Merge remote-tracking branch 'nccl/master' into develop

[ROCm/rccl commit: 3f94267f21]

projects/rccl/src/device/Makefile

@@ -36,9 +36,8 @@ define COMPILE
  $(call COMPILE$(or $3,$(suffix $2)),$1,$2)
 endef
 
-ifeq ($(shell echo "$$((1000*$(CUDA_MAJOR) + 10*$(CUDA_MINOR) >= 12080))"),1)
-	NVCC_GENCODE_LDMC_FP8 = -gencode=arch=compute_100a,code=sm_100a \
-                     -gencode=arch=compute_120a,code=sm_120a
+ifeq ($(shell echo "$$((1000*$(CUDA_MAJOR) + 10*$(CUDA_MINOR) >= 12090))"),1)
+	NVCC_GENCODE_LDMC_FP8 = -gencode=arch=compute_100f,code=sm_100f
 else ifeq ($(shell echo "$$((1000*$(CUDA_MAJOR) + 10*$(CUDA_MINOR) >= 12070))"),1)
   NVCC_GENCODE_LDMC_FP8 = -gencode=arch=compute_100a,code=sm_100a
 else

projects/rccl/src/device/common.h

@@ -636,6 +636,7 @@ __device__ __forceinline__ void ncclKernelMain(struct ncclDevKernelArgs const* a
     __syncthreads();
     if (COLLTRACE && tid%WARP_SIZE == 0) traceKernelLaunch(ncclCollTraceCollLaunchType, batchIx);
   }
+  profiler(FINI);
   if (COLLTRACE && tid%WARP_SIZE == 0) traceKernelEnd(ncclCollTraceKernelEndType);
 
 #ifdef ENABLE_PROFILING

projects/rccl/src/device/reduce_kernel.h

@@ -1073,7 +1073,7 @@ struct Apply_LoadMultimem {
   #endif
 
   #if defined(RCCL_BFLOAT16)
-    #if NCCL_CUDA_ARCH_FAMILY_SPECIFIC == 1000 || NCCL_CUDA_ARCH_FAMILY_SPECIFIC == 1010 || NCCL_CUDA_ARCH_SPECIFIC == 1200 || NCCL_CUDA_ARCH_SPECIFIC == 1210
+    #if NCCL_CUDA_ARCH_SPECIFIC == 1000 || NCCL_CUDA_ARCH_SPECIFIC == 1010 || NCCL_CUDA_ARCH_FAMILY_SPECIFIC == 1000 || NCCL_CUDA_ARCH_FAMILY_SPECIFIC == 1010 || NCCL_CUDA_ARCH_SPECIFIC == 1200 || NCCL_CUDA_ARCH_SPECIFIC == 1210
       DEFINE_Apply_LoadMultimem_sum_v4_and_xparts(__nv_fp8_e4m3, e4m3x4, 4)
       DEFINE_Apply_LoadMultimem_minmax_v4_and_xparts(__nv_fp8_e4m3, e4m3x4, 4)
       DEFINE_Apply_LoadMultimem_sum_v4_and_xparts(__nv_fp8_e5m2, e5m2x4, 4)

projects/rccl/src/device/symmetric/generate.py

@@ -108,7 +108,7 @@ def required_cuda(k):
       if k.algo in ldmc_algos:
         cudart = 12070
         arch = None
-        specific_sms = [100, 120]
+        specific_sms = ["100a", "101a", "100f", "101f", "120a", "121a"]
   return (cudart, arch, specific_sms)
 
 ################################################################################
@@ -145,7 +145,7 @@ def kernel_conds(k):
   if not specific_sms:
     arch_cond = "__CUDA_ARCH__ >= %d"%arch
   else:
-    arch_cond = " || ".join(["0"] + ["NCCL_CUDA_ARCH_SPECIFIC==%d"%(10*sm) for sm in specific_sms])
+    arch_cond = " || ".join(["0"] + ["NCCL_CUDA_ARCH_%sSPECIFIC==%d"%("FAMILY_" if sm[-1] == "f" else "", 10*int(sm.replace('a', '').replace('f', ''))) for sm in specific_sms])
   return cudart_cond, arch_cond
 
 def instantiate(k):