# RUN: env SUPPORT_LIB=%mlir_cuda_runtime \ # RUN: %PYTHON %s | FileCheck %s # ===----------------------------------------------------------------------===// # Chapter 5 : Warp Specialized GEMM with Tensor Core # ===----------------------------------------------------------------------===// # # This program demonstrates a GEMM operation for `f32+=f16*f16`, utilizing the # Warp Specialized method with a tile size of 128x128x64. The code completely # parallelizes the two outermost loops into thread blocks. It launches two Warp # Groups (256 threads in total): one for the producer and the other for the consumer. # Each group takes a different control-flow. The producer thread group is responsible # for loading data into shared memory, while the consumer group executes the Tensor # Core GEMM operation and epilogue. # # for ti in range(M//128): # -> blockIdx.x # for tj in range(N//128): # -> blockIdx.y # with wg_producer: # for tk in range(K//64): # TMA_128x64_64x128... # with wg_consumer: # for tk in range(K//64): # MMA_128x128x64... # Epilogue.. # # This chapter demonstrates: # 2 WG (warpgroups) # Producer: # 2.1.1 Wait MMA Barrier # 2.1.1 Load TMA with TMA barrier # 2.1.1 Arrive TMA barrier with txcount # Consumer: # Loop # Wait TMA barrier # Performs Tensor Core GEMM 64x128x64 by warpgroup # Arrive MMA Barrier # Epilogue # Store fragmented registers to shared memory # Store shared memory to global # # ===----------------------------------------------------------------------===// from mlir import ir from mlir.dialects import gpu, scf, nvgpu, nvvm from mlir.extras import types as T from tools.nvdsl import * import numpy as np def partition_shape(): """ Calculate the partition shape based on the block IDs. It parallelizes the two outermost loops into thread blocks. for ti in range(M//128): # -> blockIdx.x for tj in range(N//128): # -> blockIdx.y D = 0 for tk in range(K//64): for i in range(128): for j in range(128): for k in range(64): FMA Returns: dimX (int): Dimension along the x-axis. dimY (int): Dimension along the y-axis. """ bidx = gpu.block_id(gpu.Dimension.x) bidy = gpu.block_id(gpu.Dimension.y) dimX = bidx * TILE_M dimY = bidy * TILE_N return dimX, dimY def tma_load( mbar_group: Mbarriers, a_tma: TMA, b_tma: TMA, slot, stage, num_stages, p=None, ): """ TMA loads two input matrices from global memory to shared memory. It performs the following operations: - tma.load a_shared_memory[off_x] at coordinate [x, z] (Loads 128x64) - tma.load b_shared_memory[off_y1] at coordinate [y, x] (Loads 64x64) - tma.load b_shared_memory[off_y2] at coordinate [y + 64, x] (Loads 64x64) mbarrier.arrive ta_count = 128x64x2x4 """ dimX, dimY = partition_shape() tidx = gpu.thread_id(gpu.Dimension.x) begin_b = num_stages * get_type_size(a_tma.tma_memref) size_tma_a = get_type_size(a_tma.tma_memref) size_tma_b = get_type_size(b_tma.tma_memref) ta_count = size_tma_a + (size_tma_b * 2) off_a = slot * size_tma_a off_b = (slot * size_tma_a) + begin_b off_b2 = off_b + size_tma_b a_elem_ty = a_tma.tma_memref.element_type b_elem_ty = b_tma.tma_memref.element_type a = get_dynamic_shared_memory(a_tma.tma_memref.shape, a_elem_ty, off_a) b1 = get_dynamic_shared_memory(b_tma.tma_memref.shape, b_elem_ty, off_b) b2 = get_dynamic_shared_memory(b_tma.tma_memref.shape, b_elem_ty, off_b2) mbar_group[slot].arrive(ta_count, predicate=p) p = (tidx % WARP_GROUP_SIZE) == 0 c1 = stage * 64 a_tma.load(a, mbar_group[slot], coords=[c1, dimX], predicate=p) b_tma.load(b1, mbar_group[slot], coords=[dimY, c1], predicate=p) b_tma.load(b2, mbar_group[slot], coords=[dimY + 64, c1], predicate=p) def initialize(a_tma: TMA, b_tma: TMA, num_stages): """ Initialize mbarriers and prefetch TMA descriptors. """ tidx = gpu.thread_id(gpu.Dimension.x) mbar_group_tma = Mbarriers(number_of_barriers=num_stages) mbar_group_mma = Mbarriers(number_of_barriers=num_stages) isThread0 = tidx == const(0) with ir.InsertionPoint(scf.IfOp(isThread0).then_block): for i in scf.for_(0, num_stages, 1): mbar_group_tma[i].init(1) mbar_group_mma[i].init(1) scf.yield_([]) a_tma.prefetch() b_tma.prefetch() scf.yield_([]) return mbar_group_tma, mbar_group_mma def switch_phase(stage, phase, num_stages): p = stage == (num_stages - 1) phase = arith.select( p, (phase ^ const(True, ty=T.bool())), phase, ) return phase def producer_loop( mbar_tma: Mbarriers, mbar_mma: Mbarriers, a_tma: TMA, b_tma: TMA, wg_me: Warpgroup, num_stages, ): phase = const(True, ty=T.bool()) for iv, phase in scf.for_(0, (K // TILE_K), 1, [phase]): stage = iv % num_stages # Wait MMA to be done mbar_mma[stage].try_wait(phase) # New phase for mbarrier phase = switch_phase(stage, phase, num_stages) # TMA Load tma_load(mbar_tma, a_tma, b_tma, stage, iv, num_stages, wg_me.is_wg_primary) scf.yield_([phase]) def consumer_loop( mbar_tma: Mbarriers, mbar_mma: Mbarriers, a_tma: TMA, b_tma: TMA, wg_me: Warpgroup, num_stages, ): begin_b = num_stages * get_type_size(a_tma.tma_memref) size_a = TILE_M * TILE_K * get_type_size(T.f16()) phase = const(False, ty=T.bool()) A = WGMMAMatrix(WGMMAType.Descriptor, [TILE_M, TILE_K], desc=a_tma) B = WGMMAMatrix(WGMMAType.Descriptor, [TILE_K, TILE_N], desc=b_tma) D = WGMMAMatrix(WGMMAType.Accumulator, shape=[TILE_M, TILE_N], ty=T.f32()) for_op = scf.ForOp(const(0), const(K // TILE_K), const(1), [D.acc_op, phase]) with ir.InsertionPoint(for_op.body): phase = for_op.inner_iter_args[1] iv = for_op.induction_variable stage = iv % num_stages # Wait TMA for current stage mbar_tma[stage].try_wait(phase) # Find shared memory slot offset_a = stage * size_a offset_b = offset_a + begin_b a_smem = get_dynamic_shared_memory([TILE_M, TILE_K], T.f16(), offset_a) b_smem = get_dynamic_shared_memory([TILE_K, TILE_N], T.f16(), offset_b) # Iterate input matrices, update accumulator A.update_smem(a_smem) B.update_smem(b_smem) D.update_accumulator(for_op.inner_iter_args[0]) # Matrix Multiply D += A @ B # MMA Barrier Arrive p_arrive = (iv > 0) & wg_me.is_wg_primary with ir.InsertionPoint(scf.IfOp(p_arrive).then_block): barId = arith.select((stage == 0), const(num_stages - 1), (stage - 1)) mbar_mma[barId].arrive() scf.yield_([]) phase = switch_phase(stage, phase, num_stages) scf.yield_([D.acc_op, phase]) nvvm.WgmmaWaitGroupSyncOp(0) D.update_accumulator(for_op.results[0]) return D def epilogue(D: WGMMAMatrix, d_dev): """ Epilogue of the GEMM kernel. It stores the fragmented registers to global memory. MatrixAccumulator D # Fragmented results store D -> Shared Memory # Store Shared Memory Shared Memory -> Z[dimX][dimY] # Store Shared Memory to Global Memory """ tidx = gpu.thread_id(gpu.Dimension.x) dimX, dimY = partition_shape() # s = tidx - WARP_GROUP_SIZE # debug_print("[Epilogue] store to global memory @ s={}", s) d_smem = get_dynamic_shared_memory([TILE_M, TILE_N], T.f32()) d_gmem = memref.subview(d_dev, [dimX, dimY], [TILE_M, TILE_N], [1, 1]) # Store (registers -> shared memory) D.store_accumulator(d_smem) gpu.barrier() # Store (shared memory --> global memory) for i in scf.for_(0, TILE_M, 1): val = memref.load(d_smem, [i, tidx]) memref.store(val, d_gmem, [i, tidx]) scf.yield_([]) @NVDSL.mlir_func def gemm_warp_specialized(a, b, d, num_stages): token_ty = gpu.AsyncTokenType.get() t1 = gpu.wait(token_ty, []) a_dev, t2 = gpu.alloc(a.type, token_ty, [t1], [], []) b_dev, t3 = gpu.alloc(b.type, token_ty, [t2], [], []) d_dev, t4 = gpu.alloc(d.type, token_ty, [t3], [], []) t5 = gpu.memcpy(token_ty, [t4], a_dev, a) t6 = gpu.memcpy(token_ty, [t5], b_dev, b) t7 = gpu.wait(token_ty, [t6]) sw = nvgpu.TensorMapSwizzleKind.SWIZZLE_128B a_tma = TMA([128, 64], a.type, swizzle=sw) b_tma = TMA([64, 64], b.type, swizzle=sw) a_tma.create_descriptor(a_dev) b_tma.create_descriptor(b_dev) grid = [(M // TILE_M), (N // TILE_N), 1] block = [256, 1, 1] size_a = get_type_size(a.type.element_type) * TILE_M * TILE_K size_b = get_type_size(b.type.element_type) * TILE_N * TILE_K smem_size_in_bytes = (size_a + size_b) * num_stages @NVDSL.mlir_gpu_launch(grid=grid, block=block, smem=smem_size_in_bytes) def gemm_warp_specialized_kernel(): # Init Warpgroups wg_producer = Warpgroup(primary_thread=128, register_size=40) wg_consumer = Warpgroup(primary_thread=0, register_size=232) # Initialize mbarriers and prefetch TMA descriptors mbar_mma, mbar_tma = initialize(a_tma, b_tma, num_stages) # Producer performs TMA with wg_producer: producer_loop(mbar_tma, mbar_mma, a_tma, b_tma, wg_producer, num_stages) # Consumer performs MMA/Tensor Core with wg_consumer: D = consumer_loop(mbar_tma, mbar_mma, a_tma, b_tma, wg_consumer, num_stages) epilogue(D, d_dev) gemm_warp_specialized_kernel() t8 = gpu.memcpy(token_ty, [t7], d, d_dev) gpu.wait(None, [t8]) # Python pass arguments to MLIR N = 256 M = 512 K = 1024 TILE_M = 128 TILE_N = 128 TILE_K = 64 a = np.random.randn(M, K).astype(np.float16) b = np.random.randn(K, N).astype(np.float16) d = np.zeros((M, N), np.float32) gemm_warp_specialized(a, b, d, num_stages=7) # Verify MLIR with reference computation ref_d = a.astype(np.float16) @ b.astype(np.float16) np.testing.assert_allclose(d, ref_d, rtol=5e-03, atol=1e-01) print("PASS") # CHECK-NOT: Mismatched elements