cory-master/BabelStream
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Add comment about blockIdx/workgroupIdx in Julia

Browse Source
This commit is contained in:
Tom Lin 2021-06-30 19:31:42 +01:00
parent 2e957d3f60
commit 7c1e04a42b
2 changed files with 12 additions and 12 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

12
JuliaStream.jl/src/AMDGPUStream.jl
View File

@ -64,7 +64,7 @@ end
function copy!(data::ROCData{T}) where {T} function copy!(data::ROCData{T}) where {T}
function kernel(a, c) function kernel(a, c)
i = (workgroupIdx().x - 1) * workgroupDim().x + workitemIdx().x i = (workgroupIdx().x - 1) * workgroupDim().x + workitemIdx().x # only workgroupIdx starts at 1
@inbounds c[i] = a[i] @inbounds c[i] = a[i]
return return
end end
@ -76,7 +76,7 @@ end
function mul!(data::ROCData{T}) where {T} function mul!(data::ROCData{T}) where {T}
function kernel(b, c, scalar) function kernel(b, c, scalar)
i = (workgroupIdx().x - 1) * workgroupDim().x + workitemIdx().x i = (workgroupIdx().x - 1) * workgroupDim().x + workitemIdx().x # only workgroupIdx starts at 1
@inbounds b[i] = scalar * c[i] @inbounds b[i] = scalar * c[i]
return return
end end
@ -88,7 +88,7 @@ end
function add!(data::ROCData{T}) where {T} function add!(data::ROCData{T}) where {T}
function kernel(a, b, c) function kernel(a, b, c)
i = (workgroupIdx().x - 1) * workgroupDim().x + workitemIdx().x i = (workgroupIdx().x - 1) * workgroupDim().x + workitemIdx().x # only workgroupIdx starts at 1
@inbounds c[i] = a[i] + b[i] @inbounds c[i] = a[i] + b[i]
return return
end end
@ -100,7 +100,7 @@ end
function triad!(data::ROCData{T}) where {T} function triad!(data::ROCData{T}) where {T}
function kernel(a, b, c, scalar) function kernel(a, b, c, scalar)
i = (workgroupIdx().x - 1) * workgroupDim().x + workitemIdx().x i = (workgroupIdx().x - 1) * workgroupDim().x + workitemIdx().x # only workgroupIdx starts at 1
@inbounds a[i] = b[i] + (scalar * c[i]) @inbounds a[i] = b[i] + (scalar * c[i])
return return
end end
@ -117,7 +117,7 @@ end
function nstream!(data::ROCData{T}) where {T} function nstream!(data::ROCData{T}) where {T}
function kernel(a, b, c, scalar) function kernel(a, b, c, scalar)
i = (workgroupIdx().x - 1) * workgroupDim().x + workitemIdx().x i = (workgroupIdx().x - 1) * workgroupDim().x + workitemIdx().x # only workgroupIdx starts at 1
@inbounds a[i] += b[i] + scalar * c[i] @inbounds a[i] += b[i] + scalar * c[i]
return return
end end
@ -139,7 +139,7 @@ function dot(data::ROCData{T}) where {T}
@inbounds tb_sum[local_i] = 0.0 @inbounds tb_sum[local_i] = 0.0
# do dot first # do dot first
i = (workgroupIdx().x - 1) * workgroupDim().x + workitemIdx().x i = (workgroupIdx().x - 1) * workgroupDim().x + workitemIdx().x # only workgroupIdx starts at 1
while i <= size while i <= size
@inbounds tb_sum[local_i] += a[i] * b[i] @inbounds tb_sum[local_i] += a[i] * b[i]
i += TBSize * DotBlocks # XXX don't use (workgroupDim().x * gridDimWG().x) here i += TBSize * DotBlocks # XXX don't use (workgroupDim().x * gridDimWG().x) here

12
JuliaStream.jl/src/CUDAStream.jl
View File

@ -51,7 +51,7 @@ end
function copy!(data::CuData{T}) where {T} function copy!(data::CuData{T}) where {T}
function kernel(a, c) function kernel(a, c)
i = (blockIdx().x - 1) * blockDim().x + threadIdx().x i = (blockIdx().x - 1) * blockDim().x + threadIdx().x # only blockIdx starts at 1
@inbounds c[i] = a[i] @inbounds c[i] = a[i]
return return
end end
@ -61,7 +61,7 @@ end
function mul!(data::CuData{T}) where {T} function mul!(data::CuData{T}) where {T}
function kernel(b, c, scalar) function kernel(b, c, scalar)
i = (blockIdx().x - 1) * blockDim().x + threadIdx().x i = (blockIdx().x - 1) * blockDim().x + threadIdx().x # only blockIdx starts at 1
@inbounds b[i] = scalar * c[i] @inbounds b[i] = scalar * c[i]
return return
end end
@ -71,7 +71,7 @@ end
function add!(data::CuData{T}) where {T} function add!(data::CuData{T}) where {T}
function kernel(a, b, c) function kernel(a, b, c)
i = (blockIdx().x - 1) * blockDim().x + threadIdx().x i = (blockIdx().x - 1) * blockDim().x + threadIdx().x # only blockIdx starts at 1
@inbounds c[i] = a[i] + b[i] @inbounds c[i] = a[i] + b[i]
return return
end end
@ -81,7 +81,7 @@ end
function triad!(data::CuData{T}) where {T} function triad!(data::CuData{T}) where {T}
function kernel(a, b, c, scalar) function kernel(a, b, c, scalar)
i = (blockIdx().x - 1) * blockDim().x + threadIdx().x i = (blockIdx().x - 1) * blockDim().x + threadIdx().x # only blockIdx starts at 1
@inbounds a[i] = b[i] + (scalar * c[i]) @inbounds a[i] = b[i] + (scalar * c[i])
return return
end end
@ -96,7 +96,7 @@ end
function nstream!(data::CuData{T}) where {T} function nstream!(data::CuData{T}) where {T}
function kernel(a, b, c, scalar) function kernel(a, b, c, scalar)
i = (blockIdx().x - 1) * blockDim().x + threadIdx().x i = (blockIdx().x - 1) * blockDim().x + threadIdx().x # only blockIdx starts at 1
@inbounds a[i] += b[i] + scalar * c[i] @inbounds a[i] += b[i] + scalar * c[i]
return return
end end
@ -117,7 +117,7 @@ function dot(data::CuData{T}) where {T}
@inbounds tb_sum[local_i] = 0.0 @inbounds tb_sum[local_i] = 0.0
# do dot first # do dot first
i = (blockIdx().x - 1) * blockDim().x + threadIdx().x i = (blockIdx().x - 1) * blockDim().x + threadIdx().x # only blockIdx starts at 1
while i <= size while i <= size
@inbounds tb_sum[local_i] += a[i] * b[i] @inbounds tb_sum[local_i] += a[i] * b[i]
i += blockDim().x * gridDim().x i += blockDim().x * gridDim().x