A 32x32 shared-memory tile read with swapped indices

This tile-based transpose loads a 32x32 tile into shared memory, then writes it out transposed. Shared memory has 32 banks, each 4 bytes wide, and consecutive 4-byte words map to consecutive banks (bank = address / 4 mod 32). During the WRITE-out phase, warp lane threadIdx.x reads tile[threadIdx.x][threadIdx.y] (note the swapped indices). What happens on that read, for a fixed threadIdx.y across the warp's 32 threadIdx.x values?

kernel.cu
__shared__ float tile[32][32];

__global__ void transpose(float* out, const float* in, int width) {
    int x = blockIdx.x * 32 + threadIdx.x;
    int y = blockIdx.y * 32 + threadIdx.y;
    tile[threadIdx.y][threadIdx.x] = in[y * width + x];
    __syncthreads();

    int outX = blockIdx.y * 32 + threadIdx.x;
    int outY = blockIdx.x * 32 + threadIdx.y;
    out[outY * width + outX] = tile[threadIdx.x][threadIdx.y];  // swapped indices
}

7 of 46 in GPU / CUDA