Skip to content

Commit

Permalink
clean test cas3es
Browse files Browse the repository at this point in the history
  • Loading branch information
@Sibylau
Sibylau committed May 9, 2024
1 parent 15ddc00 commit 6bc4db7
Show file tree
Hide file tree
Showing 6 changed files with 22 additions and 317 deletions.
68 changes: 0 additions & 68 deletions test/UniSparse/KernelGen/HLS/codegen.mlir
View file

This file was deleted.

Binary file removed test/UniSparse/KernelGen/HLS/csr_spmv
View file
Binary file not shown.
Binary file removed test/UniSparse/KernelGen/HLS/csr_spmv.o
View file
Binary file not shown.
206 changes: 0 additions & 206 deletions test/UniSparse/KernelGen/HLS/device.cpp
View file

This file was deleted.

0 .../KernelGen/HLS/unisparse_host_device.mlir → ...arse/KernelGen/HLS/unisparse_csr_1PE.mlir
View file
File renamed without changes.
65 changes: 22 additions & 43 deletions ...arse/KernelGen/HLS/unisparse_lowered.mlir → ...arse/KernelGen/HLS/unisparse_dia_1PE.mlir
View file
Original file line number Diff line number Diff line change
Expand Up @@ -17,9 +17,9 @@
compressMap = #unisparse.compress<trim(0,1)>
}>

#CSR = #unisparse.encoding<{
crdMap = #unisparse.crd<(i,j)->(i,j)>,
compressMap = #unisparse.compress<fuse(0), trim(1,1)>
#DIA = #unisparse.encoding<{
crdMap = #unisparse.crd<(i,j)->(j-i,i)>,
compressMap = #unisparse.compress<trim(0)>
}>

#trait1 = {
Expand All @@ -36,13 +36,13 @@ module {
func.func private @rtclock() -> f64
func.func private @getTensorFilename(index) -> (!Filename)

// func.func @kernel_csr_spmv(%arg0: tensor<?x?xf32, #CSR>, %arg1: tensor<?xf32>, %argx: tensor<?xf32>) -> tensor<?xf32> {
// func.func @kernel_csr_spmv(%arg0: tensor<?x?xf32, #DIA>, %arg1: tensor<?xf32>, %argx: tensor<?xf32>) -> tensor<?xf32> {
// %0 = linalg.generic #trait1
// ins(%arg0, %arg1 : tensor<?x?xf32, #CSR>, tensor<?xf32>)
// ins(%arg0, %arg1 : tensor<?x?xf32, #DIA>, tensor<?xf32>)
// outs(%argx: tensor<?xf32>) {
// ^bb0(%a: f32, %x: f32, %o: f32):
// %2 = arith.mulf %a, %x : f32
// %3 = arith.addf %o, %2 : f32
// ^bb0(%a: f32, %b: f32, %x: f32):
// %2 = arith.mulf %a, %b : f32
// %3 = arith.addf %x, %2 : f32
// linalg.yield %3 : f32
// } -> tensor<?xf32>
// return %0 : tensor<?xf32>
Expand All @@ -60,7 +60,7 @@ module {
%t_start0 = call @rtclock() : () -> f64
%A_0 = unisparse.fromFile (%fileName) : !Filename to tensor<?x?xf32, #COO>
%c256 = tensor.dim %A_0, %c1 : tensor<?x?xf32, #COO>
%a = unisparse.convert (%A_0): tensor<?x?xf32, #COO> to tensor<?x?xf32, #CSR>
%a0 = unisparse.convert (%A_0): tensor<?x?xf32, #COO> to tensor<?x?xf32, #DIA>
%t_end0 = call @rtclock() : () -> f64
%t_0 = arith.subf %t_end0, %t_start0: f64
vector.print %t_0 : f64
Expand All @@ -70,54 +70,33 @@ module {
%ts_dim_i = arith.index_cast %c256 : index to i32
%ts_dim = arith.sitofp %ts_dim_i : i32 to f32
%elm = arith.divf %i1, %ts_dim : f32
%x = scf.for %i = %c0 to %c256 step %c1 iter_args(%t = %init_256_4) -> tensor<?xf32> {
// %k0 = arith.muli %i, %c1 : index
// %k1 = arith.index_cast %k0 : index to i32
// %k1 = arith.index_cast %i : index to i32
// %k = arith.sitofp %k1 : i32 to f32
%b = scf.for %i = %c0 to %c256 step %c1 iter_args(%t = %init_256_4) -> tensor<?xf32> {
%t3 = tensor.insert %elm into %t[%i] : tensor<?xf32>
scf.yield %t3 : tensor<?xf32>
}

%y_4_4 = bufferization.alloc_tensor(%c256) : tensor<?xf32>
%y = scf.for %i = %c0 to %c256 step %c1 iter_args(%t = %y_4_4) -> tensor<?xf32> {
%o0_4_4 = bufferization.alloc_tensor(%c256) : tensor<?xf32>
%o0 = scf.for %i = %c0 to %c256 step %c1 iter_args(%t = %o0_4_4) -> tensor<?xf32> {
%t3 = tensor.insert %i0 into %t[%i] : tensor<?xf32>
scf.yield %t3 : tensor<?xf32>
}

// %t_start4 = call @rtclock() : () -> f64
%0 = unisparse.device (%a, %x, %y) {target = "HLS"}:
(tensor<?x?xf32, #CSR>, tensor<?xf32>, tensor<?xf32>) to tensor<?xf32> {
// // read in
// %a_d0_ptr = unisparse.read_ptr (%a) : tensor<?x?xf32> -> tensor<?xf32>
// %a_d1_crd = unisparse.read_crd (%a) : tensor<?x?xf32> -> tensor<?xf32>
// %a_val = unisparse.read_val (%a) : tensor<?x?xf32> -> tensor<?xf32>
// %x_val = unisparse.read_val (%x) : tensor<?xf32> -> tensor<?xf32>
// // decode sparse matrix indices
// %a_d0_crd = unisparse.repeater (%a_d0_ptr) : tensor<?xf32> -> tensor<?xf32> // subject to change
// %a_ori_d0, %a_ori_d1 = unisparse.index_calc (%a_d0_crd, %a_d1_crd) : tensor<?xf32>, tensor<?xf32> -> tensor<?xf32>, tensor<?xf32>
// // compute
// %y_d0, %psum = unisparse.PE_mul (%a_ori_d0, %a_ori_d1, %a_val, %x_val) :
// tensor<?xf32>, tensor<?xf32>, tensor<?xf32>, tensor<?xf32> -> tensor<?xf32>, tensor<?xf32>
// %y = unisparse.PE_accum (%y_d0, %psum):
// tensor<?xf32>, tensor<?xf32> -> tensor<?xf32>
// // write out
// unisparse.write_val (%y) : tensor<?xf32>
// // generate top
// unisparse.top (%a, %x, %y) : tensor<?x?xf32, #CSR>, tensor<?xf32>, tensor<?xf32>
// // unisparse.terminator
}
// %0 = call @kernel_csr_spmv(%a, %x, %y) : (tensor<?x?xf32, #CSR>, tensor<?xf32>, tensor<?xf32>) -> tensor<?xf32>
// %t_end4 = call @rtclock() : () -> f64
%t_4 = arith.subf %t_end4, %t_start4: f64
vector.print %t_4 : f64
%0 = unisparse.device (%a0, %b, %o0) {target = "HLS"} {
%0 = linalg.generic #trait1
ins(%a0, %b : tensor<?x?xf32, #DIA>, tensor<?xf32>)
outs(%o0: tensor<?xf32>) {
^bb0(%a: f32, %b1: f32, %x: f32):
%2 = arith.mulf %a, %b1 : f32
%3 = arith.addf %x, %2 : f32
linalg.yield %3 : f32
} -> tensor<?xf32>
}: tensor<?x?xf32, #DIA>, tensor<?xf32>, tensor<?xf32> -> tensor<?xf32>
%v0 = vector.transfer_read %0[%c0], %i0: tensor<?xf32>, vector<4xf32>
vector.print %v0 : vector<4xf32>

//Release the resources
bufferization.dealloc_tensor %A_0 : tensor<?x?xf32, #COO>
// bufferization.dealloc_tensor %init_256_4 : tensor<?xf32>
// bufferization.dealloc_tensor %y_4_4 : tensor<?xf32>
return
}
}

0 comments on commit 6bc4db7

Please sign in to comment.