Add Fused Multi-Head Attention example

[AntonOresten]

See also #15

Seems to fall slightly short of my NNop / ONIONop baseline (no WMMA), although I haven't compared it to the Python version. On my GPU, it compiles and runs fastest with tile_n=32 and tile_m=32:

julia> begin
           T = Float32
           D, QL, KL, H, B = 64, 4096, 4096, 4, 4
           q = CUDA.randn(T, D, QL, H, B)
           k = CUDA.randn(T, D, KL, H, B)  
           v = CUDA.randn(T, D, KL, H, B)
       end;

julia> @b CUDA.@sync ONIONop.flash_attention(q, k, v, causal=false)
9.559 ms (339 allocs: 7.875 KiB)

julia> @b CUDA.@sync cutile_fmha(q, k, v, causal=false, tile_m=32, tile_n=32)
11.058 ms (540 allocs: 23.109 KiB)

Notably, cutile-python has a latency argument for ct.load, as well as num_ctas and occupancy arguments for the kernel, which might affect performance. The python version also does a kernel config autotune by searching a space of hand-picked configurations.

Currently only tested on Float32.

Another thing that might be important for correctness or covering edge cases is exposing flush_to_zero? Used in e.g. exp2.

[AntonOresten]

Seeing some weird erroring when branching:

This works:

        qk = if !EVEN_K[] && j >= mask_start
            offs_n = ((j-Int32(1)) * TILE_N[]) .+ offs_n_tile
            mask = ct.full((TILE_N[], TILE_M[]), true, Bool)
            mask = mask .& (offs_n .<= k_seqlen)
            mask = ct.where(mask, ct.zeros((TILE_N[], TILE_M[],), Float32), ct.full((TILE_N[], TILE_M[],), -Inf32, Float32))
            qk .+ mask
        else
            qk
        end

but this doesn't:

        if !EVEN_K[] && j >= mask_start
            offs_n = ((j-Int32(1)) * TILE_N[]) .+ offs_n_tile
            mask = ct.full((TILE_N[], TILE_M[]), true, Bool)
            mask = mask .& (offs_n .<= k_seqlen)
            mask = ct.where(mask, ct.zeros((TILE_N[], TILE_M[],), Float32), ct.full((TILE_N[], TILE_M[],), -Inf32, Float32))
            qk = qk .+ mask
        end

nor does this:

        qk = if !EVEN_K[] && j >= mask_start
            offs_n = ((j-Int32(1)) * TILE_N[]) .+ offs_n_tile
            mask = ct.full((TILE_N[], TILE_M[]), true, Bool)
            if !EVEN_K[]
                mask .& (offs_n .<= k_seqlen)
            end
            mask = ct.where(mask, ct.zeros((TILE_N[], TILE_M[],), Float32), ct.full((TILE_N[], TILE_M[],), -Inf32, Float32))
            qk .+ mask
        else
            qk
        end

In the second and third block, I get "ERROR: SSAValue %___ not found in context"

after removing the second condition, I can suddenly have a nested if block, and I don't need the outer else block:

        if !EVEN_K[]
            offs_n = ((j-Int32(1)) * TILE_N[]) .+ offs_n_tile
            mask = ct.full((TILE_N[], TILE_M[]), true, Bool)
            if !EVEN_K[]
                mask = mask .& (offs_n .<= k_seqlen)
            end
            mask = ct.where(mask, ct.zeros((TILE_N[], TILE_M[],), Float32), ct.full((TILE_N[], TILE_M[],), -Inf32, Float32))
            qk = qk .+ mask
        end

Does the if block need to depend on compile time constants?

I'd need this to make the padding and causal mask properly.

[maleadt]

In the second and third block, I get "ERROR: SSAValue %___ not found in context"

That's an IRStructurizer error. Can you provide an MWE?