Skip to content

Commit

Permalink
Add & test tiling in python
Browse files Browse the repository at this point in the history
  • Loading branch information
Aba committed Oct 18, 2023
1 parent 3bca141 commit 0b4fe23
Show file tree
Hide file tree
Showing 4 changed files with 16 additions and 9 deletions.
16 changes: 9 additions & 7 deletions c/model.h
View file
Original file line number Diff line number Diff line change
@@ -1,16 +1,18 @@
#define N_BUNDLES 7
Bundle_t bundles [N_BUNDLES] = {
{.n=8, .l=2, .kw=11, .coe=2, .coe_tl=2, .r_ll=8, .h=16, .w=8, .w_kw2=3, .t=8, .p=3, .cm=1, .cm_p0=1, .w_bpt=140, .w_bpt_p0=140, .x_bpt=840, .x_bpt_p0=840, .is_bias=1, .b_offset=0, .b_val_shift=9, .b_bias_shift=0, .ca_nzero=0, .ca_shift=12, .ca_pl_scale=0, .x_header=414341061322735616, .x_header_p0=414341061322735616, .w_header=414587437826703360, .w_header_p0=414341061322735616 },
{.n=8, .l=2, .kw=1, .coe=24, .coe_tl=0, .r_ll=8, .h=16, .w=8, .w_kw2=8, .t=1, .p=1, .cm=20, .cm_p0=16, .w_bpt=200, .w_bpt_p0=200, .x_bpt=13320, .x_bpt_p0=13320, .is_bias=0, .b_offset=16, .b_val_shift=0, .b_bias_shift=0, .ca_nzero=1, .ca_shift=3, .ca_pl_scale=0, .x_header=8700964375684448256, .x_header_p0=8700964375684448256, .w_header=8701210795138088960, .w_header_p0=8700964375684448256 },
{.n=8, .l=2, .kw=7, .coe=3, .coe_tl=4, .r_ll=8, .h=16, .w=8, .w_kw2=5, .t=6, .p=8, .cm=2, .cm_p0=2, .w_bpt=176, .w_bpt_p0=176, .x_bpt=1672, .x_bpt_p0=1672, .is_bias=1, .b_offset=16, .b_val_shift=9, .b_bias_shift=0, .ca_nzero=1, .ca_shift=12, .ca_pl_scale=0, .x_header=846686625550303232, .x_header_p0=846686625550303232, .w_header=846933027824074752, .w_header_p0=846686625550303232 },
{.n=8, .l=2, .kw=5, .coe=4, .coe_tl=4, .r_ll=8, .h=16, .w=8, .w_kw2=6, .t=4, .p=4, .cm=4, .cm_p0=4, .w_bpt=248, .w_bpt_p0=248, .x_bpt=3336, .x_bpt_p0=3336, .is_bias=0, .b_offset=34, .b_val_shift=0, .b_bias_shift=0, .ca_nzero=1, .ca_shift=6, .ca_pl_scale=3, .x_header=1927550536119222272, .x_header_p0=1927550536119222272, .w_header=1927796989932601344, .w_header_p0=1927550536119222272 },
{.n=8, .l=2, .kw=3, .coe=8, .coe_tl=8, .r_ll=8, .h=16, .w=8, .w_kw2=7, .t=3, .p=3, .cm=6, .cm_p0=4, .w_bpt=224, .w_bpt_p0=152, .x_bpt=5000, .x_bpt_p0=3336, .is_bias=1, .b_offset=34, .b_val_shift=9, .b_bias_shift=0, .ca_nzero=0, .ca_shift=12, .ca_pl_scale=0, .x_header=3008414446688141312, .x_header_p0=1855492942081294336, .w_header=3008660883321651200, .w_header_p0=1855492942081294336 },
{.n=8, .l=2, .kw=1, .coe=24, .coe_tl=2, .r_ll=8, .h=16, .w=8, .w_kw2=8, .t=3, .p=2, .cm=20, .cm_p0=4, .w_bpt=248, .w_bpt_p0=56, .x_bpt=16648, .x_bpt_p0=3336, .is_bias=0, .b_offset=58, .b_val_shift=0, .b_bias_shift=0, .ca_nzero=1, .ca_shift=6, .ca_pl_scale=3, .x_header=11006807384898142208, .x_header_p0=1783435348043366400, .w_header=11007053838711521280, .w_header_p0=1783435348043366400 },
{.n=1, .l=1, .kw=1, .coe=24, .coe_tl=0, .r_ll=8, .h=8, .w=1, .w_kw2=1, .t=1, .p=320, .cm=20, .cm_p0=20, .w_bpt=248, .w_bpt_p0=248, .x_bpt=138, .x_bpt_p0=138, .is_bias=1, .b_offset=58, .b_val_shift=9, .b_bias_shift=0, .ca_nzero=1, .ca_shift=15, .ca_pl_scale=3, .x_header=10952754293765046272, .x_header_p0=10952754293765046272, .w_header=10952754456973803520, .w_header_p0=10952754293765046272 }
{.n=8, .l=2, .kw=11, .coe=2, .coe_tl=2, .r_ll=8, .h=16, .w=8, .ci=3, .co=16, .w_kw2=3, .t=8, .p=3, .cm=1, .cm_p0=1, .w_bpt=140, .w_bpt_p0=140, .x_bpt=840, .x_bpt_p0=840, .is_bias=1, .b_offset=0, .b_val_shift=9, .b_bias_shift=0, .ca_nzero=0, .ca_shift=12, .ca_pl_scale=0, .x_header=414341061322735616, .x_header_p0=414341061322735616, .w_header=414587437826703360, .w_header_p0=414341061322735616 },
{.n=8, .l=2, .kw=1, .coe=24, .coe_tl=0, .r_ll=8, .h=16, .w=8, .ci=16, .co=16, .w_kw2=8, .t=1, .p=1, .cm=20, .cm_p0=16, .w_bpt=200, .w_bpt_p0=200, .x_bpt=13320, .x_bpt_p0=13320, .is_bias=0, .b_offset=16, .b_val_shift=0, .b_bias_shift=0, .ca_nzero=1, .ca_shift=3, .ca_pl_scale=0, .x_header=8700964375684448256, .x_header_p0=8700964375684448256, .w_header=8701210795138088960, .w_header_p0=8700964375684448256 },
{.n=8, .l=2, .kw=7, .coe=3, .coe_tl=4, .r_ll=8, .h=16, .w=8, .ci=16, .co=16, .w_kw2=5, .t=6, .p=8, .cm=2, .cm_p0=2, .w_bpt=176, .w_bpt_p0=176, .x_bpt=1672, .x_bpt_p0=1672, .is_bias=1, .b_offset=16, .b_val_shift=9, .b_bias_shift=0, .ca_nzero=1, .ca_shift=12, .ca_pl_scale=0, .x_header=846686625550303232, .x_header_p0=846686625550303232, .w_header=846933027824074752, .w_header_p0=846686625550303232 },
{.n=8, .l=2, .kw=5, .coe=4, .coe_tl=4, .r_ll=8, .h=16, .w=8, .ci=16, .co=16, .w_kw2=6, .t=4, .p=4, .cm=4, .cm_p0=4, .w_bpt=248, .w_bpt_p0=248, .x_bpt=3336, .x_bpt_p0=3336, .is_bias=0, .b_offset=34, .b_val_shift=0, .b_bias_shift=0, .ca_nzero=1, .ca_shift=6, .ca_pl_scale=3, .x_header=1927550536119222272, .x_header_p0=1927550536119222272, .w_header=1927796989932601344, .w_header_p0=1927550536119222272 },
{.n=8, .l=2, .kw=3, .coe=8, .coe_tl=8, .r_ll=8, .h=16, .w=8, .ci=16, .co=24, .w_kw2=7, .t=3, .p=3, .cm=6, .cm_p0=4, .w_bpt=224, .w_bpt_p0=152, .x_bpt=5000, .x_bpt_p0=3336, .is_bias=1, .b_offset=34, .b_val_shift=9, .b_bias_shift=0, .ca_nzero=0, .ca_shift=12, .ca_pl_scale=0, .x_header=3008414446688141312, .x_header_p0=1855492942081294336, .w_header=3008660883321651200, .w_header_p0=1855492942081294336 },
{.n=8, .l=2, .kw=1, .coe=24, .coe_tl=2, .r_ll=8, .h=16, .w=8, .ci=24, .co=50, .w_kw2=8, .t=3, .p=2, .cm=20, .cm_p0=4, .w_bpt=248, .w_bpt_p0=56, .x_bpt=16648, .x_bpt_p0=3336, .is_bias=0, .b_offset=58, .b_val_shift=0, .b_bias_shift=0, .ca_nzero=1, .ca_shift=6, .ca_pl_scale=3, .x_header=11006807384898142208, .x_header_p0=1783435348043366400, .w_header=11007053838711521280, .w_header_p0=1783435348043366400 },
{.n=1, .l=1, .kw=1, .coe=24, .coe_tl=0, .r_ll=8, .h=8, .w=1, .ci=6400, .co=10, .w_kw2=1, .t=1, .p=320, .cm=20, .cm_p0=20, .w_bpt=248, .w_bpt_p0=248, .x_bpt=138, .x_bpt_p0=138, .is_bias=1, .b_offset=58, .b_val_shift=9, .b_bias_shift=0, .ca_nzero=1, .ca_shift=15, .ca_pl_scale=3, .x_header=10952754293765046272, .x_header_p0=10952754293765046272, .w_header=10952754456973803520, .w_header_p0=10952754293765046272 }
};

#define X_BITS_L2 2
#define W_BITS_L2 2
#define X_PAD 5
#define KH_MAX 11
#define PE_ROWS 8
#define PE_COLS 24

Expand Down
2 changes: 1 addition & 1 deletion c/runtime.h
View file
Original file line number Diff line number Diff line change
Expand Up @@ -7,7 +7,7 @@
#endif

typedef struct {
const int n, l, kw, coe, coe_tl, r_ll, h, w, w_kw2, t, p, cm, cm_p0;
const int n, l, kw, coe, coe_tl, r_ll, h, w, ci, co, w_kw2, t, p, cm, cm_p0;
const int w_bpt, w_bpt_p0, x_bpt, x_bpt_p0; // bytes per transfer
const char is_bias;
const int b_offset, b_val_shift, b_bias_shift;
Expand Down
1 change: 1 addition & 0 deletions test/py/bundle.py
View file
Original file line number Diff line number Diff line change
Expand Up @@ -456,6 +456,7 @@ def export (self, c):

self.xe = self.reorder_x_q2e_conv(x_int, c, r)
self.ye_exp = self.reorder_y_q2e_conv(y_int, c, r)
self.o_int = o_int
self.oe_exp = self.reorder_y_q2e_conv(o_int, c, r)
print(f"x reshape: [int]:{self.inp['int'].shape}, int:{x_int.shape}. xe:{self.xe[0].shape}")

Expand Down
6 changes: 5 additions & 1 deletion test/py/param_test.py
View file
Original file line number Diff line number Diff line change
Expand Up @@ -280,7 +280,7 @@ def test_dnn_engine(COMPILE):

ca_nzero, ca_shift, ca_pl_scale = b.core['act']['non_zero'], b.core['act']['shift_bits'], b.core['act']['plog_slope']

ch.write(f" {{.n={b.r.XN}, .l={b.r.L}, .kw={b.r.KW}, .coe={y_coe}, .coe_tl={y_coe_tl}, .r_ll={y_r_ll}, .h={b.r.XH}, .w={b.r.XW}, .w_kw2={b.r.XW-b.r.KW//2}, .t={b.r.IT}, .p={b.r.CP}, .cm={b.r.CM}, .cm_p0={b.r.CM_0}, .w_bpt={w_bpt}, .w_bpt_p0={w_bpt_p0}, .x_bpt={x_bpt}, .x_bpt_p0={x_bpt_p0}, .is_bias={1*(b.b is not None)}, .b_offset={b_words}, .b_val_shift={b.bias_val_shift}, .b_bias_shift={b.bias_b_shift}, .ca_nzero={ca_nzero}, .ca_shift={ca_shift}, .ca_pl_scale={ca_pl_scale}, .x_header={b.r.x_header_be_p[-1][0]}, .x_header_p0={b.r.x_header_be_p[0][0]}, .w_header={b.r.w_header_be_p[-1][0]}, .w_header_p0={b.r.x_header_be_p[0][0]} }}")
ch.write(f" {{.n={b.r.XN}, .l={b.r.L}, .kw={b.r.KW}, .coe={y_coe}, .coe_tl={y_coe_tl}, .r_ll={y_r_ll}, .h={b.r.XH}, .w={b.r.XW}, .ci={b.r.CI}, .co={b.r.CO}, .w_kw2={b.r.XW-b.r.KW//2}, .t={b.r.IT}, .p={b.r.CP}, .cm={b.r.CM}, .cm_p0={b.r.CM_0}, .w_bpt={w_bpt}, .w_bpt_p0={w_bpt_p0}, .x_bpt={x_bpt}, .x_bpt_p0={x_bpt_p0}, .is_bias={1*(b.b is not None)}, .b_offset={b_words}, .b_val_shift={b.bias_val_shift}, .b_bias_shift={b.bias_b_shift}, .ca_nzero={ca_nzero}, .ca_shift={ca_shift}, .ca_pl_scale={ca_pl_scale}, .x_header={b.r.x_header_be_p[-1][0]}, .x_header_p0={b.r.x_header_be_p[0][0]}, .w_header={b.r.w_header_be_p[-1][0]}, .w_header_p0={b.r.x_header_be_p[0][0]} }}")

b_words += b.be.size if b.b else 0
if b.idx != len(bundles)-1:
Expand All @@ -289,6 +289,8 @@ def test_dnn_engine(COMPILE):
ch.write(f"\n}};\n\n")
ch.write(f"#define X_BITS_L2 {int(np.log2(c.X_BITS))}\n")
ch.write(f"#define W_BITS_L2 {int(np.log2(c.K_BITS))}\n")
ch.write(f"#define X_PAD {c.X_PAD}\n")
ch.write(f"#define KH_MAX {c.KH_MAX}\n")
ch.write(f"#define PE_ROWS {c.ROWS}\n")
ch.write(f"#define PE_COLS {c.COLS}\n\n")

Expand Down Expand Up @@ -333,6 +335,8 @@ def test_dnn_engine(COMPILE):
'''
for b in bundles:
np.savetxt(f"{DATA_DIR}/{b.idx}_y_exp.txt", b.oe_exp.flatten(), fmt='%d')
np.savetxt(f"{DATA_DIR}/{b.idx}_y_hwc.txt", b.o_int.flatten(), fmt='%d')
np.savetxt(f"{DATA_DIR}/{b.idx}_xe.txt", np.concatenate([a.flatten() for a in b.xe]), fmt='%d')
for ip in range(b.r.CP):
CM_p = b.r.CM_0 if ip==0 else b.r.CM
x_config = b.r.x_header_le_p[ip!=0][0]
Expand Down

0 comments on commit 0b4fe23

Please sign in to comment.