You are viewing a plain text version of this content. The canonical link for it is here.
Posted to commits@tvm.apache.org by gu...@apache.org on 2023/01/23 03:16:55 UTC
[tvm] branch acc-tests-1 updated: Layer 2
This is an automated email from the ASF dual-hosted git repository.
guberti pushed a commit to branch acc-tests-1
in repository https://gitbox.apache.org/repos/asf/tvm.git
The following commit(s) were added to refs/heads/acc-tests-1 by this push:
new ff1496818f Layer 2
ff1496818f is described below
commit ff1496818f10d6980b1a410daff5bb21e3f72451
Author: Gavin Uberti <ga...@gmail.com>
AuthorDate: Sat Jan 21 23:02:15 2023 -0800
Layer 2
---
vww/modified.c | 139 +++++++++++++++++++++++++++++++++++++--------------------
1 file changed, 90 insertions(+), 49 deletions(-)
diff --git a/vww/modified.c b/vww/modified.c
index ad6272ac97..0852d1ac49 100644
--- a/vww/modified.c
+++ b/vww/modified.c
@@ -34022,6 +34022,78 @@ TVM_DLL int32_t tvmgen_default_fused_nn_conv2d_add_cast_multiply_add_right_shift
return 0;
}
+static const float __attribute__((section(".rodata.tvm"), aligned(16))) scale_19774001[16] = {
+ 0x1.f7da38p-115, 0x1.8af172p-116, 0x1.a11d4ap-120, 0x1.da0cdep-124, 0x1.ed70dep-125, 0x1.fe72eep-123, 0x1.9c6f06p-117, 0x1.a14884p-114,
+ 0x1.dc58acp-124, 0x1.899598p-123, 0x1.92e09ep-122, 0x1.541228p-122, 0x1.228cd8p-114, 0x1.bd9a8ep-123, 0x1.751a8ap-117, 0x1.6526a4p-115
+};
+
+static const int32_t __attribute__((section(".rodata.tvm"), aligned(16))) bias_8643672[16] = {
+ +0x00006b58, -0x000023aa, +0x00005cf3, +0x00004cc6, +0x0000605e, +0x00006eed, +0x0000512e, -0x00002526,
+ -0x00004c37, +0x00009c70, -0x00005a3f, -0x00003aa9, +0x00005ab4, +0x00009c23, -0x0000ad55, +0x00000839
+};
+
+static const int16_t __attribute__((section(".rodata.tvm"), aligned(16))) kernel_24195235[128] = {
+ +0x007f, -0x0016, +0x003c, -0x0027, +0x0056, +0x0002, +0x0012, -0x0003,
+ -0x0006, +0x001e, -0x0044, -0x007f, +0x0047, -0x0048, +0x007d, -0x002c,
+ +0x005c, +0x007f, -0x0007, -0x0015, -0x0033, -0x0002, +0x0038, -0x0001,
+ +0x0011, -0x0036, +0x0062, +0x0034, +0x007f, +0x0010, -0x001c, -0x0063,
+ -0x001f, +0x0022, +0x0060, -0x0007, +0x0046, -0x0003, +0x0074, -0x007f,
+ -0x000b, +0x000c, +0x000f, +0x0027, +0x001e, +0x007f, -0x0007, +0x000d,
+ -0x0005, +0x0006, -0x0010, +0x007f, +0x002a, -0x0007, +0x001f, +0x0006,
+ -0x0007, -0x0007, +0x0020, -0x007f, -0x0011, +0x0010, -0x0012, +0x0018,
+ +0x0000, -0x002e, -0x004d, +0x0016, +0x0029, -0x0008, -0x0017, -0x007f,
+ -0x001d, +0x007f, +0x000d, +0x0019, +0x0040, +0x001d, +0x0065, -0x0010,
+ +0x0016, -0x0014, -0x0019, -0x0029, -0x000e, -0x007f, -0x0009, -0x001b,
+ -0x007f, -0x0020, -0x000b, -0x0009, -0x000b, -0x0011, +0x0061, -0x0026,
+ +0x001d, -0x0019, +0x0021, +0x007f, -0x001a, -0x0036, +0x0033, +0x004b,
+ -0x0020, +0x0023, +0x007f, +0x0011, +0x0010, +0x0048, +0x0023, +0x002c,
+ -0x006f, -0x003c, +0x0057, +0x0024, -0x0065, -0x0028, -0x007f, -0x0047,
+ -0x000a, +0x0015, -0x000d, -0x0022, -0x002e, -0x000f, +0x007f, -0x000b
+};
+
+#include <arm_acle.h>
+__attribute__((always_inline)) static inline int32_t tensordot_opt_x2_int16_w384_1x8_000_8_1(
+ int32_t *output, int32_t *tensor, int32_t *kernel, int32_t *bias, int32_t *scale
+) {
+ int32_t sum_0 = *bias, sum_1 = *bias;
+
+ int32_t tensor__y00_x00__y00_x01 = tensor[0];
+ int32_t tensor__y00_x02__y00_x03 = tensor[1];
+ int32_t tensor__y00_x04__y00_x05 = tensor[2];
+ int32_t tensor__y00_x06__y00_x07 = tensor[3];
+ int32_t tensor__y00_x08__y00_x09 = tensor[4];
+ int32_t tensor__y00_x0a__y00_x0b = tensor[5];
+ int32_t tensor__y00_x0c__y00_x0d = tensor[6];
+ int32_t tensor__y00_x0e__y00_x0f = tensor[7];
+
+ int32_t kernel__y00_x00__y00_x01 = kernel[0];
+ int32_t kernel__y00_x02__y00_x03 = kernel[1];
+ int32_t kernel__y00_x04__y00_x05 = kernel[2];
+ int32_t kernel__y00_x06__y00_x07 = kernel[3];
+
+ sum_0 = __smlad(tensor__y00_x00__y00_x01, kernel__y00_x00__y00_x01, sum_0);
+ sum_0 = __smlad(tensor__y00_x02__y00_x03, kernel__y00_x02__y00_x03, sum_0);
+ sum_0 = __smlad(tensor__y00_x04__y00_x05, kernel__y00_x04__y00_x05, sum_0);
+ sum_0 = __smlad(tensor__y00_x06__y00_x07, kernel__y00_x06__y00_x07, sum_0);
+ sum_1 = __smlad(tensor__y00_x08__y00_x09, kernel__y00_x00__y00_x01, sum_1);
+ sum_1 = __smlad(tensor__y00_x0a__y00_x0b, kernel__y00_x02__y00_x03, sum_1);
+ sum_1 = __smlad(tensor__y00_x0c__y00_x0d, kernel__y00_x04__y00_x05, sum_1);
+ sum_1 = __smlad(tensor__y00_x0e__y00_x0f, kernel__y00_x06__y00_x07, sum_1);
+
+ int32_t scale_val = *scale;
+ int32_t requant_0 = (sum_0 * (int64_t) scale_val) >> 32;
+ requant_0 = (requant_0 + 1) >> 1;
+ requant_0 = __ssat(requant_0 + -128, 8);
+ int32_t requant_1 = (sum_1 * (int64_t) scale_val) >> 32;
+ requant_1 = (requant_1 + 1) >> 1;
+ requant_1 = __ssat(requant_1 + -128, 8);
+
+ int packed_res_0;
+ __asm__ ("pkhbt %0, %1, %2, lsl #16" : "=r" (packed_res_0) : "r" (requant_0), "r" (requant_1));
+ output[0] = packed_res_0;
+ return 0;
+}
+
#ifdef __cplusplus
extern "C"
#endif
@@ -34054,48 +34126,19 @@ TVM_DLL int32_t tvmgen_default_fused_nn_conv2d_add_cast_multiply_add_right_shift
if (conv == NULL) {
return -1;
}
- for (int32_t i1 = 0; i1 < 48; ++i1) {
- for (int32_t i2 = 0; i2 < 48; ++i2) {
- for (int32_t i3 = 0; i3 < 8; ++i3) {
- int32_t cse_var_1 = (((i1 * 384) + (i2 * 8)) + i3);
- ((int16_t*)PadInput)[cse_var_1] = ((int16_t*)placeholder)[cse_var_1];
- }
- }
- }
for (int32_t n_oho_fused = 0; n_oho_fused < 12; ++n_oho_fused) {
for (int32_t owo = 0; owo < 6; ++owo) {
for (int32_t ohi = 0; ohi < 4; ++ohi) {
for (int32_t ic = 0; ic < 8; ++ic) {
int32_t cse_var_2 = (n_oho_fused * 1536);
- ((int16_t*)data_vec)[(((cse_var_2 + (owo * 256)) + (ohi * 64)) + ic)] = ((int16_t*)PadInput)[(((cse_var_2 + (ohi * 384)) + (owo * 64)) + ic)];
- }
- for (int32_t ic_1 = 0; ic_1 < 8; ++ic_1) {
- int32_t cse_var_3 = (n_oho_fused * 1536);
- ((int16_t*)data_vec)[((((cse_var_3 + (owo * 256)) + (ohi * 64)) + ic_1) + 8)] = ((int16_t*)PadInput)[((((cse_var_3 + (ohi * 384)) + (owo * 64)) + ic_1) + 8)];
- }
- for (int32_t ic_2 = 0; ic_2 < 8; ++ic_2) {
- int32_t cse_var_4 = (n_oho_fused * 1536);
- ((int16_t*)data_vec)[((((cse_var_4 + (owo * 256)) + (ohi * 64)) + ic_2) + 16)] = ((int16_t*)PadInput)[((((cse_var_4 + (ohi * 384)) + (owo * 64)) + ic_2) + 16)];
- }
- for (int32_t ic_3 = 0; ic_3 < 8; ++ic_3) {
- int32_t cse_var_5 = (n_oho_fused * 1536);
- ((int16_t*)data_vec)[((((cse_var_5 + (owo * 256)) + (ohi * 64)) + ic_3) + 24)] = ((int16_t*)PadInput)[((((cse_var_5 + (ohi * 384)) + (owo * 64)) + ic_3) + 24)];
- }
- for (int32_t ic_4 = 0; ic_4 < 8; ++ic_4) {
- int32_t cse_var_6 = (n_oho_fused * 1536);
- ((int16_t*)data_vec)[((((cse_var_6 + (owo * 256)) + (ohi * 64)) + ic_4) + 32)] = ((int16_t*)PadInput)[((((cse_var_6 + (ohi * 384)) + (owo * 64)) + ic_4) + 32)];
- }
- for (int32_t ic_5 = 0; ic_5 < 8; ++ic_5) {
- int32_t cse_var_7 = (n_oho_fused * 1536);
- ((int16_t*)data_vec)[((((cse_var_7 + (owo * 256)) + (ohi * 64)) + ic_5) + 40)] = ((int16_t*)PadInput)[((((cse_var_7 + (ohi * 384)) + (owo * 64)) + ic_5) + 40)];
- }
- for (int32_t ic_6 = 0; ic_6 < 8; ++ic_6) {
- int32_t cse_var_8 = (n_oho_fused * 1536);
- ((int16_t*)data_vec)[((((cse_var_8 + (owo * 256)) + (ohi * 64)) + ic_6) + 48)] = ((int16_t*)PadInput)[((((cse_var_8 + (ohi * 384)) + (owo * 64)) + ic_6) + 48)];
- }
- for (int32_t ic_7 = 0; ic_7 < 8; ++ic_7) {
- int32_t cse_var_9 = (n_oho_fused * 1536);
- ((int16_t*)data_vec)[((((cse_var_9 + (owo * 256)) + (ohi * 64)) + ic_7) + 56)] = ((int16_t*)PadInput)[((((cse_var_9 + (ohi * 384)) + (owo * 64)) + ic_7) + 56)];
+ ((int16_t*)data_vec)[(((cse_var_2 + (owo * 256)) + (ohi * 64)) + ic)] = ((int16_t*)placeholder)[(((cse_var_2 + (ohi * 384)) + (owo * 64)) + ic)];
+ ((int16_t*)data_vec)[((((cse_var_2 + (owo * 256)) + (ohi * 64)) + ic) + 8)] = ((int16_t*)placeholder)[((((cse_var_2 + (ohi * 384)) + (owo * 64)) + ic) + 8)];
+ ((int16_t*)data_vec)[((((cse_var_2 + (owo * 256)) + (ohi * 64)) + ic) + 16)] = ((int16_t*)placeholder)[((((cse_var_2 + (ohi * 384)) + (owo * 64)) + ic) + 16)];
+ ((int16_t*)data_vec)[((((cse_var_2 + (owo * 256)) + (ohi * 64)) + ic) + 24)] = ((int16_t*)placeholder)[((((cse_var_2 + (ohi * 384)) + (owo * 64)) + ic) + 24)];
+ ((int16_t*)data_vec)[((((cse_var_2 + (owo * 256)) + (ohi * 64)) + ic) + 32)] = ((int16_t*)placeholder)[((((cse_var_2 + (ohi * 384)) + (owo * 64)) + ic) + 32)];
+ ((int16_t*)data_vec)[((((cse_var_2 + (owo * 256)) + (ohi * 64)) + ic) + 40)] = ((int16_t*)placeholder)[((((cse_var_2 + (ohi * 384)) + (owo * 64)) + ic) + 40)];
+ ((int16_t*)data_vec)[((((cse_var_2 + (owo * 256)) + (ohi * 64)) + ic) + 48)] = ((int16_t*)placeholder)[((((cse_var_2 + (ohi * 384)) + (owo * 64)) + ic) + 48)];
+ ((int16_t*)data_vec)[((((cse_var_2 + (owo * 256)) + (ohi * 64)) + ic) + 56)] = ((int16_t*)placeholder)[((((cse_var_2 + (ohi * 384)) + (owo * 64)) + ic) + 56)];
}
}
}
@@ -34175,18 +34218,16 @@ TVM_DLL int32_t tvmgen_default_fused_nn_conv2d_add_cast_multiply_add_right_shift
int32_t cse_var_26 = (cse_var_27 + 1);
int32_t cse_var_25 = (((cse_var_28 + (ax2_outer * 512)) + (ax3_outer * 64)) + (ax2_inner * 2));
int32_t cse_var_24 = (((cse_var_28 + (ax2_outer * 128)) + (ax2_inner * 16)) + cse_var_27);
- int32_t __1 = ((int32_t)((((((int64_t)((int32_t*)conv)[cse_var_25]) + ((int64_t)((int32_t*)fused_nn_conv2d_constant_11)[cse_var_27])) * ((int64_t*)fused_nn_conv2d_add_cast_constant_12)[cse_var_27]) + ((int64_t*)fused_nn_conv2d_add_cast_multiply_constant_13)[cse_var_27]) >> ((int64_t*)fused_nn_conv2d_add_cast_multiply_add_constant_14)[cse_var_27])) - 128;
- int32_t __2 = (__1) < (127) ? (__1) : (127);
- int8_t __3 = (int8_t)((__2) > (-128) ? (__2) : (-128));
- int8_t __4 = (int8_t)127;
- int8_t __5 = (__3) < (__4) ? (__3) : (__4);
- int8_t __6 = (int8_t)-128;
- ((int16_t*)T_subtract)[cse_var_24] = (((int16_t)((__5) > (__6) ? (__5) : (__6))) - (int16_t)-128);
- int32_t __7 = ((int32_t)((((((int64_t)((int32_t*)conv)[(cse_var_25 + 1)]) + ((int64_t)((int32_t*)fused_nn_conv2d_constant_11)[cse_var_26])) * ((int64_t*)fused_nn_conv2d_add_cast_constant_12)[cse_var_26]) + ((int64_t*)fused_nn_conv2d_add_cast_multiply_constant_13)[cse_var_26]) >> ((int64_t*)fused_nn_conv2d_add_cast_multiply_add_constant_14)[cse_var_26])) - 128;
- int32_t __8 = (__7) < (127) ? (__7) : (127);
- int8_t __9 = (int8_t)((__8) > (-128) ? (__8) : (-128));
- int8_t __10 = (__9) < (__4) ? (__9) : (__4);
- ((int16_t*)T_subtract)[(cse_var_24 + 1)] = (((int16_t)((__10) > (__6) ? (__10) : (__6))) - (int16_t)-128);
+ int32_t scale_val = ((int32_t*)scale_19774001)[cse_var_27];
+ int32_t __1 = ((int32_t)((((((int64_t)((int32_t*)conv)[cse_var_25]) + ((int64_t)((int32_t*)fused_nn_conv2d_constant_11)[cse_var_27])) * ((int64_t*)fused_nn_conv2d_add_cast_constant_12)[cse_var_27]) + ((int64_t*)fused_nn_conv2d_add_cast_multiply_constant_13)[cse_var_27]) >> ((int64_t*)fused_nn_conv2d_add_cast_multiply_add_constant_14)[cse_var_27]));
+ //int32_t sum_0 = ((int32_t*)conv)[cse_var_25] + ((int32_t*)bias_8643672)[cse_var_27];
+ //int32_t __1 = (sum_0 * (int64_t) scale_val) >> 32;
+ //__1 = (__1 + 1) >> 1;
+ int32_t requant_0 = __ssat(__1 - 128, 8);
+ ((int16_t*)T_subtract)[cse_var_24] = (((int16_t) requant_0) - (int16_t)-128);
+ int32_t __7 = ((int32_t)((((((int64_t)((int32_t*)conv)[(cse_var_25 + 1)]) + ((int64_t)((int32_t*)fused_nn_conv2d_constant_11)[cse_var_26])) * ((int64_t*)fused_nn_conv2d_add_cast_constant_12)[cse_var_26]) + ((int64_t*)fused_nn_conv2d_add_cast_multiply_constant_13)[cse_var_26]) >> ((int64_t*)fused_nn_conv2d_add_cast_multiply_add_constant_14)[cse_var_26]));
+ int32_t requant_1 = __ssat(__7 - 128, 8);
+ ((int16_t*)T_subtract)[(cse_var_24 + 1)] = (((int16_t) requant_1) - (int16_t)-128);
}
for (int32_t ax2_inner_1 = 0; ax2_inner_1 < 8; ++ax2_inner_1) {
int32_t cse_var_33 = (ax0_ax1_outer_fused * 3072);