Some generated cuda kernel's input's shape is 0

[Vincent Wendong XU]

Hello, I noticed that some generated cuda kernel's input's shape is 0. Here is the hidet python model-define code:

I wonder know it would happen and is it a bug? Thanks.

import torch
import torch._dynamo
from torch import nn
import hidet 
import math


class LayerNorm(nn.Module):
    def __init__(self, hidden_size, eps=1e-12):
        """Construct a layernorm module in the TF style (epsilon inside the square root).
        """
        super(LayerNorm, self).__init__()
        self.weight = nn.Parameter(torch.ones(hidden_size))
        self.bias = nn.Parameter(torch.zeros(hidden_size))
        self.variance_epsilon = eps

    def forward(self, x):
        u = x.mean(-1, keepdim=True)
        s = (x - u).pow(2).mean(-1, keepdim=True)
        x = (x - u) / torch.sqrt(s + self.variance_epsilon)
        return self.weight * x + self.bias
        
class SelfAttention(nn.Module):
    def __init__(self, num_attention_heads, input_size, hidden_size, attention_probs_dropout_prob, hidden_dropout_prob):
        super(SelfAttention, self).__init__()
        if hidden_size % num_attention_heads != 0:
            raise ValueError(
                "The hidden size (%d) is not a multiple of the number of attention "
                "heads (%d)" % (hidden_size, num_attention_heads))
        self.num_attention_heads = num_attention_heads
        self.attention_head_size = int(hidden_size / num_attention_heads)
        self.all_head_size = hidden_size

        self.query = nn.Linear(input_size, self.all_head_size)
        self.key = nn.Linear(input_size, self.all_head_size)
        self.value = nn.Linear(input_size, self.all_head_size)

        self.attn_dropout = nn.Dropout(attention_probs_dropout_prob)

        self.dense = nn.Linear(hidden_size, hidden_size)
        self.LayerNorm = LayerNorm(hidden_size, eps=1e-12)
        self.out_dropout = nn.Dropout(hidden_dropout_prob)

    def transpose_for_scores(self, x):
        new_x_shape = x.size()[:-1] + (self.num_attention_heads, self.attention_head_size)
        x = x.view(*new_x_shape)
        return x.permute(0, 2, 1, 3)

    def forward(self, input_tensor):
        mixed_query_layer = self.query(input_tensor)
        mixed_key_layer = self.key(input_tensor)
        mixed_value_layer = self.value(input_tensor)

        query_layer = self.transpose_for_scores(mixed_query_layer)
        key_layer = self.transpose_for_scores(mixed_key_layer)
        value_layer = self.transpose_for_scores(mixed_value_layer)

        # Take the dot product between "query" and "key" to get the raw attention scores.
        attention_scores = torch.matmul(query_layer, key_layer.transpose(-1, -2))

        attention_scores = attention_scores / math.sqrt(self.attention_head_size)
        # Apply the attention mask is (precomputed for all layers in BertModel forward() function)
        # [batch_size heads seq_len seq_len] scores
        # [batch_size 1 1 seq_len]

        # attention_scores = attention_scores + attention_mask

        # Normalize the attention scores to probabilities.
        attention_probs = nn.Softmax(dim=-1)(attention_scores)
        # This is actually dropping out entire tokens to attend to, which might
        # seem a bit unusual, but is taken from the original Transformer paper.
        # Fixme
        attention_probs = self.attn_dropout(attention_probs)
        context_layer = torch.matmul(attention_probs, value_layer)
        context_layer = context_layer.permute(0, 2, 1, 3).contiguous()
        new_context_layer_shape = context_layer.size()[:-2] + (self.all_head_size,)
        context_layer = context_layer.view(*new_context_layer_shape)
        hidden_states = self.dense(context_layer)
        hidden_states = self.out_dropout(hidden_states)
        hidden_states = self.LayerNorm(hidden_states + input_tensor)

        return hidden_states


hidet.option.cache_dir('./outs/cache')
model = SelfAttention(num_attention_heads = 12, input_size = 768, hidden_size = 768, attention_probs_dropout_prob = 0.5, hidden_dropout_prob = 0.5).cuda().eval()
x = torch.rand(1, 128, 768).cuda()
# print(model)
model_opt = torch.compile(model, backend='hidet')  
y = model_opt(x)

Here is the 12nd kernel meta.json:

{
  "name": "fused_subtract_pow",
  "symbols": [],
  "inputs": [
    {
      "device": "cuda",
      "dtype": "float32",
      "shape": []
    },
    {
      "device": "cuda",
      "dtype": "float32",
      "shape": [
        1,
        128,
        768
      ]
    },
    {
      "device": "cuda",
      "dtype": "float32",
      "shape": [
        1,
        128,
        1
      ]
    }
  ],
  "outputs": [
    {
      "device": "cuda",
      "dtype": "float32",
      "shape": [
        1,
        128,
        768
      ]
    }
  ],
  "target": "cuda",
  "num_candidates": 1,
  "hidet_version": "0.3.1.dev"
}

Here is the generated kernel:

#include <stdint.h>
#include <hidet/runtime/symbols.h>
#include <hidet/runtime/memory_planner.h>
#include <hidet/runtime/cpu/context.h>
#include <hidet/runtime/cuda/complex.h>
#include <hidet/runtime/cuda/context.h>
#include <hidet/runtime/logging.h>


static __global__ void __launch_bounds__(512) hidet_fused_compute_z(float * __restrict__ x, float * __restrict__ y, float * __restrict__ y_1, float * __restrict__ z) {
  z[((((((int)blockIdx.x * 512) + (int)threadIdx.x) / 768) * 768) + ((((int)blockIdx.x * 512) + (int)threadIdx.x) % 768))] = powf((x[((((((int)blockIdx.x * 512) + (int)threadIdx.x) / 768) * 768) + ((((int)blockIdx.x * 512) + (int)threadIdx.x) % 768))] - y[((((int)blockIdx.x * 512) + (int)threadIdx.x) / 768)]), y_1[0]);
}

DLL void hidet_get_input_shape(int32_t idx, int32_t * __restrict__ dims) {
  if (idx == 0) {
  } 
  if (idx == 1) {
    dims[0] = 1;
    dims[1] = 128;
    dims[2] = 768;
  } 
  if (idx == 2) {
    dims[0] = 1;
    dims[1] = 128;
    dims[2] = 1;
  } 
}

DLL void hidet_get_output_shape(int32_t idx, int32_t * __restrict__ dims) {
  if (idx == 0) {
    dims[0] = 1;
    dims[1] = 128;
    dims[2] = 768;
  } 
}

DLL void hidet_launch_0(float * __restrict__ y, float * __restrict__ x, float * __restrict__ y_1, float * __restrict__ z) {
  hidet_fused_compute_z<<<dim3(192, 1, 1), dim3(512, 1, 1), 0, (cudaStream_t)get_cuda_stream()>>>(x, y_1, y, z);
  {cudaError_t err = cudaGetLastError(); if (err != cudaSuccess) LOG(ERROR) << "CUDA error: " << cudaGetErrorString(err) << "\n";}
}