Coverage for src/flag_gems/runtime/backend/_metax/ops/min.py: 0%

1import logging 

2import math 

3from collections import namedtuple 

4 

5import torch 

6import triton 

7import triton.language as tl 

8 

9from flag_gems import runtime 

10from flag_gems.runtime import torch_device_fn 

11from flag_gems.utils import libentry, libtuner 

12from flag_gems.utils import triton_lang_extension as tle 

13from flag_gems.utils.limits import get_dtype_max 

14 

15logger = logging.getLogger("flag_gems." + __name__) 

16 

17 

18@libentry() 

19@triton.jit 

20def min_kernel_1( 

21 inp, 

22 mid, 

23 M, 

24 BLOCK_SIZE: tl.constexpr, 

25): 

26 pid = tle.program_id(0) 

27 offset = pid * BLOCK_SIZE + tl.arange(0, BLOCK_SIZE) 

28 inp_ptrs = inp + offset 

29 mask = offset < M 

30 max_value = get_dtype_max(inp.type.element_ty) 

31 inp_val = tl.load(inp_ptrs, mask=mask, other=max_value) 

32 min_val = tl.min(inp_val) 

33 mid_ptr = mid + pid 

34 tl.store(mid_ptr, min_val) 

35 

36 

37@libentry() 

38@triton.jit 

39def min_kernel_2(mid, out, mid_size, BLOCK_MID: tl.constexpr): 

40 offset = tl.arange(0, BLOCK_MID) 

41 mid_ptrs = mid + offset 

42 mask = offset < mid_size 

43 max_value = get_dtype_max(mid.type.element_ty) 

44 mid_val = tl.load(mid_ptrs, mask=mask, other=max_value) 

45 min_val = tl.min(mid_val) 

46 tl.store(out, min_val) 

47 

48 

49def heur_block_n(args): 

50 return triton.next_power_of_2(args["N"]) 

51 

52 

53@libentry() 

54@libtuner( 

55 configs=runtime.get_tuned_config("min"), 

56 key=[ 

57 "M", 

58 "N", 

59 ], 

60) 

61@triton.heuristics( 

62 { 

63 "BLOCK_N": heur_block_n, 

64 } 

65) 

66@triton.jit 

67def min_kernel( 

68 inp, 

69 out_value, 

70 out_index, 

71 M, 

72 N, 

73 K, 

74 BLOCK_M: tl.constexpr, 

75 BLOCK_N: tl.constexpr, 

76): 

77 # set offset 

78 pid_m = tle.program_id(0) 

79 pid_k = tle.program_id(1) 

80 m_offset = pid_m * BLOCK_M + tl.arange(0, BLOCK_M) 

81 

82 dtype = inp.type.element_ty 

83 # you just cannot create a function that return a tl.dtype in triton lang 

84 acc_type = tl.float32 if dtype is tl.bfloat16 else dtype 

85 max_value = get_dtype_max(dtype) 

86 min_values = tl.full([BLOCK_M], dtype=acc_type, value=max_value) 

87 argmin_values = tl.full([BLOCK_M], dtype=tl.int64, value=0) 

88 for start_n in range(0, N, BLOCK_N): 

89 n_offset = start_n + tl.arange(0, BLOCK_N) 

90 offset = m_offset[:, None] * N * K + n_offset[None, :] * K + pid_k 

91 mask = m_offset[:, None] < M and n_offset[None, :] < N 

92 inp_ptrs = inp + offset 

93 inp_vals = tl.load(inp_ptrs, mask=mask, other=max_value) 

94 local_min, local_argmin = tl.min(inp_vals, 1, return_indices=True) 

95 # if return indices is not supported, call a tl.argmax in addition 

96 # local_argmin = tl.argmin(inp_vals, 1) 

97 update = local_min < min_values 

98 min_values = tl.where(update, local_min, min_values) 

99 argmin_values = tl.where(update, start_n + local_argmin, argmin_values) 

100 

101 offset_index = m_offset * K + pid_k 

102 out_value_ptrs = out_value + offset_index 

103 out_index_ptrs = out_index + offset_index 

104 mask1 = m_offset < M 

105 tl.store(out_value_ptrs, min_values, mask=mask1) 

106 tl.store(out_index_ptrs, argmin_values, mask=mask1) 

107 

108 

109def min(inp): 

110 logger.debug("METAX GEMS MIN") 

111 M = inp.numel() 

112 block_size = triton.next_power_of_2(math.ceil(math.sqrt(M))) 

113 mid_size = triton.cdiv(M, block_size) 

114 block_mid = triton.next_power_of_2(mid_size) 

115 

116 dtype = inp.dtype 

117 mid = torch.empty((mid_size,), dtype=dtype, device=inp.device) 

118 out = torch.empty([], dtype=dtype, device=inp.device) 

119 

120 with torch_device_fn.device(inp.device): 

121 min_kernel_1[(mid_size, 1, 1)](inp, mid, M, block_size) 

122 min_kernel_2[(1, 1, 1)](mid, out, mid_size, block_mid) 

123 return out 

124 

125 

126def min_dim(inp, dim=None, keepdim=False): 

127 logger.debug("METAX GEMS MIN DIM") 

128 assert dim >= -inp.ndim and dim < inp.ndim, "Invalid dim" 

129 shape = inp.shape 

130 dim = dim % inp.ndim 

131 N = shape[dim] 

132 M = math.prod(shape[:dim]) 

133 K = inp.numel() // M // N 

134 

135 inp = inp.contiguous() 

136 

137 shape_list = list(shape) 

138 shape_list[dim] = 1 

139 out_value = torch.empty(shape_list, dtype=inp.dtype, device=inp.device) 

140 out_index = torch.empty(shape_list, dtype=torch.int64, device=inp.device) 

141 

142 if not keepdim: 

143 out_value = torch.squeeze(out_value, dim) 

144 out_index = torch.squeeze(out_index, dim) 

145 

146 grid = lambda meta: ( 

147 triton.cdiv(M, meta["BLOCK_M"]), 

148 K, 

149 ) 

150 with torch_device_fn.device(inp.device): 

151 min_kernel[grid](inp, out_value, out_index, M, N, K) 

152 Min_out = namedtuple("min", ["values", "indices"]) 

153 out = Min_out(values=out_value, indices=out_index) 

154 return out