP port
+--------
+
+Linux:
+* Install [ROCm](https://rocm.docs.amd.com/projects/install-on-linux/en/latest/install/native-install/ubuntu.html)
+* Download and extract [HIPRT](https://gpuopen.com/hiprt/)
+  * Note that HIP is very sensitive to the difference between versions of HIPRT and the application on Linux
+  * If you encounter linking error, we recommend to compile HIPRT by yourself
+* `cmake -DCMAKE_C_COMPILER=/opt/rocm/llvm/bin/clang -DCMAKE_CXX_COMPILER=/opt/rocm/llvm/bin/clang++ -DPBRT_HIPRT=ON -DPBRT_HIPRT_PATH=~/hiprtSdk ..`
+* `make` or `make pbrt_exe`
+
+Windows:
+* Download and install [HIP SDK](https://www.amd.com/en/developer/resources/rocm-hub/hip-sdk.html)
+  * Make sure that you checked HIPRT to be installed
+* Download and install [Strawberry Perl](https://strawberryperl.com/)
+* Run `x64 Native Tools Command Prompt for VS 2022` as administrator
+* `set CC=clang`
+* `set CXX=clang++`
+* `mkdir build` and `cd build`
+* `cmake -G "Ninja" -DCMAKE_BUILD_TYPE=Release -DPBRT_HIPRT=ON ..`
+* `cmake --build . --config=Release`
+
+Example: `./pbrt --gpu ~/pbrt-v4-scenes/killeroos/killeroo-simple.pbrt`
+Add `--interactive` for the interactive mode.
+
+GPU architecture and ROCm version
+--------
+
+The instructions above assume the `gfx1100` architecture. You can specify other architectures via the CMake `AMDGPU_TARGETS` variable. Note that PBRT should be compiled with the same version as HIPRT binaries (e.g., using ROCm 6.0 and `hiprt02004_6.0_amd_lib_linux.bc`).

src/ext/flip/flip.cpp

@@ -50,6 +50,10 @@
 #include <fstream>
 #include <cassert>
 
+#ifndef M_PI
+#define M_PI 3.14159265358979323846f
+#endif
+
 namespace flip_detail {
 
 class histogram

src/pbrt/base/medium.h

@@ -88,7 +88,6 @@ class Medium
 
     std::string ToString() const;
 
-    PBRT_CPU_GPU
     bool IsEmissive() const;
 
     PBRT_CPU_GPU

src/pbrt/cmd/imgtool.cpp

@@ -7,9 +7,9 @@
 #include <pbrt/filters.h>
 #include <pbrt/options.h>
 #ifdef PBRT_BUILD_GPU_RENDERER
-#ifndef __HIP_PLATFORM_AMD__
+#ifdef __NVCC__
 #include <pbrt/gpu/optix/denoiser.h>
-#endif // __HIP_PLATFORM_AMD__
+#endif // __NVCC__
 #include <pbrt/gpu/util.h>
 #endif  // PBRT_BUILD_GPU_RENDERER
 #include <pbrt/util/args.h>
@@ -2220,7 +2220,7 @@ int makeequiarea(std::vector<std::string> args) {
     return 0;
 }
 
-#ifdef PBRT_BUILD_GPU_RENDERER
+#ifdef __NVCC__
 int denoise_optix(std::vector<std::string> args) {
     std::string inFilename, outFilename;
 
@@ -2335,7 +2335,7 @@ int denoise_optix(std::vector<std::string> args) {
 
     return 0;
 }
-#endif  // PBRT_BUILD_GPU_RENDERER
+#endif  // __NVCC__
 
 int main(int argc, char *argv[]) {
     PBRTOptions opt;
@@ -2362,10 +2362,10 @@ int main(int argc, char *argv[]) {
         return convert(args);
     else if (cmd == "diff")
         return diff(args);
-#ifdef PBRT_BUILD_GPU_RENDERER
+#ifdef __NVCC__
     else if (cmd == "denoise-optix")
         return denoise_optix(args);
-#endif  // PBRT_BUILD_GPU_RENDERER
+#endif  // __NVCC__
     else if (cmd == "error")
         return error(args);
     else if (cmd == "falsecolor")

src/pbrt/cmd/pbrt_test.cpp

@@ -2,14 +2,14 @@
 // The pbrt source code is licensed under the Apache License, Version 2.0.
 // SPDX: Apache-2.0
 
+#include <gtest/gtest.h>
 #include <pbrt/pbrt.h>
 
 #include <pbrt/options.h>
 #include <pbrt/util/args.h>
 #include <pbrt/util/error.h>
 #include <pbrt/util/print.h>
 
-#include <gtest/gtest.h>
 #include <string>
 
 using namespace pbrt;

src/pbrt/cmd/pspec_gpu.cpp

@@ -16,8 +16,12 @@
 #include <pbrt/util/image.h>
 #include <pbrt/util/vecmath.h>
 
+#if defined(__HIPCC__)
+#include <pbrt/util/hip_aliases.h>
+#else
 #include <cuda.h>
 #include <cuda_runtime_api.h>
+#endif
 
 #include <vector>
 

src/pbrt/gpu/common.h

@@ -0,0 +1,111 @@
+#ifndef PBRT_GPU_COMMON_H
+#define PBRT_GPU_COMMON_H
+
+#include <pbrt/pbrt.h>
+
+#include <pbrt/base/light.h>
+#include <pbrt/base/material.h>
+#include <pbrt/base/medium.h>
+#include <pbrt/base/shape.h>
+#include <pbrt/base/texture.h>
+#include <pbrt/util/pstd.h>
+#include <pbrt/wavefront/workitems.h>
+#include <pbrt/wavefront/workqueue.h>
+
+#if defined(__HIPCC__)
+#include <hiprt/hiprt.h>
+#include <hiprt/hiprt_vec.h>
+#else
+#include <optix.h>
+#endif
+
+namespace pbrt {
+
+class TriangleMesh;
+class BilinearPatchMesh;
+
+struct TriangleMeshRecord {
+    const TriangleMesh *mesh;
+    Material material;
+    FloatTexture alphaTexture;
+    pstd::span<Light> areaLights;
+    MediumInterface *mediumInterface;
+};
+
+struct BilinearMeshRecord {
+    const BilinearPatchMesh *mesh;
+    Material material;
+    FloatTexture alphaTexture;
+    pstd::span<Light> areaLights;
+    MediumInterface *mediumInterface;
+};
+
+struct QuadricRecord {
+    Shape shape;
+    Material material;
+    FloatTexture alphaTexture;
+    Light areaLight;
+    MediumInterface *mediumInterface;
+};
+
+#if defined(__HIP_PLATFORM_AMD__)
+static constexpr size_t HitgroupAlignment = 16u;
+
+struct alignas(HitgroupAlignment) HitgroupRecord {
+    PBRT_CPU_GPU HitgroupRecord() {}
+    PBRT_CPU_GPU HitgroupRecord(const HitgroupRecord &r) {
+        memcpy(this, &r, sizeof(HitgroupRecord));
+    }
+    PBRT_CPU_GPU HitgroupRecord &operator=(const HitgroupRecord &r) {
+        if (this != &r)
+            memcpy(this, &r, sizeof(HitgroupRecord));
+        return *this;
+    }
+
+    union {
+        TriangleMeshRecord triRec;
+        BilinearMeshRecord blpRec;
+        QuadricRecord quadricRec;
+    };
+    enum { TriangleMesh, BilinearMesh, Quadric } type;
+};
+#endif
+
+struct RayIntersectParameters {
+#if defined(__HIPCC__)
+    hiprtScene traversable;
+#else
+    OptixTraversableHandle traversable;
+#endif
+
+    const RayQueue *rayQueue;
+
+    // Closest hit
+    RayQueue *nextRayQueue;
+    EscapedRayQueue *escapedRayQueue;
+    HitAreaLightQueue *hitAreaLightQueue;
+    MaterialEvalQueue *basicEvalMaterialQueue, *universalEvalMaterialQueue;
+    MediumSampleQueue *mediumSampleQueue;
+
+    // Shadow rays
+    ShadowRayQueue *shadowRayQueue;
+    SOA<PixelSampleState> pixelSampleState;
+
+    // Subsurface scattering...
+    SubsurfaceScatterQueue *subsurfaceScatterQueue;
+
+#if defined(__HIPCC__)
+    // Stack buffers
+    hiprtGlobalStackBuffer globalStackBuffer;
+    hiprtGlobalStackBuffer globalInstanceStackBuffer;
+    // Custom function table
+    hiprtFuncTable funcTable;
+    // Hitgroup records
+    HitgroupRecord *hgRecords;
+    // Offsets for hitgroup records
+    uint32_t *offsets;
+#endif
+};
+}  // namespace pbrt
+
+#endif  // PBRT_GPU_COMMON_H

src/pbrt/gpu/cudagl.h

@@ -34,9 +34,13 @@
 
 #include <glad/glad.h>
 
+#if defined(__HIPCC__)
+#include <pbrt/util/hip_aliases.h>
+#else
 #include <cuda.h>
 #include <cuda_gl_interop.h>
 #include <cuda_runtime.h>
+#endif
 
 #define GL_CHECK(call)                                                   \
     do {                                                                 \
@@ -370,6 +374,15 @@ CUDAOutputBuffer<PIXEL_FORMAT>::CUDAOutputBuffer(int32_t width, int32_t height)
                           nullptr, GL_STREAM_DRAW));
     GL_CHECK(glBindBuffer(GL_ARRAY_BUFFER, 0u));
 
+#ifdef __HIPCC__
+    uint32_t num_gl_devices = 0;
+
+    int glDevice;
+    cudaGLGetDevices(&num_gl_devices, &glDevice, 1, cudaGLDeviceListAll);
+
+    if (glDevice != current_device)
+        LOG_FATAL("Multi-GPU not supported with GL interop yet");
+#endif
     CUDA_CHECK(cudaGraphicsGLRegisterBuffer(&m_cuda_gfx_resource, m_pbo,
                                             cudaGraphicsMapFlagsWriteDiscard));