Skip to content

[Tests][S-TIR] Fix stale MetaSchedule sketch expectations and migrate let binds to T.let #19729

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
tqchen merged 4 commits into from
Jun 11, 2026
Merged

[Tests][S-TIR] Fix stale MetaSchedule sketch expectations and migrate let binds to T.let #19729

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
6e4a4da
[Tests][MetaSchedule] Update s_tir sketch tests for current defaults
tlopex Jun 11, 2026
a33b2ba
[Tests][S-TIR] Use T.let binds in cross-thread tuple reduction tests
tlopex Jun 11, 2026
f7aa998
[Tests][S-TIR] Migrate remaining let-binding tests to T.let spelling
tlopex Jun 11, 2026
85ffba1
[Tests][S-TIR] Use T.let for reduction temporaries in green tests
tlopex Jun 11, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
4 changes: 2 additions & 2 deletions tests/python/relax/test_transform_rewrite_cuda_graph.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -475,10 +475,10 @@ def layer_norm(
with T.init():
A_red_temp_v0[v_ax0, v_ax1, v_ax2] = T.float32(0)
A_red_temp_v1[v_ax0, v_ax1, v_ax2] = T.float32(0)
v_A_red_temp_v0: T.float32 = A_red_temp_v0[v_ax0, v_ax1, v_ax2] + T.Cast(
v_A_red_temp_v0: T.let[T.float32] = A_red_temp_v0[v_ax0, v_ax1, v_ax2] + T.Cast(
"float32", A[v_ax0, v_ax1, v_ax2, v_k3]
)
v_A_red_temp_v1: T.float32 = A_red_temp_v1[v_ax0, v_ax1, v_ax2] + T.Cast(
v_A_red_temp_v1: T.let[T.float32] = A_red_temp_v1[v_ax0, v_ax1, v_ax2] + T.Cast(
"float32", A[v_ax0, v_ax1, v_ax2, v_k3]
) * T.Cast("float32", A[v_ax0, v_ax1, v_ax2, v_k3])
A_red_temp_v0[v_ax0, v_ax1, v_ax2] = v_A_red_temp_v0
Expand Down
16 changes: 8 additions & 8 deletions tests/python/s_tir/dlight/test_gpu_general_reduction.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -337,8 +337,8 @@ def main(p_lv6: T.handle, weight1: T.Buffer((T.int64(2560),), "float32"), bias:
with T.init():
A_red_temp_v0[v_ax0, v_ax1] = T.float32(0)
A_red_temp_v1[v_ax0, v_ax1] = T.float32(0)
v_A_red_temp_v0: T.float32 = A_red_temp_v0[v_ax0, v_ax1] + lv6[v_ax0, v_ax1, v_k2]
v_A_red_temp_v1: T.float32 = A_red_temp_v1[v_ax0, v_ax1] + lv6[v_ax0, v_ax1, v_k2] * lv6[v_ax0, v_ax1, v_k2]
v_A_red_temp_v0: T.let[T.float32] = A_red_temp_v0[v_ax0, v_ax1] + lv6[v_ax0, v_ax1, v_k2]
v_A_red_temp_v1: T.let[T.float32] = A_red_temp_v1[v_ax0, v_ax1] + lv6[v_ax0, v_ax1, v_k2] * lv6[v_ax0, v_ax1, v_k2]
A_red_temp_v0[v_ax0, v_ax1] = v_A_red_temp_v0
A_red_temp_v1[v_ax0, v_ax1] = v_A_red_temp_v1
for ax0, ax1, ax2 in T.grid(T.int64(1), n, T.int64(2560)):
Expand Down Expand Up @@ -377,8 +377,8 @@ def main(p_lv6: T.handle, weight1: T.Buffer((T.int64(2560),), "float32"), bias:
with T.init():
A_red_temp_v0_shared[T.int64(0), v0] = T.float32(0)
A_red_temp_v1_shared[T.int64(0), v0] = T.float32(0)
v_A_red_temp_v0: T.float32 = A_red_temp_v0_shared[T.int64(0), v0] + lv6[T.int64(0), v0, v1]
v_A_red_temp_v1: T.float32 = A_red_temp_v1_shared[T.int64(0), v0] + lv6[T.int64(0), v0, v1] * lv6[T.int64(0), v0, v1]
v_A_red_temp_v0: T.let[T.float32] = A_red_temp_v0_shared[T.int64(0), v0] + lv6[T.int64(0), v0, v1]
v_A_red_temp_v1: T.let[T.float32] = A_red_temp_v1_shared[T.int64(0), v0] + lv6[T.int64(0), v0, v1] * lv6[T.int64(0), v0, v1]
A_red_temp_v0_shared[T.int64(0), v0] = v_A_red_temp_v0
A_red_temp_v1_shared[T.int64(0), v0] = v_A_red_temp_v1
for ax1_1 in T.thread_binding(T.int64(256), thread="threadIdx.x"):
Expand Down Expand Up @@ -481,8 +481,8 @@ def main(A: T.Buffer((1, 2048), "float32"), B: T.Buffer((2048,), "float32"), C:
with T.init():
A_red_temp_v0[v_ax0, v_ax1] = T.float32(0)
A_red_temp_v1[v_ax0, v_ax1] = T.float32(0)
v_A_red_temp_v0: T.float32 = A_red_temp_v0[v_ax0, v_ax1] + T_reshape_1[v_ax0, v_ax1, v_k2]
v_A_red_temp_v1: T.float32 = A_red_temp_v1[v_ax0, v_ax1] + T_reshape_1[v_ax0, v_ax1, v_k2] * T_reshape_1[v_ax0, v_ax1, v_k2]
v_A_red_temp_v0: T.let[T.float32] = A_red_temp_v0[v_ax0, v_ax1] + T_reshape_1[v_ax0, v_ax1, v_k2]
v_A_red_temp_v1: T.let[T.float32] = A_red_temp_v1[v_ax0, v_ax1] + T_reshape_1[v_ax0, v_ax1, v_k2] * T_reshape_1[v_ax0, v_ax1, v_k2]
A_red_temp_v0[v_ax0, v_ax1] = v_A_red_temp_v0
A_red_temp_v1[v_ax0, v_ax1] = v_A_red_temp_v1
for ax0, ax1 in T.grid(32, 64):
Expand Down Expand Up @@ -531,8 +531,8 @@ def main(A: T.Buffer((1, 2048), "float32"), B: T.Buffer((2048,), "float32"), C:
with T.init():
A_red_temp_v0_shared[0, v0] = T.float32(0)
A_red_temp_v1_shared[0, v0] = T.float32(0)
v_A_red_temp_v0: T.float32 = A_red_temp_v0_shared[0, v0] + A[0, v0 * 64 + v1]
v_A_red_temp_v1: T.float32 = A_red_temp_v1_shared[0, v0] + A[0, v0 * 64 + v1] * A[0, v0 * 64 + v1]
v_A_red_temp_v0: T.let[T.float32] = A_red_temp_v0_shared[0, v0] + A[0, v0 * 64 + v1]
v_A_red_temp_v1: T.let[T.float32] = A_red_temp_v1_shared[0, v0] + A[0, v0 * 64 + v1] * A[0, v0 * 64 + v1]
A_red_temp_v0_shared[0, v0] = v_A_red_temp_v0
A_red_temp_v1_shared[0, v0] = v_A_red_temp_v1
for ax1_1 in T.thread_binding(256, thread="threadIdx.x"):
Expand Down
8 changes: 4 additions & 4 deletions ...thon/s_tir/meta_schedule/test_meta_schedule_postproc_rewrite_parallel_vectorize_unroll.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -241,8 +241,8 @@ def layer_norm(A: T.Buffer((1, 4, 4, 32), "float32"), B: T.Buffer((4, 4, 32), "f
with T.init():
A_red_temp_v0[v_ax0] = T.float32(0)
A_red_temp_v1[v_ax0] = T.float32(0)
v_A_red_temp_v0: T.float32 = A_red_temp_v0[v_ax0] + A[v_ax0, v_k1, v_k2, v_k3]
v_A_red_temp_v1: T.float32 = A_red_temp_v1[v_ax0] + A[v_ax0, v_k1, v_k2, v_k3] * A[v_ax0, v_k1, v_k2, v_k3]
v_A_red_temp_v0: T.let[T.float32] = A_red_temp_v0[v_ax0] + A[v_ax0, v_k1, v_k2, v_k3]
v_A_red_temp_v1: T.let[T.float32] = A_red_temp_v1[v_ax0] + A[v_ax0, v_k1, v_k2, v_k3] * A[v_ax0, v_k1, v_k2, v_k3]
A_red_temp_v0[v_ax0] = v_A_red_temp_v0
A_red_temp_v1[v_ax0] = v_A_red_temp_v1
for ax0, ax1, ax2, ax3 in T.grid(1, 4, 4, 32):
Expand All @@ -267,8 +267,8 @@ def expected(A: T.Buffer((1, 4, 4, 32), "float32"), B: T.Buffer((4, 4, 32), "flo
with T.init():
A_red_temp_v0[0] = T.float32(0)
A_red_temp_v1[0] = T.float32(0)
v_A_red_temp_v0: T.float32 = A_red_temp_v0[0] + A[0, v_k1, v_k2, v_k3]
v_A_red_temp_v1: T.float32 = A_red_temp_v1[0] + A[0, v_k1, v_k2, v_k3] * A[0, v_k1, v_k2, v_k3]
v_A_red_temp_v0: T.let[T.float32] = A_red_temp_v0[0] + A[0, v_k1, v_k2, v_k3]
v_A_red_temp_v1: T.let[T.float32] = A_red_temp_v1[0] + A[0, v_k1, v_k2, v_k3] * A[0, v_k1, v_k2, v_k3]
A_red_temp_v0[0] = v_A_red_temp_v0
A_red_temp_v1[0] = v_A_red_temp_v1
for ax0, ax1, ax2, ax3 in T.grid(1, 4, 4, 32):
Expand Down
28 changes: 16 additions & 12 deletions tests/python/s_tir/meta_schedule/test_meta_schedule_schedule_rule_add_rfactor.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -138,8 +138,10 @@ def argmax(
with T.init():
argmax_v0[i] = -1
argmax_v1[i] = T.min_value("float32")
v_argmax_v0: T.int32 = T.Select(argmax_v1[i] >= val[i, k], argmax_v0[i], idx[i, k])
v_argmax_v1: T.float32 = T.Select(
v_argmax_v0: T.let[T.int32] = T.Select(
argmax_v1[i] >= val[i, k], argmax_v0[i], idx[i, k]
)
v_argmax_v1: T.let[T.float32] = T.Select(
argmax_v1[i] >= val[i, k], argmax_v1[i], val[i, k]
)
argmax_v0[i] = v_argmax_v0
Expand All @@ -160,8 +162,10 @@ def argmax_0(
with T.init():
argmax_v0[i] = -1
argmax_v1[i] = T.float32(-3.4028234663852886e38)
v_argmax_v0: T.int32 = T.Select(argmax_v1[i] >= val[i, k], argmax_v0[i], idx[i, k])
v_argmax_v1: T.float32 = T.Select(
v_argmax_v0: T.let[T.int32] = T.Select(
argmax_v1[i] >= val[i, k], argmax_v0[i], idx[i, k]
)
v_argmax_v1: T.let[T.float32] = T.Select(
argmax_v1[i] >= val[i, k], argmax_v1[i], val[i, k]
)
argmax_v0[i] = v_argmax_v0
Expand All @@ -184,12 +188,12 @@ def argmax_1(
with T.init():
argmax_v0_rf[i, vi1_1] = -1
argmax_v1_rf[i, vi1_1] = T.float32(-3.4028234663852886e38)
v_argmax_v0_rf: T.int32 = T.Select(
v_argmax_v0_rf: T.let[T.int32] = T.Select(
argmax_v1_rf[i, vi1_1] >= val[i, vi1_0 * 16 + vi1_1],
argmax_v0_rf[i, vi1_1],
idx[i, vi1_0 * 16 + vi1_1],
)
v_argmax_v1_rf: T.float32 = T.Select(
v_argmax_v1_rf: T.let[T.float32] = T.Select(
argmax_v1_rf[i, vi1_1] >= val[i, vi1_0 * 16 + vi1_1],
argmax_v1_rf[i, vi1_1],
val[i, vi1_0 * 16 + vi1_1],
Expand All @@ -205,10 +209,10 @@ def argmax_1(
with T.init():
argmax_v0[i] = -1
argmax_v1[i] = T.float32(-3.4028234663852886e38)
v_argmax_v0: T.int32 = T.Select(
v_argmax_v0: T.let[T.int32] = T.Select(
argmax_v1[i] >= argmax_v1_rf[i, vi1_1], argmax_v0[i], argmax_v0_rf[i, vi1_1]
)
v_argmax_v1: T.float32 = T.Select(
v_argmax_v1: T.let[T.float32] = T.Select(
argmax_v1[i] >= argmax_v1_rf[i, vi1_1], argmax_v1[i], argmax_v1_rf[i, vi1_1]
)
argmax_v0[i] = v_argmax_v0
Expand All @@ -233,12 +237,12 @@ def argmax_2(
with T.init():
argmax_v0_rf[i, vi1_0] = -1
argmax_v1_rf[i, vi1_0] = T.float32(-3.4028234663852886e38)
v_argmax_v0_rf: T.int32 = T.Select(
v_argmax_v0_rf: T.let[T.int32] = T.Select(
argmax_v1_rf[i, vi1_0] >= val[i, vi1_0 * 16 + vi1_1],
argmax_v0_rf[i, vi1_0],
idx[i, vi1_0 * 16 + vi1_1],
)
v_argmax_v1_rf: T.float32 = T.Select(
v_argmax_v1_rf: T.let[T.float32] = T.Select(
argmax_v1_rf[i, vi1_0] >= val[i, vi1_0 * 16 + vi1_1],
argmax_v1_rf[i, vi1_0],
val[i, vi1_0 * 16 + vi1_1],
Expand All @@ -254,10 +258,10 @@ def argmax_2(
with T.init():
argmax_v0[i] = -1
argmax_v1[i] = T.float32(-3.4028234663852886e38)
v_argmax_v0: T.int32 = T.Select(
v_argmax_v0: T.let[T.int32] = T.Select(
argmax_v1[i] >= argmax_v1_rf[i, vi1_0], argmax_v0[i], argmax_v0_rf[i, vi1_0]
)
v_argmax_v1: T.float32 = T.Select(
v_argmax_v1: T.let[T.float32] = T.Select(
argmax_v1[i] >= argmax_v1_rf[i, vi1_0], argmax_v1[i], argmax_v1_rf[i, vi1_0]
)
argmax_v0[i] = v_argmax_v0
Expand Down
36 changes: 24 additions & 12 deletions tests/python/s_tir/meta_schedule/test_meta_schedule_schedule_rule_cross_thread_reduction.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -582,8 +582,12 @@ def argmax(
with T.init():
argmax_v0[i] = -1
argmax_v1[i] = T.min_value("float32")
v_argmax_v0: T.int32 = T.Select(argmax_v1[i] >= val[i, k], argmax_v0[i], idx[i, k])
v_argmax_v1: T.float32 = T.Select(argmax_v1[i] >= val[i, k], argmax_v1[i], val[i, k])
v_argmax_v0: T.let[T.int32] = T.Select(
argmax_v1[i] >= val[i, k], argmax_v0[i], idx[i, k]
)
v_argmax_v1: T.let[T.float32] = T.Select(
argmax_v1[i] >= val[i, k], argmax_v1[i], val[i, k]
)
argmax_v0[i] = v_argmax_v0
argmax_v1[i] = v_argmax_v1

Expand All @@ -604,8 +608,12 @@ def argmax_32(
with T.init():
argmax_v0[i] = -1
argmax_v1[i] = T.min_value("float32")
v_argmax_v0: T.int32 = T.Select(argmax_v1[i] >= val[i, k], argmax_v0[i], idx[i, k])
v_argmax_v1: T.float32 = T.Select(argmax_v1[i] >= val[i, k], argmax_v1[i], val[i, k])
v_argmax_v0: T.let[T.int32] = T.Select(
argmax_v1[i] >= val[i, k], argmax_v0[i], idx[i, k]
)
v_argmax_v1: T.let[T.float32] = T.Select(
argmax_v1[i] >= val[i, k], argmax_v1[i], val[i, k]
)
argmax_v0[i] = v_argmax_v0
argmax_v1[i] = v_argmax_v1

Expand All @@ -628,8 +636,10 @@ def argmax_0(
with T.init():
argmax_v0[i] = -1
argmax_v1[i] = T.float32(-3.4028234663852886e38)
v_argmax_v0: T.int32 = T.Select(argmax_v1[i] >= val[i, k], argmax_v0[i], idx[i, k])
v_argmax_v1: T.float32 = T.Select(
v_argmax_v0: T.let[T.int32] = T.Select(
argmax_v1[i] >= val[i, k], argmax_v0[i], idx[i, k]
)
v_argmax_v1: T.let[T.float32] = T.Select(
argmax_v1[i] >= val[i, k], argmax_v1[i], val[i, k]
)
argmax_v0[i] = v_argmax_v0
Expand All @@ -654,10 +664,10 @@ def argmax_1(
with T.init():
argmax_v0[i] = -1
argmax_v1[i] = T.float32(-3.4028234663852886e38)
v_argmax_v0: T.int32 = T.Select(
v_argmax_v0: T.let[T.int32] = T.Select(
argmax_v1[i] >= val[i, k], argmax_v0[i], idx[i, k]
)
v_argmax_v1: T.float32 = T.Select(
v_argmax_v1: T.let[T.float32] = T.Select(
argmax_v1[i] >= val[i, k], argmax_v1[i], val[i, k]
)
argmax_v0[i] = v_argmax_v0
Expand Down Expand Up @@ -701,8 +711,10 @@ def argmax_0(
with T.init():
argmax_v0[i] = -1
argmax_v1[i] = T.float32(-3.4028234663852886e38)
v_argmax_v0: T.int32 = T.Select(argmax_v1[i] >= val[i, k], argmax_v0[i], idx[i, k])
v_argmax_v1: T.float32 = T.Select(
v_argmax_v0: T.let[T.int32] = T.Select(
argmax_v1[i] >= val[i, k], argmax_v0[i], idx[i, k]
)
v_argmax_v1: T.let[T.float32] = T.Select(
argmax_v1[i] >= val[i, k], argmax_v1[i], val[i, k]
)
argmax_v0[i] = v_argmax_v0
Expand All @@ -728,10 +740,10 @@ def argmax_1(
with T.init():
argmax_v0[i] = -1
argmax_v1[i] = T.float32(-3.4028234663852886e38)
v_argmax_v0: T.int32 = T.Select(
v_argmax_v0: T.let[T.int32] = T.Select(
argmax_v1[i] >= val[i, k], argmax_v0[i], idx[i, k]
)
v_argmax_v1: T.float32 = T.Select(
v_argmax_v1: T.let[T.float32] = T.Select(
argmax_v1[i] >= val[i, k], argmax_v1[i], val[i, k]
)
argmax_v0[i] = v_argmax_v0
Expand Down
18 changes: 9 additions & 9 deletions tests/python/s_tir/meta_schedule/test_meta_schedule_space_cuda.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -375,7 +375,7 @@ def cap_0(inputs: T.Buffer((1, 16, 16, 4, 4, 32), "float32"), weight: T.Buffer((
("SamplePerfectTile", [8, 4, 1]),
("SampleCategorical", 1),
("SampleCategorical", 3),
("SampleCategorical", 2),
("SampleCategorical", 3),
]
mod = create_te_workload("CAP", 0)
actual = _design_space(mod)
Expand Down Expand Up @@ -537,7 +537,7 @@ def dil_0(inputs: T.Buffer((1, 224, 224, 3), "float32"), weight: T.Buffer((7, 7,
("SamplePerfectTile", [3, 1, 1]),
("SampleCategorical", 1),
("SampleCategorical", 3),
("SampleCategorical", 3),
("SampleCategorical", 6),
]
mod = create_te_workload("DIL", 0)
actual = _design_space(mod)
Expand Down Expand Up @@ -611,7 +611,7 @@ def gmm_0(X: T.Buffer((1, 128, 128), "float32"), Y: T.Buffer((1, 128, 128), "flo
("SamplePerfectTile", [1, 32, 4]),
("SampleCategorical", 1),
("SampleCategorical", 0),
("SampleCategorical", 4),
("SampleCategorical", 7),
]
mod = create_te_workload("GMM", 0)
actual = _design_space(mod)
Expand Down Expand Up @@ -776,7 +776,7 @@ def t2d_0(inputs: T.Buffer((1, 4, 4, 512), "float32"), weight: T.Buffer((4, 4, 5
("SamplePerfectTile", [16, 4, 8]),
("SampleCategorical", 1),
("SampleCategorical", 3),
("SampleCategorical", 2),
("SampleCategorical", 3),
]
mod = create_te_workload("T2D", 0)
actual = _design_space(mod)
Expand Down Expand Up @@ -846,11 +846,11 @@ def nrm_1(A: T.Buffer((1, 256, 256), "float32"), D: T.Buffer(1, "float32")) -> N
D[v_b] = T.sqrt(C_shared[v_b])
# fmt: on
decision_0 = [
("SampleCategorical", 3),
("SampleCategorical", 6),
]
decision_1 = [
("SampleCategorical", 5),
("SampleCategorical", 4),
("SampleCategorical", 7),
]
mod = create_te_workload("NRM", 0)
actual = _design_space(mod)
Expand Down Expand Up @@ -1043,7 +1043,7 @@ def sfm_3(A: T.Buffer((256, 256), "float32"), T_softmax_norm: T.Buffer((256, 256
]
decision_2 = [
("SampleCategorical", 7),
("SampleCategorical", 3),
("SampleCategorical", 6),
("SampleCategorical", 0),
]
decision_3 = [
Expand Down Expand Up @@ -1132,7 +1132,7 @@ def cbr_0(data: T.Buffer((1, 224, 224, 3), "float32"), kernel: T.Buffer((7, 7, 3
("SamplePerfectTile", [3, 1, 1]),
("SampleCategorical", 0),
("SampleCategorical", 0),
("SampleCategorical", 3),
("SampleCategorical", 6),
]
mod = create_te_workload("CBR", 0)
actual = _design_space(mod)
Expand Down Expand Up @@ -1211,7 +1211,7 @@ def tbg_0(query: T.Buffer((1, 128, 12, 64), "float32"), value: T.Buffer((1, 128,
("SamplePerfectTile", [8, 4, 2]),
("SampleCategorical", 2),
("SampleCategorical", 3),
("SampleCategorical", 4),
("SampleCategorical", 7),
]
mod = create_te_workload("TBG", 0)
actual = _design_space(mod)
Expand Down
2 changes: 1 addition & 1 deletion tests/python/s_tir/meta_schedule/test_meta_schedule_space_cuda_async.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -178,7 +178,7 @@ def test_cuda_c2d():
("SamplePerfectTile", [3, 1, 1]),
("SampleCategorical", 3),
("SampleCategorical", 2),
("SampleCategorical", 4),
("SampleCategorical", 7),
]

mod = create_te_workload("C2D", 0)
Expand Down
8 changes: 4 additions & 4 deletions tests/python/s_tir/schedule/test_tir_schedule_compute_inline.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1424,8 +1424,8 @@ def before(p_lv6: T.handle, weight1: T.Buffer((T.int64(2560),), "float32"), bias
with T.init():
A_red_temp_v0_shared[v_ax0, v_ax1] = T.float32(0)
A_red_temp_v1_shared[v_ax0, v_ax1] = T.float32(0)
v_A_red_temp_v0: T.float32 = A_red_temp_v0_shared[v_ax0, v_ax1] + lv6[v_ax0, v_ax1, v_k2]
v_A_red_temp_v1: T.float32 = A_red_temp_v1_shared[v_ax0, v_ax1] + lv6[v_ax0, v_ax1, v_k2] * lv6[v_ax0, v_ax1, v_k2]
v_A_red_temp_v0: T.let[T.float32] = A_red_temp_v0_shared[v_ax0, v_ax1] + lv6[v_ax0, v_ax1, v_k2]
v_A_red_temp_v1: T.let[T.float32] = A_red_temp_v1_shared[v_ax0, v_ax1] + lv6[v_ax0, v_ax1, v_k2] * lv6[v_ax0, v_ax1, v_k2]
A_red_temp_v0_shared[v_ax0, v_ax1] = v_A_red_temp_v0
A_red_temp_v1_shared[v_ax0, v_ax1] = v_A_red_temp_v1
for ax2_0 in range(T.int64(10)):
Expand Down Expand Up @@ -1465,8 +1465,8 @@ def after(p_lv6: T.handle, weight1: T.Buffer((T.int64(2560),), "float32"), bias:
with T.init():
A_red_temp_v0_shared[v_ax0, v_ax1] = T.float32(0)
A_red_temp_v1_shared[v_ax0, v_ax1] = T.float32(0)
v_A_red_temp_v0: T.float32 = A_red_temp_v0_shared[v_ax0, v_ax1] + lv6[v_ax0, v_ax1, v_k2]
v_A_red_temp_v1: T.float32 = A_red_temp_v1_shared[v_ax0, v_ax1] + lv6[v_ax0, v_ax1, v_k2] * lv6[v_ax0, v_ax1, v_k2]
v_A_red_temp_v0: T.let[T.float32] = A_red_temp_v0_shared[v_ax0, v_ax1] + lv6[v_ax0, v_ax1, v_k2]
v_A_red_temp_v1: T.let[T.float32] = A_red_temp_v1_shared[v_ax0, v_ax1] + lv6[v_ax0, v_ax1, v_k2] * lv6[v_ax0, v_ax1, v_k2]
A_red_temp_v0_shared[v_ax0, v_ax1] = v_A_red_temp_v0
A_red_temp_v1_shared[v_ax0, v_ax1] = v_A_red_temp_v1
for ax2_0 in range(T.int64(10)):
Expand Down
Loading
Loading