[Kernel][ROCm][AMD] enable fused topk_softmax kernel for moe layer (v…

…llm-project#4927)

This PR enables the fused topk_softmax kernel used in moe layer for HIP

CMakeLists.txt

@@ -311,6 +311,9 @@ if(VLLM_GPU_LANG STREQUAL "CUDA" OR VLLM_GPU_LANG STREQUAL "HIP")
   message(STATUS "Enabling C extension.")
   add_dependencies(default _C)
 
+  message(STATUS "Enabling moe extension.")
+  add_dependencies(default _moe_C)
+
   # Enable punica if -DVLLM_INSTALL_PUNICA_KERNELS=ON or
   # VLLM_INSTALL_PUNICA_KERNELS is set in the environment and
   # there are supported target arches.
@@ -320,8 +323,3 @@ if(VLLM_GPU_LANG STREQUAL "CUDA" OR VLLM_GPU_LANG STREQUAL "HIP")
     add_dependencies(default _punica_C)
   endif()
 endif()
-
-if(VLLM_GPU_LANG STREQUAL "CUDA")
-  message(STATUS "Enabling moe extension.")
-  add_dependencies(default _moe_C)
-endif()

Dockerfile.rocm

@@ -108,6 +108,7 @@ RUN --mount=type=cache,target=/root/.cache/pip \
     && python3 setup.py install \
     && cp build/lib.linux-x86_64-cpython-39/vllm/_C.cpython-39-x86_64-linux-gnu.so vllm/ \
     && cp build/lib.linux-x86_64-cpython-39/vllm/_punica_C.cpython-39-x86_64-linux-gnu.so vllm/ \
+    && cp build/lib.linux-x86_64-cpython-39/vllm/_moe_C.cpython-39-x86_64-linux-gnu.so vllm/ \
     && cd ..
 
 

csrc/cuda_compat.h

@@ -19,8 +19,12 @@
 #ifndef USE_ROCM
   #define VLLM_SHFL_XOR_SYNC(var, lane_mask) \
     __shfl_xor_sync(uint32_t(-1), var, lane_mask)
+  #define VLLM_SHFL_XOR_SYNC_WIDTH(var, lane_mask, width) \
+    __shfl_xor_sync(uint32_t(-1), var, lane_mask, width)
 #else
   #define VLLM_SHFL_XOR_SYNC(var, lane_mask) __shfl_xor(var, lane_mask)
+  #define VLLM_SHFL_XOR_SYNC_WIDTH(var, lane_mask, width) \
+    __shfl_xor(var, lane_mask, width)
 #endif
 
 #ifndef USE_ROCM

csrc/moe/topk_softmax_kernels.cu

@@ -19,15 +19,22 @@
 #include <torch/extension.h>
 #include <ATen/cuda/CUDAContext.h>
 #include <c10/cuda/CUDAGuard.h>
+#include "../cuda_compat.h"
 
-#include <cub/cub.cuh>
-#include <cub/util_type.cuh>
+#ifndef USE_ROCM
+    #include <cub/util_type.cuh>
+    #include <cub/cub.cuh>
+#else
+    #include <hipcub/util_type.hpp>
+    #include <hipcub/hipcub.hpp>
+#endif
+
+#define MAX(a, b) ((a) > (b) ? (a) : (b))
+#define MIN(a, b) ((a) < (b) ? (a) : (b))
 
 namespace vllm {
 namespace moe {
 
-static constexpr int WARP_SIZE = 32;
-
 /// Aligned array type
 template <
     typename T,
@@ -265,7 +272,7 @@ __launch_bounds__(WARPS_PER_CTA* WARP_SIZE) __global__
 #pragma unroll
     for (int mask = THREADS_PER_ROW / 2; mask > 0; mask /= 2)
     {
-        thread_max = max(thread_max, __shfl_xor_sync(0xFFFFFFFF, thread_max, mask, THREADS_PER_ROW));
+        thread_max = max(thread_max, VLLM_SHFL_XOR_SYNC_WIDTH(thread_max, mask, THREADS_PER_ROW));
     }
 
     // From this point, thread max in all the threads have the max within the row.
@@ -282,7 +289,7 @@ __launch_bounds__(WARPS_PER_CTA* WARP_SIZE) __global__
 #pragma unroll
     for (int mask = THREADS_PER_ROW / 2; mask > 0; mask /= 2)
     {
-        row_sum += __shfl_xor_sync(0xFFFFFFFF, row_sum, mask, THREADS_PER_ROW);
+        row_sum += VLLM_SHFL_XOR_SYNC_WIDTH(row_sum, mask, THREADS_PER_ROW);
     }
 
     // From this point, all threads have the max and the sum for their rows in the thread_max and thread_sum variables
@@ -332,8 +339,8 @@ __launch_bounds__(WARPS_PER_CTA* WARP_SIZE) __global__
 #pragma unroll
         for (int mask = THREADS_PER_ROW / 2; mask > 0; mask /= 2)
         {
-            float other_max = __shfl_xor_sync(0xFFFFFFFF, max_val, mask, THREADS_PER_ROW);
-            int other_expert = __shfl_xor_sync(0xFFFFFFFF, expert, mask, THREADS_PER_ROW);
+            float other_max = VLLM_SHFL_XOR_SYNC_WIDTH(max_val, mask, THREADS_PER_ROW);
+            int other_expert = VLLM_SHFL_XOR_SYNC_WIDTH(expert, mask, THREADS_PER_ROW);
 
             // We want lower indices to "win" in every thread so we break ties this way
             if (other_max > max_val || (other_max == max_val && other_expert < expert))
@@ -383,7 +390,7 @@ struct TopkConstants
 {
     static constexpr int ELTS_PER_LDG = BYTES_PER_LDG / sizeof(float);
     static_assert(EXPERTS / (ELTS_PER_LDG * WARP_SIZE) == 0 || EXPERTS % (ELTS_PER_LDG * WARP_SIZE) == 0, "");
-    static constexpr int VECs_PER_THREAD = std::max(1, EXPERTS / (ELTS_PER_LDG * WARP_SIZE));
+    static constexpr int VECs_PER_THREAD = MAX(1, EXPERTS / (ELTS_PER_LDG * WARP_SIZE));
     static constexpr int VPT = VECs_PER_THREAD * ELTS_PER_LDG;
     static constexpr int THREADS_PER_ROW = EXPERTS / VPT;
     static constexpr int ROWS_PER_WARP = WARP_SIZE / THREADS_PER_ROW;
@@ -396,7 +403,7 @@ void topkGatingSoftmaxLauncherHelper(const float* input, const bool* finished, f
 {
     static constexpr std::size_t MAX_BYTES_PER_LDG = 16;
 
-    static constexpr int BYTES_PER_LDG = std::min(MAX_BYTES_PER_LDG, sizeof(float) * EXPERTS);
+    static constexpr int BYTES_PER_LDG = MIN(MAX_BYTES_PER_LDG, sizeof(float) * EXPERTS);
     using Constants = detail::TopkConstants<EXPERTS, BYTES_PER_LDG>;
     static constexpr int VPT = Constants::VPT;
     static constexpr int ROWS_PER_WARP = Constants::ROWS_PER_WARP;

setup.py

@@ -382,7 +382,7 @@ def _read_requirements(filename: str) -> List[str]:
 
 ext_modules = []
 
-if _is_cuda():
+if _is_cuda() or _is_hip():
     ext_modules.append(CMakeExtension(name="vllm._moe_C"))
 
 if not _is_neuron():

vllm/model_executor/layers/fused_moe/fused_moe.py

@@ -8,9 +8,9 @@
 import triton
 import triton.language as tl
 
+import vllm._moe_C as moe_kernels
 from vllm import _custom_ops as ops
 from vllm.logger import init_logger
-from vllm.utils import is_hip
 
 logger = init_logger(__name__)
 
@@ -319,34 +319,26 @@ def fused_topk(
 
     M, _ = hidden_states.shape
 
-    if is_hip():
-        # The MoE kernels are not yet supported on ROCm.
-        routing_weights = torch.softmax(gating_output,
-                                        dim=-1,
-                                        dtype=torch.float32)
-        topk_weights, topk_ids = torch.topk(routing_weights, topk, dim=-1)
-    else:
-        import vllm._moe_C as moe_kernels
-
-        topk_weights = torch.empty(M,
-                                   topk,
-                                   dtype=torch.float32,
-                                   device=hidden_states.device)
-        topk_ids = torch.empty(M,
+    topk_weights = torch.empty(M,
                                topk,
-                               dtype=torch.int32,
+                               dtype=torch.float32,
                                device=hidden_states.device)
-        token_expert_indicies = torch.empty(M,
-                                            topk,
-                                            dtype=torch.int32,
-                                            device=hidden_states.device)
-        moe_kernels.topk_softmax(
-            topk_weights,
-            topk_ids,
-            token_expert_indicies,
-            gating_output.float(),  # TODO(woosuk): Optimize this.
-        )
-        del token_expert_indicies  # Not used. Will be used in the future.
+    topk_ids = torch.empty(M,
+                           topk,
+                           dtype=torch.int32,
+                           device=hidden_states.device)
+    token_expert_indicies = torch.empty(M,
+                                        topk,
+                                        dtype=torch.int32,
+                                        device=hidden_states.device)
+    moe_kernels.topk_softmax(
+        topk_weights,
+        topk_ids,
+        token_expert_indicies,
+        gating_output.float(),  # TODO(woosuk): Optimize this.
+    )
+    del token_expert_indicies  # Not used. Will be used in the future.
+
     if renormalize:
         topk_weights = topk_weights / topk_weights.sum(dim=-1, keepdim=True)
     return topk_weights, topk_ids