This is an archive of the discontinued LLVM Phabricator instance.

Differential D84699

[mlir] [VectorOps] [integration_test] Sparse matrix times vector (jagged SAXPY version)
ClosedPublic

Authored by aartbik on Jul 27 2020, 1:17 PM.

Details

Reviewers
nicolasvasilache
bondhugula
ftynse
arpith-jacob
reidtatge
mehdi_amini
Commits
rG4f92ad508fcc: [mlir] [VectorOps] [integration_test] Sparse matrix times vector (jagged SAXPY…
Summary

Transposed jagged diagonal storage yields longer vector lengths. Also, in
contrast with naive SAXPY (one gather/scatter), this only performs one gather.

Diff Detail

Revision Contents

PathSize
mlir/
integration_test/
Dialect/
Vector/
CPU/
232 lines

Diff 281715

mlir/integration_test/Dialect/Vector/CPU/test-sparse-saxpy-jagged-matvec.mlir

  • This file was added.
// RUN: mlir-opt %s -convert-scf-to-std -convert-vector-to-llvm -convert-std-to-llvm | \
// RUN: mlir-cpu-runner -e entry -entry-point-result=void \
// RUN: -shared-libs=%mlir_integration_test_dir/libmlir_c_runner_utils%shlibext | \
// RUN: FileCheck %s
// Illustrates an 8x8 Sparse Matrix x Vector implemented with only operations
// of the vector dialect (and some std/scf). Essentially, this example performs
// the following multiplication:
//
// 0 1 2 3 4 5 6 7
// +------------------------+
// 0 | 1 0 2 0 0 1 0 1 | | 1 | | 21 |
// 1 | 1 8 0 0 3 0 1 0 | | 2 | | 39 |
// 2 | 0 0 1 0 0 2 6 2 | | 3 | | 73 |
// 3 | 0 3 0 1 0 1 0 1 | x | 4 | = | 24 |
// 4 | 5 0 0 1 1 1 0 0 | | 5 | | 20 |
// 5 | 0 3 0 0 2 1 2 0 | | 6 | | 36 |
// 6 | 4 0 7 0 1 0 1 0 | | 7 | | 37 |
// 7 | 0 3 0 2 0 0 1 1 | | 8 | | 29 |
// +------------------------+
//
// The sparse storage scheme used is an extended column scheme (also referred
// to as jagged diagonal, which is essentially a vector friendly variant of
// the general sparse row-wise scheme (also called compressed row storage),
// using fixed length vectors and no explicit pointer indexing into the
// value array to find the rows.
//
// The extended column storage for the matrix shown above is as follows.
//
// VALUE INDEX
// +---------+ +---------+
// 0 | 1 2 1 1 | | 0 2 5 7 |
// 1 | 1 8 3 1 | | 0 1 4 6 |
// 2 | 1 2 6 2 | | 2 5 6 7 |
// 3 | 3 1 1 1 | | 1 3 5 7 |
// 4 | 5 1 1 1 | | 0 3 4 5 |
// 5 | 3 2 1 2 | | 1 4 5 6 |
// 6 | 4 7 1 1 | | 0 2 4 6 |
// 7 | 3 2 1 1 | | 1 3 6 7 |
// +---------+ +---------+
//
// This example illustrates an effective SAXPY version that operates
// on the transposed jagged diagonal storage to obtain higher vector
// lengths. Another example in this directory illustrates a DOT
// version of the operation.
func @spmv8x8(%AVAL: memref<4xvector<8xf32>>,
%AIDX: memref<4xvector<8xi32>>,
%X: memref<?xf32>, %B: memref<1xvector<8xf32>>) {
%c0 = constant 0 : index
%c1 = constant 1 : index
%cn = constant 4 : index
%mask = vector.constant_mask [8] : vector<8xi1>
%b = load %B[%c0] : memref<1xvector<8xf32>>
%b_out = scf.for %k = %c0 to %cn step %c1 iter_args(%b_iter = %b) -> (vector<8xf32>) {
%aval = load %AVAL[%k] : memref<4xvector<8xf32>>
%aidx = load %AIDX[%k] : memref<4xvector<8xi32>>
%0 = vector.gather %X, %aidx, %mask : (memref<?xf32>, vector<8xi32>, vector<8xi1>) -> vector<8xf32>
%b_new = vector.fma %aval, %0, %b_iter : vector<8xf32>
scf.yield %b_new : vector<8xf32>
}
store %b_out, %B[%c0] : memref<1xvector<8xf32>>
return
}
func @entry() {
%c0 = constant 0 : index
%c1 = constant 1 : index
%c2 = constant 2 : index
%c3 = constant 3 : index
%c4 = constant 4 : index
%c5 = constant 5 : index
%c6 = constant 6 : index
%c7 = constant 7 : index
%c8 = constant 8 : index
%f0 = constant 0.0 : f32
%f1 = constant 1.0 : f32
%f2 = constant 2.0 : f32
%f3 = constant 3.0 : f32
%f4 = constant 4.0 : f32
%f5 = constant 5.0 : f32
%f6 = constant 6.0 : f32
%f7 = constant 7.0 : f32
%f8 = constant 8.0 : f32
%i0 = constant 0 : i32
%i1 = constant 1 : i32
%i2 = constant 2 : i32
%i3 = constant 3 : i32
%i4 = constant 4 : i32
%i5 = constant 5 : i32
%i6 = constant 6 : i32
%i7 = constant 7 : i32
//
// Allocate.
//
%AVAL = alloc() {alignment = 64} : memref<4xvector<8xf32>>
%AIDX = alloc() {alignment = 64} : memref<4xvector<8xi32>>
%X = alloc(%c8) {alignment = 64} : memref<?xf32>
%B = alloc() {alignment = 64} : memref<1xvector<8xf32>>
//
// Initialize.
//
%vf1 = vector.broadcast %f1 : f32 to vector<8xf32>
%0 = vector.insert %f3, %vf1[3] : f32 into vector<8xf32>
%1 = vector.insert %f5, %0[4] : f32 into vector<8xf32>
%2 = vector.insert %f3, %1[5] : f32 into vector<8xf32>
%3 = vector.insert %f4, %2[6] : f32 into vector<8xf32>
%4 = vector.insert %f3, %3[7] : f32 into vector<8xf32>
store %4, %AVAL[%c0] : memref<4xvector<8xf32>>
%5 = vector.insert %f2, %vf1[0] : f32 into vector<8xf32>
%6 = vector.insert %f8, %5[1] : f32 into vector<8xf32>
%7 = vector.insert %f2, %6[2] : f32 into vector<8xf32>
%8 = vector.insert %f2, %7[5] : f32 into vector<8xf32>
%9 = vector.insert %f7, %8[6] : f32 into vector<8xf32>
%10 = vector.insert %f2, %9[7] : f32 into vector<8xf32>
store %10, %AVAL[%c1] : memref<4xvector<8xf32>>
%11 = vector.insert %f3, %vf1[1] : f32 into vector<8xf32>
%12 = vector.insert %f6, %11[2] : f32 into vector<8xf32>
store %12, %AVAL[%c2] : memref<4xvector<8xf32>>
%13 = vector.insert %f2, %vf1[2] : f32 into vector<8xf32>
%14 = vector.insert %f2, %13[5] : f32 into vector<8xf32>
store %14, %AVAL[%c3] : memref<4xvector<8xf32>>
%vi0 = vector.broadcast %i0 : i32 to vector<8xi32>
%20 = vector.insert %i2, %vi0[2] : i32 into vector<8xi32>
%21 = vector.insert %i1, %20[3] : i32 into vector<8xi32>
%22 = vector.insert %i1, %21[5] : i32 into vector<8xi32>
%23 = vector.insert %i1, %22[7] : i32 into vector<8xi32>
store %23, %AIDX[%c0] : memref<4xvector<8xi32>>
%24 = vector.insert %i2, %vi0[0] : i32 into vector<8xi32>
%25 = vector.insert %i1, %24[1] : i32 into vector<8xi32>
%26 = vector.insert %i5, %25[2] : i32 into vector<8xi32>
%27 = vector.insert %i3, %26[3] : i32 into vector<8xi32>
%28 = vector.insert %i3, %27[4] : i32 into vector<8xi32>
%29 = vector.insert %i4, %28[5] : i32 into vector<8xi32>
%30 = vector.insert %i2, %29[6] : i32 into vector<8xi32>
%31 = vector.insert %i3, %30[7] : i32 into vector<8xi32>
store %31, %AIDX[%c1] : memref<4xvector<8xi32>>
%32 = vector.insert %i5, %vi0[0] : i32 into vector<8xi32>
%33 = vector.insert %i4, %32[1] : i32 into vector<8xi32>
%34 = vector.insert %i6, %33[2] : i32 into vector<8xi32>
%35 = vector.insert %i5, %34[3] : i32 into vector<8xi32>
%36 = vector.insert %i4, %35[4] : i32 into vector<8xi32>
%37 = vector.insert %i5, %36[5] : i32 into vector<8xi32>
%38 = vector.insert %i4, %37[6] : i32 into vector<8xi32>
%39 = vector.insert %i6, %38[7] : i32 into vector<8xi32>
store %39, %AIDX[%c2] : memref<4xvector<8xi32>>
%40 = vector.insert %i7, %vi0[0] : i32 into vector<8xi32>
%41 = vector.insert %i6, %40[1] : i32 into vector<8xi32>
%42 = vector.insert %i7, %41[2] : i32 into vector<8xi32>
%43 = vector.insert %i7, %42[3] : i32 into vector<8xi32>
%44 = vector.insert %i5, %43[4] : i32 into vector<8xi32>
%45 = vector.insert %i6, %44[5] : i32 into vector<8xi32>
%46 = vector.insert %i6, %45[6] : i32 into vector<8xi32>
%47 = vector.insert %i7, %46[7] : i32 into vector<8xi32>
store %47, %AIDX[%c3] : memref<4xvector<8xi32>>
%vf0 = vector.broadcast %f0 : f32 to vector<8xf32>
store %vf0, %B[%c0] : memref<1xvector<8xf32>>
scf.for %i = %c0 to %c8 step %c1 {
%ix = addi %i, %c1 : index
%kx = index_cast %ix : index to i32
%fx = sitofp %kx : i32 to f32
store %fx, %X[%i] : memref<?xf32>
}
//
// Multiply.
//
call @spmv8x8(%AVAL, %AIDX, %X, %B) : (memref<4xvector<8xf32>>,
memref<4xvector<8xi32>>,
memref<?xf32>,
memref<1xvector<8xf32>>) -> ()
//
// Print and verify.
//
scf.for %i = %c0 to %c4 step %c1 {
%aval = load %AVAL[%i] : memref<4xvector<8xf32>>
vector.print %aval : vector<8xf32>
}
scf.for %i = %c0 to %c4 step %c1 {
%aidx = load %AIDX[%i] : memref<4xvector<8xi32>>
vector.print %aidx : vector<8xi32>
}
%ldb = load %B[%c0] : memref<1xvector<8xf32>>
vector.print %ldb : vector<8xf32>
//
// CHECK: ( 1, 1, 1, 3, 5, 3, 4, 3 )
// CHECK-NEXT: ( 2, 8, 2, 1, 1, 2, 7, 2 )
// CHECK-NEXT: ( 1, 3, 6, 1, 1, 1, 1, 1 )
// CHECK-NEXT: ( 1, 1, 2, 1, 1, 2, 1, 1 )
//
// CHECK-NEXT: ( 0, 0, 2, 1, 0, 1, 0, 1 )
// CHECK-NEXT: ( 2, 1, 5, 3, 3, 4, 2, 3 )
// CHECK-NEXT: ( 5, 4, 6, 5, 4, 5, 4, 6 )
// CHECK-NEXT: ( 7, 6, 7, 7, 5, 6, 6, 7 )
//
// CHECK-NEXT: ( 21, 39, 73, 24, 20, 36, 37, 29 )
//
//
// Free.
//
dealloc %AVAL : memref<4xvector<8xf32>>
dealloc %AIDX : memref<4xvector<8xi32>>
dealloc %X : memref<?xf32>
dealloc %B : memref<1xvector<8xf32>>
return
}