github/concrete
1
1
Fork 0
mirror of https://github.com/zama-ai/concrete.git synced 2026-04-17 03:00:54 -04:00
Code Issues Packages Projects Releases Wiki Activity

test(compiler): Add tests for batching of TFHE operations

Browse Source 
Re-introduce the previously deleted batching tests for BConcrete as
tests for TFHE with the addition of a new test, checking that
non-batchable operands generated by pure operations are hoisted.
This commit is contained in:
Andi Drebes
2023-03-15 16:38:19 +01:00
parent fa5c09a52b
commit 488877ac0b
2 changed files with 218 additions and 197 deletions
Download Patch File Download Diff File Expand all files Collapse all files

218
compilers/concrete-compiler/compiler/tests/check_tests/Transforms/batching.mlir Normal file
View File

@@ -0,0 +1,218 @@
// RUN: concretecompiler --split-input-file --action=dump-batched-tfhe --batch-tfhe-ops %s 2>&1| FileCheck %s
// CHECK-LABEL: func.func @batch_continuous_slice_keyswitch(%arg0: tensor<2x3x4x!TFHE.glwe<sk[1]<1,2048>>>) -> tensor<2x3x4x!TFHE.glwe<sk[3]<1,750>>> {
func.func @batch_continuous_slice_keyswitch(%arg0: tensor<2x3x4x!TFHE.glwe<sk[1]<1,2048>>>) -> tensor<2x3x4x!TFHE.glwe<sk[3]<1,750>>> {
%c2 = arith.constant 2 : index
%c3 = arith.constant 3 : index
%c4 = arith.constant 4 : index
%c0 = arith.constant 0 : index
%c1 = arith.constant 1 : index
%0 = bufferization.alloc_tensor() : tensor<2x3x4x!TFHE.glwe<sk[3]<1,750>>>
// CHECK: %[[V0:.*]] = tensor.collapse_shape [[ARG:.*]] {{\[\[0, 1, 2\]\]}} : tensor<2x3x4x!TFHE.glwe<sk[1]<1,2048>>> into tensor<24x!TFHE.glwe<sk[1]<1,2048>>>
// CHECK: %[[V1:.*]] = "TFHE.batched_keyswitch_glwe"(%[[V0]]) {key = #TFHE.ksk<sk[1]<1,2048>, sk[3]<1,750>, 3, 4>} : (tensor<24x!TFHE.glwe<sk[1]<1,2048>>>) -> tensor<24x!TFHE.glwe<sk[3]<1,750>>>
// CHECK: %[[V2:.*]] = tensor.expand_shape %[[V1]] {{\[\[0, 1, 2\]\]}} : tensor<24x!TFHE.glwe<sk[3]<1,750>>> into tensor<2x3x4x!TFHE.glwe<sk[3]<1,750>>>
// CHECK: return %[[V2]]
%1 = scf.for %arg2 = %c0 to %c2 step %c1 iter_args(%arg3 = %0) -> (tensor<2x3x4x!TFHE.glwe<sk[3]<1,750>>>) {
%2 = scf.for %arg4 = %c0 to %c3 step %c1 iter_args(%arg5 = %arg3) -> (tensor<2x3x4x!TFHE.glwe<sk[3]<1,750>>>) {
%3 = scf.for %arg6 = %c0 to %c4 step %c1 iter_args(%arg7 = %arg5) -> (tensor<2x3x4x!TFHE.glwe<sk[3]<1,750>>>) {
%4 = tensor.extract %arg0[%arg2, %arg4, %arg6] : tensor<2x3x4x!TFHE.glwe<sk[1]<1,2048>>>
%5 = "TFHE.keyswitch_glwe"(%4) {key = #TFHE.ksk<sk[1]<1,2048>, sk[3]<1,750>, 3, 4>} : (!TFHE.glwe<sk[1]<1,2048>>) -> !TFHE.glwe<sk[3]<1,750>>
%7 = tensor.insert %5 into %arg7[%arg2, %arg4, %arg6] : tensor<2x3x4x!TFHE.glwe<sk[3]<1,750>>>
scf.yield %7 : tensor<2x3x4x!TFHE.glwe<sk[3]<1,750>>>
}
scf.yield %3 : tensor<2x3x4x!TFHE.glwe<sk[3]<1,750>>>
}
scf.yield %2 : tensor<2x3x4x!TFHE.glwe<sk[3]<1,750>>>
}
return %1 : tensor<2x3x4x!TFHE.glwe<sk[3]<1,750>>>
}
// -----
// CHECK-LABEL: func.func @batch_continuous_slice_keyswitch_1dim(%arg0: tensor<4x!TFHE.glwe<sk[1]<1,2048>>>) -> tensor<4x!TFHE.glwe<sk[3]<1,750>>> {
func.func @batch_continuous_slice_keyswitch_1dim(%arg0: tensor<4x!TFHE.glwe<sk[1]<1,2048>>>) -> tensor<4x!TFHE.glwe<sk[3]<1,750>>> {
%c4 = arith.constant 4 : index
%c0 = arith.constant 0 : index
%c1 = arith.constant 1 : index
%0 = bufferization.alloc_tensor() : tensor<4x!TFHE.glwe<sk[3]<1,750>>>
// CHECK: %[[V0:.*]] = "TFHE.batched_keyswitch_glwe"(%[[ARG:.*]]) {key = #TFHE.ksk<sk[1]<1,2048>, sk[3]<1,750>, 3, 4>} : (tensor<4x!TFHE.glwe<sk[1]<1,2048>>>) -> tensor<4x!TFHE.glwe<sk[3]<1,750>>>
// CHECK: return %[[V0]]
%1 = scf.for %arg2 = %c0 to %c4 step %c1 iter_args(%arg3 = %0) -> (tensor<4x!TFHE.glwe<sk[3]<1,750>>>) {
%2 = tensor.extract %arg0[%arg2] : tensor<4x!TFHE.glwe<sk[1]<1,2048>>>
%3 = "TFHE.keyswitch_glwe"(%2) {key = #TFHE.ksk<sk[1]<1,2048>, sk[3]<1,750>, 3, 4>} : (!TFHE.glwe<sk[1]<1,2048>>) -> !TFHE.glwe<sk[3]<1,750>>
%4 = tensor.insert %3 into %arg3[%arg2] : tensor<4x!TFHE.glwe<sk[3]<1,750>>>
scf.yield %4 : tensor<4x!TFHE.glwe<sk[3]<1,750>>>
}
return %1 : tensor<4x!TFHE.glwe<sk[3]<1,750>>>
}
// -----
// CHECK-LABEL: func.func @batch_offset_extract_keyswitch(%arg0: tensor<99x2x3x4x99x99x!TFHE.glwe<sk[1]<1,2048>>>) -> tensor<2x3x4x!TFHE.glwe<sk[3]<1,750>>> {
func.func @batch_offset_extract_keyswitch(%arg0: tensor<99x2x3x4x99x99x!TFHE.glwe<sk[1]<1,2048>>>) -> tensor<2x3x4x!TFHE.glwe<sk[3]<1,750>>> {
%c2 = arith.constant 2 : index
%c3 = arith.constant 3 : index
%c4 = arith.constant 4 : index
%c0 = arith.constant 0 : index
%c1 = arith.constant 1 : index
%c97 = arith.constant 97 : index
%0 = bufferization.alloc_tensor() : tensor<2x3x4x!TFHE.glwe<sk[3]<1,750>>>
// CHECK: %[[VDROP1DIMS:.*]] = tensor.collapse_shape [[ARG:.*]] {{\[\[0, 1\], \[2\], \[3, 4, 5\]\]}} : tensor<1x2x3x4x1x1x!TFHE.glwe<sk[1]<1,2048>>> into tensor<2x3x4x!TFHE.glwe<sk[1]<1,2048>>>
// CHECK: %[[V0:.*]] = tensor.collapse_shape %[[VDROP1DIMS]] {{\[\[0, 1, 2\]\]}} : tensor<2x3x4x!TFHE.glwe<sk[1]<1,2048>>> into tensor<24x!TFHE.glwe<sk[1]<1,2048>>>
// CHECK: %[[V1:.*]] = "TFHE.batched_keyswitch_glwe"(%[[V0]]) {key = #TFHE.ksk<sk[1]<1,2048>, sk[3]<1,750>, 3, 4>} : (tensor<24x!TFHE.glwe<sk[1]<1,2048>>>) -> tensor<24x!TFHE.glwe<sk[3]<1,750>>>
// CHECK: %[[V2:.*]] = tensor.expand_shape %[[V1]] {{\[\[0, 1, 2\]\]}} : tensor<24x!TFHE.glwe<sk[3]<1,750>>> into tensor<2x3x4x!TFHE.glwe<sk[3]<1,750>>>
// CHECK: return %[[V2]]
%1 = scf.for %arg2 = %c0 to %c2 step %c1 iter_args(%arg3 = %0) -> (tensor<2x3x4x!TFHE.glwe<sk[3]<1,750>>>) {
%2 = scf.for %arg4 = %c0 to %c3 step %c1 iter_args(%arg5 = %arg3) -> (tensor<2x3x4x!TFHE.glwe<sk[3]<1,750>>>) {
%3 = scf.for %arg6 = %c0 to %c4 step %c1 iter_args(%arg7 = %arg5) -> (tensor<2x3x4x!TFHE.glwe<sk[3]<1,750>>>) {
%4 = tensor.extract %arg0[%c0, %arg2, %arg4, %arg6, %c97, %c1] : tensor<99x2x3x4x99x99x!TFHE.glwe<sk[1]<1,2048>>>
%5 = "TFHE.keyswitch_glwe"(%4) {key = #TFHE.ksk<sk[1]<1,2048>, sk[3]<1,750>, 3, 4>} : (!TFHE.glwe<sk[1]<1,2048>>) -> !TFHE.glwe<sk[3]<1,750>>
%7 = tensor.insert %5 into %arg7[%arg2, %arg4, %arg6] : tensor<2x3x4x!TFHE.glwe<sk[3]<1,750>>>
scf.yield %7 : tensor<2x3x4x!TFHE.glwe<sk[3]<1,750>>>
}
scf.yield %3 : tensor<2x3x4x!TFHE.glwe<sk[3]<1,750>>>
}
scf.yield %2 : tensor<2x3x4x!TFHE.glwe<sk[3]<1,750>>>
}
return %1 : tensor<2x3x4x!TFHE.glwe<sk[3]<1,750>>>
}
// -----
// CHECK-LABEL: func.func @batch_offset_shifted_bounds_nonunitstep_extract_keyswitch(%arg0: tensor<99x20x30x40x99x99x!TFHE.glwe<sk[1]<1,2048>>>) -> tensor<2x2x2x!TFHE.glwe<sk[3]<1,750>>> {
func.func @batch_offset_shifted_bounds_nonunitstep_extract_keyswitch(%arg0: tensor<99x20x30x40x99x99x!TFHE.glwe<sk[1]<1,2048>>>) -> tensor<2x2x2x!TFHE.glwe<sk[3]<1,750>>> {
%c2 = arith.constant 2 : index
%c3 = arith.constant 3 : index
%c4 = arith.constant 4 : index
%c0 = arith.constant 0 : index
%c1 = arith.constant 1 : index
%c7 = arith.constant 7 : index
%c9 = arith.constant 9 : index
%c97 = arith.constant 97 : index
%c24 = arith.constant 24 : index
%0 = bufferization.alloc_tensor() : tensor<2x2x2x!TFHE.glwe<sk[3]<1,750>>>
// CHECK: %[[V0:.*]] = bufferization.alloc_tensor() : tensor<2x2x2x!TFHE.glwe<sk[3]<1,750>>>
// CHECK-NEXT: %[[V1:.*]] = tensor.extract_slice %arg0{{\[0, 3, 7, 9, 97, 1\] \[1, 2, 2, 2, 1, 1\] \[1, 2, 1, 7, 1, 1\]}} : tensor<99x20x30x40x99x99x!TFHE.glwe<sk[1]<1,2048>>> to tensor<1x2x2x2x1x1x!TFHE.glwe<sk[1]<1,2048>>>
// CHECK-NEXT: %[[V2:.*]] = tensor.collapse_shape %[[V1]] {{\[\[0, 1\], \[2\], \[3, 4, 5\]\]}} : tensor<1x2x2x2x1x1x!TFHE.glwe<sk[1]<1,2048>>> into tensor<2x2x2x!TFHE.glwe<sk[1]<1,2048>>>
// CHECK-NEXT: %[[V3:.*]] = tensor.collapse_shape %[[V2]] {{\[\[0, 1, 2\]\]}} : tensor<2x2x2x!TFHE.glwe<sk[1]<1,2048>>> into tensor<8x!TFHE.glwe<sk[1]<1,2048>>>
// CHECK-NEXT: %[[V4:.*]] = "TFHE.batched_keyswitch_glwe"(%[[V3]]) {key = #TFHE.ksk<sk[1]<1,2048>, sk[3]<1,750>, 3, 4>} : (tensor<8x!TFHE.glwe<sk[1]<1,2048>>>) -> tensor<8x!TFHE.glwe<sk[3]<1,750>>>
// CHECK-NEXT: %[[V5:.*]] = tensor.expand_shape %[[V4]] {{\[\[0, 1, 2\]\]}} : tensor<8x!TFHE.glwe<sk[3]<1,750>>> into tensor<2x2x2x!TFHE.glwe<sk[3]<1,750>>>
// CHECK-NEXT: return %[[V5]] : tensor<2x2x2x!TFHE.glwe<sk[3]<1,750>>>
%1 = scf.for %arg2 = %c3 to %c7 step %c2 iter_args(%arg3 = %0) -> (tensor<2x2x2x!TFHE.glwe<sk[3]<1,750>>>) {
%2 = scf.for %arg4 = %c7 to %c9 step %c1 iter_args(%arg5 = %arg3) -> (tensor<2x2x2x!TFHE.glwe<sk[3]<1,750>>>) {
%3 = scf.for %arg6 = %c9 to %c24 step %c7 iter_args(%arg7 = %arg5) -> (tensor<2x2x2x!TFHE.glwe<sk[3]<1,750>>>) {
%4 = tensor.extract %arg0[%c0, %arg2, %arg4, %arg6, %c97, %c1] : tensor<99x20x30x40x99x99x!TFHE.glwe<sk[1]<1,2048>>>
%5 = "TFHE.keyswitch_glwe"(%4) {key = #TFHE.ksk<sk[1]<1,2048>, sk[3]<1,750>, 3, 4>} : (!TFHE.glwe<sk[1]<1,2048>>) -> !TFHE.glwe<sk[3]<1,750>>
%od00 = arith.subi %arg2, %c3 : index
%od01 = arith.divsi %od00, %c2 : index
%od10 = arith.subi %arg4, %c7 : index
%od20 = arith.subi %arg6, %c9 : index
%od21 = arith.divsi %od20, %c7 : index
%7 = tensor.insert %5 into %arg7[%od01, %od10, %od21] : tensor<2x2x2x!TFHE.glwe<sk[3]<1,750>>>
scf.yield %7 : tensor<2x2x2x!TFHE.glwe<sk[3]<1,750>>>
}
scf.yield %3 : tensor<2x2x2x!TFHE.glwe<sk[3]<1,750>>>
}
scf.yield %2 : tensor<2x2x2x!TFHE.glwe<sk[3]<1,750>>>
}
return %1 : tensor<2x2x2x!TFHE.glwe<sk[3]<1,750>>>
}
// -----
// CHECK-LABEL: func.func @apply_lookup_table_contiguous(%arg0: tensor<2x3x4x!TFHE.glwe<sk[1]<1,2048>>>, %arg1: tensor<4xi64>) -> tensor<2x3x4x!TFHE.glwe<sk[1]<1,2048>>> {
// CHECK: %[[Vcollapsed:.*]] = tensor.collapse_shape %[[Varg0:.*]] {{\[\[0, 1, 2\]\]}} : tensor<2x3x4x!TFHE.glwe<sk[1]<1,2048>>> into tensor<24x!TFHE.glwe<sk[1]<1,2048>>>
// CHECK-NEXT: %[[V1:.*]] = "TFHE.batched_keyswitch_glwe"(%[[Vcollapsed]]) {key = #TFHE.ksk<sk[1]<1,2048>, sk[3]<1,750>, 3, 4>} : (tensor<24x!TFHE.glwe<sk[1]<1,2048>>>) -> tensor<24x!TFHE.glwe<sk[3]<1,750>>>
// CHECK-NEXT: %[[V2:.*]] = "TFHE.encode_expand_lut_for_bootstrap"(%[[Varg1:.*]]) {isSigned = false, outputBits = 2 : i32, polySize = 256 : i32} : (tensor<4xi64>) -> tensor<256xi64>
// CHECK-NEXT: %[[V3:.*]] = "TFHE.batched_bootstrap_glwe"(%[[V1]], %[[V2]]) {key = #TFHE.bsk<sk[1]<1,2048>, sk[1]<1,2048>, 1024, 2, 1, 23>} : (tensor<24x!TFHE.glwe<sk[3]<1,750>>>, tensor<256xi64>) -> tensor<24x!TFHE.glwe<sk[1]<1,2048>>>
// CHECK-NEXT: %[[Vexpanded:.*]] = tensor.expand_shape %[[V3]] {{\[\[0, 1, 2\]\]}} : tensor<24x!TFHE.glwe<sk[1]<1,2048>>> into tensor<2x3x4x!TFHE.glwe<sk[1]<1,2048>
func.func @apply_lookup_table_contiguous(%arg0: tensor<2x3x4x!TFHE.glwe<sk[1]<1,2048>>>, %arg1: tensor<4xi64>) -> tensor<2x3x4x!TFHE.glwe<sk[1]<1,2048>>> {
%c0 = arith.constant 0 : index
%c2 = arith.constant 2 : index
%c1 = arith.constant 1 : index
%c3 = arith.constant 3 : index
%c4 = arith.constant 4 : index
%0 = bufferization.alloc_tensor() : tensor<2x3x4x!TFHE.glwe<sk[1]<1,2048>>>
%1 = scf.for %arg2 = %c0 to %c2 step %c1 iter_args(%arg3 = %0) -> (tensor<2x3x4x!TFHE.glwe<sk[1]<1,2048>>>) {
%2 = scf.for %arg4 = %c0 to %c3 step %c1 iter_args(%arg5 = %arg3) -> (tensor<2x3x4x!TFHE.glwe<sk[1]<1,2048>>>) {
%3 = scf.for %arg6 = %c0 to %c4 step %c1 iter_args(%arg7 = %arg5) -> (tensor<2x3x4x!TFHE.glwe<sk[1]<1,2048>>>) {
%extracted = tensor.extract %arg0[%arg2, %arg4, %arg6] : tensor<2x3x4x!TFHE.glwe<sk[1]<1,2048>>>
%4 = "TFHE.encode_expand_lut_for_bootstrap"(%arg1) {isSigned = false, outputBits = 2 : i32, polySize = 256 : i32} : (tensor<4xi64>) -> tensor<256xi64>
%5 = "TFHE.keyswitch_glwe"(%extracted) {key = #TFHE.ksk<sk[1]<1,2048>, sk[3]<1,750>, 3, 4>} : (!TFHE.glwe<sk[1]<1,2048>>) -> !TFHE.glwe<sk[3]<1,750>>
%6 = "TFHE.bootstrap_glwe"(%5, %4) {key = #TFHE.bsk<sk[1]<1,2048>, sk[1]<1,2048>, 1024, 2, 1, 23>} : (!TFHE.glwe<sk[3]<1,750>>, tensor<256xi64>) -> !TFHE.glwe<sk[1]<1,2048>>
%inserted = tensor.insert %6 into %arg7[%arg2, %arg4, %arg6] : tensor<2x3x4x!TFHE.glwe<sk[1]<1,2048>>>
scf.yield %inserted : tensor<2x3x4x!TFHE.glwe<sk[1]<1,2048>>>
}
scf.yield %3 : tensor<2x3x4x!TFHE.glwe<sk[1]<1,2048>>>
}
scf.yield %2 : tensor<2x3x4x!TFHE.glwe<sk[1]<1,2048>>>
}
return %1 : tensor<2x3x4x!TFHE.glwe<sk[1]<1,2048>>>
}
// -----
// CHECK-LABEL: func.func @apply_lookup_table_contiguous_hoistable_nonbatchable(%arg0: tensor<2x3x4x!TFHE.glwe<sk[1]<1,2048>>>, %arg1: tensor<4xi64>) -> tensor<2x3x4x!TFHE.glwe<sk[1]<1,2048>>> {
// CHECK: %[[V1:.*]] = "TFHE.batched_keyswitch_glwe"(%[[Vcollapsed]]) {key = #TFHE.ksk<sk[1]<1,2048>, sk[3]<1,750>, 3, 4>} : (tensor<24x!TFHE.glwe<sk[1]<1,2048>>>) -> tensor<24x!TFHE.glwe<sk[3]<1,750>>>
// CHECK: %[[V3:.*]] = "TFHE.batched_bootstrap_glwe"(%[[V1]], %[[V2:.*]]) {key = #TFHE.bsk<sk[1]<1,2048>, sk[1]<1,2048>, 1024, 2, 1, 23>} : (tensor<24x!TFHE.glwe<sk[3]<1,750>>>, tensor<256xi64>) -> tensor<24x!TFHE.glwe<sk[1]<1,2048>>>
func.func @apply_lookup_table_contiguous_hoistable_nonbatchable(%arg0: tensor<2x3x4x!TFHE.glwe<sk[1]<1,2048>>>, %arg1: tensor<4xi64>) -> tensor<2x3x4x!TFHE.glwe<sk[1]<1,2048>>> {
%c0 = arith.constant 0 : index
%c2 = arith.constant 2 : index
%c1 = arith.constant 1 : index
%c3 = arith.constant 3 : index
%c4 = arith.constant 4 : index
%c256 = arith.constant 256 : index
%c1_i64 = arith.constant 1 : i64
%c0_i64 = arith.constant 0 : i64
%0 = bufferization.alloc_tensor() : tensor<2x3x4x!TFHE.glwe<sk[1]<1,2048>>>
%1 = scf.for %arg2 = %c0 to %c2 step %c1 iter_args(%arg3 = %0) -> (tensor<2x3x4x!TFHE.glwe<sk[1]<1,2048>>>) {
%2 = scf.for %arg4 = %c0 to %c3 step %c1 iter_args(%arg5 = %arg3) -> (tensor<2x3x4x!TFHE.glwe<sk[1]<1,2048>>>) {
%3 = scf.for %arg6 = %c0 to %c4 step %c1 iter_args(%arg7 = %arg5) -> (tensor<2x3x4x!TFHE.glwe<sk[1]<1,2048>>>) {
%extracted = tensor.extract %arg0[%arg2, %arg4, %arg6] : tensor<2x3x4x!TFHE.glwe<sk[1]<1,2048>>>
%4 = "TFHE.encode_expand_lut_for_bootstrap"(%arg1) {isSigned = false, outputBits = 2 : i32, polySize = 256 : i32} : (tensor<4xi64>) -> tensor<256xi64>
%aa = tensor.from_elements
%c1_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64,
%c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64,
%c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64,
%c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64,
%c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64,
%c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64,
%c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64,
%c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64,
%c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64,
%c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64,
%c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64,
%c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64,
%c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64,
%c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64,
%c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64,
%c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64, %c0_i64
: tensor<256xi64>
%b = tensor.extract_slice %4[0][255][1] : tensor<256xi64> to tensor<255xi64>
%x = tensor.extract %4[%c0] : tensor<256xi64>
%c = tensor.insert_slice %b into %aa[1][255][1] : tensor<255xi64> into tensor<256xi64>
%d = tensor.insert %x into %c[%c0] : tensor<256xi64>
%5 = "TFHE.keyswitch_glwe"(%extracted) {key = #TFHE.ksk<sk[1]<1,2048>, sk[3]<1,750>, 3, 4>} : (!TFHE.glwe<sk[1]<1,2048>>) -> !TFHE.glwe<sk[3]<1,750>>
%6 = "TFHE.bootstrap_glwe"(%5, %d) {key = #TFHE.bsk<sk[1]<1,2048>, sk[1]<1,2048>, 1024, 2, 1, 23>} : (!TFHE.glwe<sk[3]<1,750>>, tensor<256xi64>) -> !TFHE.glwe<sk[1]<1,2048>>
%inserted = tensor.insert %6 into %arg7[%arg2, %arg4, %arg6] : tensor<2x3x4x!TFHE.glwe<sk[1]<1,2048>>>
scf.yield %inserted : tensor<2x3x4x!TFHE.glwe<sk[1]<1,2048>>>
}
scf.yield %3 : tensor<2x3x4x!TFHE.glwe<sk[1]<1,2048>>>
}
scf.yield %2 : tensor<2x3x4x!TFHE.glwe<sk[1]<1,2048>>>
}
return %1 : tensor<2x3x4x!TFHE.glwe<sk[1]<1,2048>>>
}

197
compilers/concrete-compiler/compiler/tests/check_tests/Transforms/batching.mlir.disabled
View File

@@ -1,197 +0,0 @@
// RUN: concretecompiler --split-input-file --action=dump-concrete --batch-concrete-ops %s 2>&1| FileCheck %s
// CHECK-LABEL: func.func @batch_continuous_slice_keyswitch(%arg0: tensor<2x3x4x!Concrete.lwe_ciphertext<572,2>>) -> tensor<2x3x4x!Concrete.lwe_ciphertext<572,2>> {
func.func @batch_continuous_slice_keyswitch(%arg0: tensor<2x3x4x!Concrete.lwe_ciphertext<572,2>>) -> tensor<2x3x4x!Concrete.lwe_ciphertext<572,2>> {
%c2 = arith.constant 2 : index
%c3 = arith.constant 3 : index
%c4 = arith.constant 4 : index
%c0 = arith.constant 0 : index
%c1 = arith.constant 1 : index
%0 = bufferization.alloc_tensor() : tensor<2x3x4x!Concrete.lwe_ciphertext<572,2>>
// CHECK: %[[V0:.*]] = tensor.collapse_shape [[ARG:.*]] {{\[\[0, 1, 2\]\]}} : tensor<2x3x4x!Concrete.lwe_ciphertext<572,2>> into tensor<24x!Concrete.lwe_ciphertext<572,2>>
// CHECK: %[[V1:.*]] = "Concrete.batched_keyswitch_lwe"(%[[V0]]) {baseLog = 2 : i32, level = 5 : i32} : (tensor<24x!Concrete.lwe_ciphertext<572,2>>) -> tensor<24x!Concrete.lwe_ciphertext<572,2>>
// CHECK: %[[V2:.*]] = tensor.expand_shape %[[V1]] {{\[\[0, 1, 2\]\]}} : tensor<24x!Concrete.lwe_ciphertext<572,2>> into tensor<2x3x4x!Concrete.lwe_ciphertext<572,2>>
// CHECK: return %[[V2]]
%1 = scf.for %arg2 = %c0 to %c2 step %c1 iter_args(%arg3 = %0) -> (tensor<2x3x4x!Concrete.lwe_ciphertext<572,2>>) {
%2 = scf.for %arg4 = %c0 to %c3 step %c1 iter_args(%arg5 = %arg3) -> (tensor<2x3x4x!Concrete.lwe_ciphertext<572,2>>) {
%3 = scf.for %arg6 = %c0 to %c4 step %c1 iter_args(%arg7 = %arg5) -> (tensor<2x3x4x!Concrete.lwe_ciphertext<572,2>>) {
%4 = tensor.extract %arg0[%arg2, %arg4, %arg6] : tensor<2x3x4x!Concrete.lwe_ciphertext<572,2>>
%5 = "Concrete.keyswitch_lwe"(%4) {baseLog = 2 : i32, level = 5 : i32} : (!Concrete.lwe_ciphertext<572,2>) -> !Concrete.lwe_ciphertext<572,2>
%7 = tensor.insert %5 into %arg7[%arg2, %arg4, %arg6] : tensor<2x3x4x!Concrete.lwe_ciphertext<572,2>>
scf.yield %7 : tensor<2x3x4x!Concrete.lwe_ciphertext<572,2>>
}
scf.yield %3 : tensor<2x3x4x!Concrete.lwe_ciphertext<572,2>>
}
scf.yield %2 : tensor<2x3x4x!Concrete.lwe_ciphertext<572,2>>
}
return %1 : tensor<2x3x4x!Concrete.lwe_ciphertext<572,2>>
}
// -----
// CHECK-LABEL: func.func @batch_continuous_slice_keyswitch_1dim(%arg0: tensor<4x!Concrete.lwe_ciphertext<572,2>>) -> tensor<4x!Concrete.lwe_ciphertext<572,2>> {
func.func @batch_continuous_slice_keyswitch_1dim(%arg0: tensor<4x!Concrete.lwe_ciphertext<572,2>>) -> tensor<4x!Concrete.lwe_ciphertext<572,2>> {
%c4 = arith.constant 4 : index
%c0 = arith.constant 0 : index
%c1 = arith.constant 1 : index
%0 = bufferization.alloc_tensor() : tensor<4x!Concrete.lwe_ciphertext<572,2>>
// CHECK: %[[V0:.*]] = "Concrete.batched_keyswitch_lwe"(%[[ARG:.*]]) {baseLog = 2 : i32, level = 5 : i32} : (tensor<4x!Concrete.lwe_ciphertext<572,2>>) -> tensor<4x!Concrete.lwe_ciphertext<572,2>>
// CHECK: return %[[V0]]
%1 = scf.for %arg2 = %c0 to %c4 step %c1 iter_args(%arg3 = %0) -> (tensor<4x!Concrete.lwe_ciphertext<572,2>>) {
%2 = tensor.extract %arg0[%arg2] : tensor<4x!Concrete.lwe_ciphertext<572,2>>
%3 = "Concrete.keyswitch_lwe"(%2) {baseLog = 2 : i32, level = 5 : i32} : (!Concrete.lwe_ciphertext<572,2>) -> !Concrete.lwe_ciphertext<572,2>
%4 = tensor.insert %3 into %arg3[%arg2] : tensor<4x!Concrete.lwe_ciphertext<572,2>>
scf.yield %4 : tensor<4x!Concrete.lwe_ciphertext<572,2>>
}
return %1 : tensor<4x!Concrete.lwe_ciphertext<572,2>>
}
// -----
// CHECK-LABEL: func.func @batch_continuous_slice_bootstrap(%arg0: tensor<2x3x4x!Concrete.lwe_ciphertext<572,2>>, %arg1: tensor<4xi64>) -> tensor<2x3x4x!Concrete.lwe_ciphertext<1024,2>> {
func.func @batch_continuous_slice_bootstrap(%arg0: tensor<2x3x4x!Concrete.lwe_ciphertext<572,2>>, %arg1: tensor<4xi64>) -> tensor<2x3x4x!Concrete.lwe_ciphertext<1024,2>> {
%c2 = arith.constant 2 : index
%c3 = arith.constant 3 : index
%c4 = arith.constant 4 : index
%c0 = arith.constant 0 : index
%c1 = arith.constant 1 : index
%0 = bufferization.alloc_tensor() : tensor<2x3x4x!Concrete.lwe_ciphertext<1024,2>>
// CHECK: %[[V0:.*]] = tensor.collapse_shape [[ARG:.*]] {{\[\[0, 1, 2\]\]}} : tensor<2x3x4x!Concrete.lwe_ciphertext<572,2>> into tensor<24x!Concrete.lwe_ciphertext<572,2>>
// CHECK: %[[V1:.*]] = "Concrete.batched_bootstrap_lwe"(%[[V0]], %arg1) {baseLog = 8 : i32, glweDimension = 1 : i32, level = 2 : i32, polySize = 1024 : i32} : (tensor<24x!Concrete.lwe_ciphertext<572,2>>, tensor<4xi64>) -> tensor<24x!Concrete.lwe_ciphertext<1024,2>>
// CHECK: %[[V2:.*]] = tensor.expand_shape %[[V1]] {{\[\[0, 1, 2\]\]}} : tensor<24x!Concrete.lwe_ciphertext<1024,2>> into tensor<2x3x4x!Concrete.lwe_ciphertext<1024,2>>
// CHECK: return %[[V2]]
%1 = scf.for %arg2 = %c0 to %c2 step %c1 iter_args(%arg3 = %0) -> (tensor<2x3x4x!Concrete.lwe_ciphertext<1024,2>>) {
%2 = scf.for %arg4 = %c0 to %c3 step %c1 iter_args(%arg5 = %arg3) -> (tensor<2x3x4x!Concrete.lwe_ciphertext<1024,2>>) {
%3 = scf.for %arg6 = %c0 to %c4 step %c1 iter_args(%arg7 = %arg5) -> (tensor<2x3x4x!Concrete.lwe_ciphertext<1024,2>>) {
%4 = tensor.extract %arg0[%arg2, %arg4, %arg6] : tensor<2x3x4x!Concrete.lwe_ciphertext<572,2>>
%5 = "Concrete.bootstrap_lwe"(%4, %arg1) {baseLog = 8 : i32, glweDimension = 1 : i32, level = 2 : i32, polySize = 1024 : i32} : (!Concrete.lwe_ciphertext<572,2>, tensor<4xi64>) -> !Concrete.lwe_ciphertext<1024,2>
%6 = tensor.insert %5 into %arg7[%arg2, %arg4, %arg6] : tensor<2x3x4x!Concrete.lwe_ciphertext<1024,2>>
scf.yield %6 : tensor<2x3x4x!Concrete.lwe_ciphertext<1024,2>>
}
scf.yield %3 : tensor<2x3x4x!Concrete.lwe_ciphertext<1024,2>>
}
scf.yield %2 : tensor<2x3x4x!Concrete.lwe_ciphertext<1024,2>>
}
return %1 : tensor<2x3x4x!Concrete.lwe_ciphertext<1024,2>>
}
// -----
// CHECK-LABEL: func.func @batch_continuous_slice_apply_lookup_table(%arg0: tensor<2x3x4x!Concrete.lwe_ciphertext<1024,2>>, %arg1: tensor<4xi64>) -> tensor<2x3x4x!Concrete.lwe_ciphertext<1024,2>> {
func.func @batch_continuous_slice_apply_lookup_table(%arg0: tensor<2x3x4x!Concrete.lwe_ciphertext<1024,2>>, %arg1: tensor<4xi64>) -> tensor<2x3x4x!Concrete.lwe_ciphertext<1024,2>> {
%c2 = arith.constant 2 : index
%c3 = arith.constant 3 : index
%c4 = arith.constant 4 : index
%c0 = arith.constant 0 : index
%c1 = arith.constant 1 : index
// CHECK: %[[V0:.*]] = tensor.collapse_shape [[ARG:.*]] {{\[\[0, 1, 2\]\]}} : tensor<2x3x4x!Concrete.lwe_ciphertext<1024,2>> into tensor<24x!Concrete.lwe_ciphertext<1024,2>>
// CHECK: %[[V1:.*]] = "Concrete.batched_keyswitch_lwe"(%[[V0]]) {baseLog = 2 : i32, level = 5 : i32} : (tensor<24x!Concrete.lwe_ciphertext<1024,2>>) -> tensor<24x!Concrete.lwe_ciphertext<572,2>>
// CHECK: %[[V2:.*]] = "Concrete.batched_bootstrap_lwe"(%[[V1]], %arg1) {baseLog = 8 : i32, glweDimension = 1 : i32, level = 2 : i32, polySize = 1024 : i32} : (tensor<24x!Concrete.lwe_ciphertext<572,2>>, tensor<4xi64>) -> tensor<24x!Concrete.lwe_ciphertext<1024,2>>
// CHECK: %[[V3:.*]] = tensor.expand_shape %[[V2]] {{\[\[0, 1, 2\]\]}} : tensor<24x!Concrete.lwe_ciphertext<1024,2>> into tensor<2x3x4x!Concrete.lwe_ciphertext<1024,2>>
// CHECK: return %[[V3]]
%0 = bufferization.alloc_tensor() : tensor<2x3x4x!Concrete.lwe_ciphertext<1024,2>>
%1 = scf.for %arg2 = %c0 to %c2 step %c1 iter_args(%arg3 = %0) -> (tensor<2x3x4x!Concrete.lwe_ciphertext<1024,2>>) {
%2 = scf.for %arg4 = %c0 to %c3 step %c1 iter_args(%arg5 = %arg3) -> (tensor<2x3x4x!Concrete.lwe_ciphertext<1024,2>>) {
%3 = scf.for %arg6 = %c0 to %c4 step %c1 iter_args(%arg7 = %arg5) -> (tensor<2x3x4x!Concrete.lwe_ciphertext<1024,2>>) {
%4 = tensor.extract %arg0[%arg2, %arg4, %arg6] : tensor<2x3x4x!Concrete.lwe_ciphertext<1024,2>>
%5 = "Concrete.keyswitch_lwe"(%4) {baseLog = 2 : i32, level = 5 : i32} : (!Concrete.lwe_ciphertext<1024,2>) -> !Concrete.lwe_ciphertext<572,2>
%6 = "Concrete.bootstrap_lwe"(%5, %arg1) {baseLog = 8 : i32, glweDimension = 1 : i32, level = 2 : i32, polySize = 1024 : i32} : (!Concrete.lwe_ciphertext<572,2>, tensor<4xi64>) -> !Concrete.lwe_ciphertext<1024,2>
%7 = tensor.insert %6 into %arg7[%arg2, %arg4, %arg6] : tensor<2x3x4x!Concrete.lwe_ciphertext<1024,2>>
scf.yield %7 : tensor<2x3x4x!Concrete.lwe_ciphertext<1024,2>>
}
scf.yield %3 : tensor<2x3x4x!Concrete.lwe_ciphertext<1024,2>>
}
scf.yield %2 : tensor<2x3x4x!Concrete.lwe_ciphertext<1024,2>>
}
return %1 : tensor<2x3x4x!Concrete.lwe_ciphertext<1024,2>>
}
// -----
// CHECK-LABEL: func.func @batch_offset_extract_keyswitch(%arg0: tensor<99x2x3x4x99x99x!Concrete.lwe_ciphertext<572,2>>) -> tensor<2x3x4x!Concrete.lwe_ciphertext<572,2>> {
func.func @batch_offset_extract_keyswitch(%arg0: tensor<99x2x3x4x99x99x!Concrete.lwe_ciphertext<572,2>>) -> tensor<2x3x4x!Concrete.lwe_ciphertext<572,2>> {
%c2 = arith.constant 2 : index
%c3 = arith.constant 3 : index
%c4 = arith.constant 4 : index
%c0 = arith.constant 0 : index
%c1 = arith.constant 1 : index
%c97 = arith.constant 97 : index
%0 = bufferization.alloc_tensor() : tensor<2x3x4x!Concrete.lwe_ciphertext<572,2>>
// CHECK: %[[VDROP1DIMS:.*]] = tensor.collapse_shape [[ARG:.*]] {{\[\[0, 1\], \[2\], \[3, 4, 5\]\]}} : tensor<1x2x3x4x1x1x!Concrete.lwe_ciphertext<572,2>> into tensor<2x3x4x!Concrete.lwe_ciphertext<572,2>>
// CHECK: %[[V0:.*]] = tensor.collapse_shape %[[VDROP1DIMS]] {{\[\[0, 1, 2\]\]}} : tensor<2x3x4x!Concrete.lwe_ciphertext<572,2>> into tensor<24x!Concrete.lwe_ciphertext<572,2>>
// CHECK: %[[V1:.*]] = "Concrete.batched_keyswitch_lwe"(%[[V0]]) {baseLog = 2 : i32, level = 5 : i32} : (tensor<24x!Concrete.lwe_ciphertext<572,2>>) -> tensor<24x!Concrete.lwe_ciphertext<572,2>>
// CHECK: %[[V2:.*]] = tensor.expand_shape %[[V1]] {{\[\[0, 1, 2\]\]}} : tensor<24x!Concrete.lwe_ciphertext<572,2>> into tensor<2x3x4x!Concrete.lwe_ciphertext<572,2>>
// CHECK: return %[[V2]]
%1 = scf.for %arg2 = %c0 to %c2 step %c1 iter_args(%arg3 = %0) -> (tensor<2x3x4x!Concrete.lwe_ciphertext<572,2>>) {
%2 = scf.for %arg4 = %c0 to %c3 step %c1 iter_args(%arg5 = %arg3) -> (tensor<2x3x4x!Concrete.lwe_ciphertext<572,2>>) {
%3 = scf.for %arg6 = %c0 to %c4 step %c1 iter_args(%arg7 = %arg5) -> (tensor<2x3x4x!Concrete.lwe_ciphertext<572,2>>) {
%4 = tensor.extract %arg0[%c0, %arg2, %arg4, %arg6, %c97, %c1] : tensor<99x2x3x4x99x99x!Concrete.lwe_ciphertext<572,2>>
%5 = "Concrete.keyswitch_lwe"(%4) {baseLog = 2 : i32, level = 5 : i32} : (!Concrete.lwe_ciphertext<572,2>) -> !Concrete.lwe_ciphertext<572,2>
%7 = tensor.insert %5 into %arg7[%arg2, %arg4, %arg6] : tensor<2x3x4x!Concrete.lwe_ciphertext<572,2>>
scf.yield %7 : tensor<2x3x4x!Concrete.lwe_ciphertext<572,2>>
}
scf.yield %3 : tensor<2x3x4x!Concrete.lwe_ciphertext<572,2>>
}
scf.yield %2 : tensor<2x3x4x!Concrete.lwe_ciphertext<572,2>>
}
return %1 : tensor<2x3x4x!Concrete.lwe_ciphertext<572,2>>
}
// -----
// CHECK-LABEL: func.func @batch_offset_shifted_bounds_nonunitstep_extract_keyswitch(%arg0: tensor<99x20x30x40x99x99x!Concrete.lwe_ciphertext<572,2>>) -> tensor<2x2x2x!Concrete.lwe_ciphertext<572,2>> {
func.func @batch_offset_shifted_bounds_nonunitstep_extract_keyswitch(%arg0: tensor<99x20x30x40x99x99x!Concrete.lwe_ciphertext<572,2>>) -> tensor<2x2x2x!Concrete.lwe_ciphertext<572,2>> {
%c2 = arith.constant 2 : index
%c3 = arith.constant 3 : index
%c4 = arith.constant 4 : index
%c0 = arith.constant 0 : index
%c1 = arith.constant 1 : index
%c7 = arith.constant 7 : index
%c9 = arith.constant 9 : index
%c97 = arith.constant 97 : index
%c24 = arith.constant 24 : index
%0 = bufferization.alloc_tensor() : tensor<2x2x2x!Concrete.lwe_ciphertext<572,2>>
// CHECK: %[[V0:.*]] = bufferization.alloc_tensor() : tensor<2x2x2x!Concrete.lwe_ciphertext<572,2>>
// CHECK-NEXT: %[[V1:.*]] = tensor.extract_slice %arg0{{\[0, 3, 7, 9, 97, 1\] \[1, 2, 2, 2, 1, 1\] \[1, 2, 1, 7, 1, 1\]}} : tensor<99x20x30x40x99x99x!Concrete.lwe_ciphertext<572,2>> to tensor<1x2x2x2x1x1x!Concrete.lwe_ciphertext<572,2>>
// CHECK-NEXT: %[[V2:.*]] = tensor.collapse_shape %[[V1]] {{\[\[0, 1\], \[2\], \[3, 4, 5\]\]}} : tensor<1x2x2x2x1x1x!Concrete.lwe_ciphertext<572,2>> into tensor<2x2x2x!Concrete.lwe_ciphertext<572,2>>
// CHECK-NEXT: %[[V3:.*]] = tensor.collapse_shape %[[V2]] {{\[\[0, 1, 2\]\]}} : tensor<2x2x2x!Concrete.lwe_ciphertext<572,2>> into tensor<8x!Concrete.lwe_ciphertext<572,2>>
// CHECK-NEXT: %[[V4:.*]] = "Concrete.batched_keyswitch_lwe"(%[[V3]]) {baseLog = 2 : i32, level = 5 : i32} : (tensor<8x!Concrete.lwe_ciphertext<572,2>>) -> tensor<8x!Concrete.lwe_ciphertext<572,2>>
// CHECK-NEXT: %[[V5:.*]] = tensor.expand_shape %[[V4]] {{\[\[0, 1, 2\]\]}} : tensor<8x!Concrete.lwe_ciphertext<572,2>> into tensor<2x2x2x!Concrete.lwe_ciphertext<572,2>>
// CHECK-NEXT: return %[[V5]] : tensor<2x2x2x!Concrete.lwe_ciphertext<572,2>>
%1 = scf.for %arg2 = %c3 to %c7 step %c2 iter_args(%arg3 = %0) -> (tensor<2x2x2x!Concrete.lwe_ciphertext<572,2>>) {
%2 = scf.for %arg4 = %c7 to %c9 step %c1 iter_args(%arg5 = %arg3) -> (tensor<2x2x2x!Concrete.lwe_ciphertext<572,2>>) {
%3 = scf.for %arg6 = %c9 to %c24 step %c7 iter_args(%arg7 = %arg5) -> (tensor<2x2x2x!Concrete.lwe_ciphertext<572,2>>) {
%4 = tensor.extract %arg0[%c0, %arg2, %arg4, %arg6, %c97, %c1] : tensor<99x20x30x40x99x99x!Concrete.lwe_ciphertext<572,2>>
%5 = "Concrete.keyswitch_lwe"(%4) {baseLog = 2 : i32, level = 5 : i32} : (!Concrete.lwe_ciphertext<572,2>) -> !Concrete.lwe_ciphertext<572,2>
%od00 = arith.subi %arg2, %c3 : index
%od01 = arith.divsi %od00, %c2 : index
%od10 = arith.subi %arg4, %c7 : index
%od20 = arith.subi %arg6, %c9 : index
%od21 = arith.divsi %od20, %c7 : index
%7 = tensor.insert %5 into %arg7[%od01, %od10, %od21] : tensor<2x2x2x!Concrete.lwe_ciphertext<572,2>>
scf.yield %7 : tensor<2x2x2x!Concrete.lwe_ciphertext<572,2>>
}
scf.yield %3 : tensor<2x2x2x!Concrete.lwe_ciphertext<572,2>>
}
scf.yield %2 : tensor<2x2x2x!Concrete.lwe_ciphertext<572,2>>
}
return %1 : tensor<2x2x2x!Concrete.lwe_ciphertext<572,2>>
}