wangwei
/
MegEngine
Not watched
1
0
Fork 0
Code
Releases 31 Wiki Activity
Issues 0 Pull Requests 0 Datasets Model Cloudbrain
Browse Source

feat(dnn): fix Werror by adding macro 

GitOrigin-RevId: 1f5fe4d46a
tags/v0.5.0
Megvii Engine Team Xu Xinran 6 years ago
parent
d7bb62cfa1
commit
4f8e60801c
10 changed files with 200 additions and 200 deletions
Split View
Diff Options
Show Stats Download Patch File Download Diff File
  1. +8
    -8
      dnn/src/aarch64/matrix_mul/fp32/kernel_general_4x16.h
  2. +8
    -8
      dnn/src/aarch64/matrix_mul/fp32/kernel_general_8x12.h
  3. +8
    -8
      dnn/src/aarch64/matrix_mul/int16/kernel_12x8x1.h
  4. +18
    -18
      dnn/src/aarch64/matrix_mul/int8/kernel_4x4x16.h
  5. +44
    -44
      dnn/src/aarch64/matrix_mul/int8/kernel_8x8x8.h
  6. +4
    -4
      dnn/src/aarch64/matrix_mul/int8/kernel_mk4_4x4x16.h
  7. +18
    -18
      dnn/src/aarch64/matrix_mul/int8x8x16/kernel_4x4x16.h
  8. +44
    -44
      dnn/src/aarch64/matrix_mul/int8x8x16/kernel_8x8x8.h
  9. +44
    -44
      dnn/src/aarch64/matrix_mul/quint8/kernel_8x8x8.h
  10. +4
    -4
      dnn/src/arm_common/conv_bias/matmul_postprocess.h

+ 8
- 8
dnn/src/aarch64/matrix_mul/fp32/kernel_general_4x16.h View File

@@ -459,9 +459,9 @@ void sgemm_4x16_pack_A_n(float * outptr, const float * inptr, int ldin, int y0,
switch ((y + 3) - ymax) {
/* Everything falls through in here */
case 2:
inptr1 = zerobuff;
inptr1 = zerobuff; MEGDNN_FALLTHRU
case 1:
inptr2 = zerobuff;
inptr2 = zerobuff; MEGDNN_FALLTHRU
case 0:
inptr3 = zerobuff;
break;
@@ -478,9 +478,9 @@ void sgemm_4x16_pack_A_n(float * outptr, const float * inptr, int ldin, int y0,
if (y + 3 >= ymax) {
switch (y + 3 - ymax) {
case 2:
inptr1 = zerobuff;
inptr1 = zerobuff; MEGDNN_FALLTHRU
case 1:
inptr2 = zerobuff;
inptr2 = zerobuff; MEGDNN_FALLTHRU
case 0:
inptr3 = zerobuff;
break;
@@ -676,9 +676,9 @@ void sgemm_4x16_pack_B_t(float* out, const float* in, int ldin,
switch ((y + 3) - ymax) {
/* Everything falls through in here */
case 2:
inptr1 = zerobuff;
inptr1 = zerobuff; MEGDNN_FALLTHRU
case 1:
inptr2 = zerobuff;
inptr2 = zerobuff; MEGDNN_FALLTHRU
case 0:
inptr3 = zerobuff;
break;
@@ -696,9 +696,9 @@ void sgemm_4x16_pack_B_t(float* out, const float* in, int ldin,
switch ((y + 3) - ymax) {
/* Everything falls through in here */
case 2:
inptr1 = zerobuff;
inptr1 = zerobuff; MEGDNN_FALLTHRU
case 1:
inptr2 = zerobuff;
inptr2 = zerobuff; MEGDNN_FALLTHRU
case 0:
inptr3 = zerobuff;
break;


+ 8
- 8
dnn/src/aarch64/matrix_mul/fp32/kernel_general_8x12.h View File

@@ -964,9 +964,9 @@ void sgemm_8x12_pack_A_n(float* outptr, const float* inptr, int ldin, int y0,
switch ((y + 3) - ymax) {
/* Everything falls through in here */
case 2:
inptr1 = zerobuff;
inptr1 = zerobuff; MEGDNN_FALLTHRU
case 1:
inptr2 = zerobuff;
inptr2 = zerobuff; MEGDNN_FALLTHRU
case 0:
inptr3 = zerobuff;
break;
@@ -983,9 +983,9 @@ void sgemm_8x12_pack_A_n(float* outptr, const float* inptr, int ldin, int y0,
if (y + 3 >= ymax) {
switch (y + 3 - ymax) {
case 2:
inptr1 = zerobuff;
inptr1 = zerobuff; MEGDNN_FALLTHRU
case 1:
inptr2 = zerobuff;
inptr2 = zerobuff; MEGDNN_FALLTHRU
case 0:
inptr3 = zerobuff;
break;
@@ -1204,9 +1204,9 @@ void sgemm_8x12_pack_B_t(float* out, const float* in, int ldin,
switch ((y + 3) - ymax) {
/* Everything falls through in here */
case 2:
inptr1 = zerobuff;
inptr1 = zerobuff; MEGDNN_FALLTHRU
case 1:
inptr2 = zerobuff;
inptr2 = zerobuff; MEGDNN_FALLTHRU
case 0:
inptr3 = zerobuff;
break;
@@ -1224,9 +1224,9 @@ void sgemm_8x12_pack_B_t(float* out, const float* in, int ldin,
switch ((y + 3) - ymax) {
/* Everything falls through in here */
case 2:
inptr1 = zerobuff;
inptr1 = zerobuff; MEGDNN_FALLTHRU
case 1:
inptr2 = zerobuff;
inptr2 = zerobuff; MEGDNN_FALLTHRU
case 0:
inptr3 = zerobuff;
break;


+ 8
- 8
dnn/src/aarch64/matrix_mul/int16/kernel_12x8x1.h View File

@@ -975,9 +975,9 @@ static void gemm_s16_12x8x1_pack_A_n(int16_t* outptr, const int16_t* inptr,
if (y + 3 >= ymax) {
switch (y + 3 - ymax) {
case 2:
inptr1 = zerobuff;
inptr1 = zerobuff; MEGDNN_FALLTHRU
case 1:
inptr2 = zerobuff;
inptr2 = zerobuff; MEGDNN_FALLTHRU
case 0:
inptr3 = zerobuff;
break;
@@ -992,9 +992,9 @@ static void gemm_s16_12x8x1_pack_A_n(int16_t* outptr, const int16_t* inptr,
if (y + 3 >= ymax) {
switch (y + 3 - ymax) {
case 2:
inptr1 = zerobuff;
inptr1 = zerobuff; MEGDNN_FALLTHRU
case 1:
inptr2 = zerobuff;
inptr2 = zerobuff; MEGDNN_FALLTHRU
case 0:
inptr3 = zerobuff;
break;
@@ -1136,9 +1136,9 @@ static void gemm_s16_12x8x1_transpose_pack_B_n(int16_t* outptr,
if (y + 3 >= ymax) {
switch (y + 3 - ymax) {
case 2:
inptr1 = zerobuff;
inptr1 = zerobuff; MEGDNN_FALLTHRU
case 1:
inptr2 = zerobuff;
inptr2 = zerobuff; MEGDNN_FALLTHRU
case 0:
inptr3 = zerobuff;
break;
@@ -1153,9 +1153,9 @@ static void gemm_s16_12x8x1_transpose_pack_B_n(int16_t* outptr,
if (y + 3 >= ymax) {
switch (y + 3 - ymax) {
case 2:
inptr1 = zerobuff;
inptr1 = zerobuff; MEGDNN_FALLTHRU
case 1:
inptr2 = zerobuff;
inptr2 = zerobuff; MEGDNN_FALLTHRU
case 0:
inptr3 = zerobuff;
break;


+ 18
- 18
dnn/src/aarch64/matrix_mul/int8/kernel_4x4x16.h View File

@@ -717,9 +717,9 @@ static void gemm_s8_4x4_pack_A_n(dt_int8* outptr, const dt_int8* inptr,
if (y + 3 >= ymax) {
switch (y + 3 - ymax) {
case 2:
inptr1 = zerobuff;
inptr1 = zerobuff; MEGDNN_FALLTHRU
case 1:
inptr2 = zerobuff;
inptr2 = zerobuff; MEGDNN_FALLTHRU
case 0:
inptr3 = zerobuff;
break;
@@ -735,9 +735,9 @@ static void gemm_s8_4x4_pack_A_n(dt_int8* outptr, const dt_int8* inptr,
if (y + 3 >= ymax) {
switch (y + 3 - ymax) {
case 2:
inptr1 = zerobuff;
inptr1 = zerobuff; MEGDNN_FALLTHRU
case 1:
inptr2 = zerobuff;
inptr2 = zerobuff; MEGDNN_FALLTHRU
case 0:
inptr3 = zerobuff;
break;
@@ -778,19 +778,19 @@ static void gemm_s8_4x4_pack_B_n(dt_int8* out, const dt_int8* in, int ldin,
if (remain >= 0) {
switch (remain) {
case 7:
inptr0 = zerobuff;
inptr0 = zerobuff; MEGDNN_FALLTHRU
case 6:
inptr1 = zerobuff;
inptr1 = zerobuff; MEGDNN_FALLTHRU
case 5:
inptr2 = zerobuff;
inptr2 = zerobuff; MEGDNN_FALLTHRU
case 4:
inptr3 = zerobuff;
inptr3 = zerobuff; MEGDNN_FALLTHRU
case 3:
inptr4 = zerobuff;
inptr4 = zerobuff; MEGDNN_FALLTHRU
case 2:
inptr5 = zerobuff;
inptr5 = zerobuff; MEGDNN_FALLTHRU
case 1:
inptr6 = zerobuff;
inptr6 = zerobuff; MEGDNN_FALLTHRU
case 0:
inptr7 = zerobuff;
break;
@@ -809,19 +809,19 @@ static void gemm_s8_4x4_pack_B_n(dt_int8* out, const dt_int8* in, int ldin,
if (remain >= 0) {
switch (remain) {
case 7:
inptr0 = zerobuff;
inptr0 = zerobuff; MEGDNN_FALLTHRU
case 6:
inptr1 = zerobuff;
inptr1 = zerobuff; MEGDNN_FALLTHRU
case 5:
inptr2 = zerobuff;
inptr2 = zerobuff; MEGDNN_FALLTHRU
case 4:
inptr3 = zerobuff;
inptr3 = zerobuff; MEGDNN_FALLTHRU
case 3:
inptr4 = zerobuff;
inptr4 = zerobuff; MEGDNN_FALLTHRU
case 2:
inptr5 = zerobuff;
inptr5 = zerobuff; MEGDNN_FALLTHRU
case 1:
inptr6 = zerobuff;
inptr6 = zerobuff; MEGDNN_FALLTHRU
case 0:
inptr7 = zerobuff;
break;


+ 44
- 44
dnn/src/aarch64/matrix_mul/int8/kernel_8x8x8.h View File

@@ -992,9 +992,9 @@ static void gemm_s8_8x8_pack_A_n(int8_t* outptr, const int8_t* inptr, int ldin,
if (y + 3 >= ymax) {
switch (y + 3 - ymax) {
case 2:
inptr1 = zerobuff;
inptr1 = zerobuff; MEGDNN_FALLTHRU
case 1:
inptr2 = zerobuff;
inptr2 = zerobuff; MEGDNN_FALLTHRU
case 0:
inptr3 = zerobuff;
break;
@@ -1010,9 +1010,9 @@ static void gemm_s8_8x8_pack_A_n(int8_t* outptr, const int8_t* inptr, int ldin,
if (y + 3 >= ymax) {
switch (y + 3 - ymax) {
case 2:
inptr1 = zerobuff;
inptr1 = zerobuff; MEGDNN_FALLTHRU
case 1:
inptr2 = zerobuff;
inptr2 = zerobuff; MEGDNN_FALLTHRU
case 0:
inptr3 = zerobuff;
break;
@@ -1064,17 +1064,17 @@ static void gemm_s8_8x8_transpose_pack_A_n(int8_t* out, const int8_t* in,
if (k + 7 >= kmax) {
switch (k + 7 - kmax) {
case 6:
inptr1 = zerobuff;
inptr1 = zerobuff; MEGDNN_FALLTHRU
case 5:
inptr2 = zerobuff;
inptr2 = zerobuff; MEGDNN_FALLTHRU
case 4:
inptr3 = zerobuff;
inptr3 = zerobuff; MEGDNN_FALLTHRU
case 3:
inptr4 = zerobuff;
inptr4 = zerobuff; MEGDNN_FALLTHRU
case 2:
inptr5 = zerobuff;
inptr5 = zerobuff; MEGDNN_FALLTHRU
case 1:
inptr6 = zerobuff;
inptr6 = zerobuff; MEGDNN_FALLTHRU
case 0:
inptr7 = zerobuff;
break;
@@ -1092,17 +1092,17 @@ static void gemm_s8_8x8_transpose_pack_A_n(int8_t* out, const int8_t* in,
if (k + 7 >= kmax) {
switch (k + 7 - kmax) {
case 6:
inptr1 = zerobuff;
inptr1 = zerobuff; MEGDNN_FALLTHRU
case 5:
inptr2 = zerobuff;
inptr2 = zerobuff; MEGDNN_FALLTHRU
case 4:
inptr3 = zerobuff;
inptr3 = zerobuff; MEGDNN_FALLTHRU
case 3:
inptr4 = zerobuff;
inptr4 = zerobuff; MEGDNN_FALLTHRU
case 2:
inptr5 = zerobuff;
inptr5 = zerobuff; MEGDNN_FALLTHRU
case 1:
inptr6 = zerobuff;
inptr6 = zerobuff; MEGDNN_FALLTHRU
case 0:
inptr7 = zerobuff;
break;
@@ -1120,17 +1120,17 @@ static void gemm_s8_8x8_transpose_pack_A_n(int8_t* out, const int8_t* in,
if (k + 7 >= kmax) {
switch (k + 7 - kmax) {
case 6:
inptr1 = zerobuff;
inptr1 = zerobuff; MEGDNN_FALLTHRU
case 5:
inptr2 = zerobuff;
inptr2 = zerobuff; MEGDNN_FALLTHRU
case 4:
inptr3 = zerobuff;
inptr3 = zerobuff; MEGDNN_FALLTHRU
case 3:
inptr4 = zerobuff;
inptr4 = zerobuff; MEGDNN_FALLTHRU
case 2:
inptr5 = zerobuff;
inptr5 = zerobuff; MEGDNN_FALLTHRU
case 1:
inptr6 = zerobuff;
inptr6 = zerobuff; MEGDNN_FALLTHRU
case 0:
inptr7 = zerobuff;
break;
@@ -1187,17 +1187,17 @@ static void gemm_s8_8x8_pack_B_n(int8_t* out, const int8_t* in, int ldin,
if (k + 7 >= kmax) {
switch (k + 7 - kmax) {
case 6:
inptr1 = zerobuff;
inptr1 = zerobuff; MEGDNN_FALLTHRU
case 5:
inptr2 = zerobuff;
inptr2 = zerobuff; MEGDNN_FALLTHRU
case 4:
inptr3 = zerobuff;
inptr3 = zerobuff; MEGDNN_FALLTHRU
case 3:
inptr4 = zerobuff;
inptr4 = zerobuff; MEGDNN_FALLTHRU
case 2:
inptr5 = zerobuff;
inptr5 = zerobuff; MEGDNN_FALLTHRU
case 1:
inptr6 = zerobuff;
inptr6 = zerobuff; MEGDNN_FALLTHRU
case 0:
inptr7 = zerobuff;
break;
@@ -1216,17 +1216,17 @@ static void gemm_s8_8x8_pack_B_n(int8_t* out, const int8_t* in, int ldin,
if (k + 7 >= kmax) {
switch (k + 7 - kmax) {
case 6:
inptr1 = zerobuff;
inptr1 = zerobuff; MEGDNN_FALLTHRU
case 5:
inptr2 = zerobuff;
inptr2 = zerobuff; MEGDNN_FALLTHRU
case 4:
inptr3 = zerobuff;
inptr3 = zerobuff; MEGDNN_FALLTHRU
case 3:
inptr4 = zerobuff;
inptr4 = zerobuff; MEGDNN_FALLTHRU
case 2:
inptr5 = zerobuff;
inptr5 = zerobuff; MEGDNN_FALLTHRU
case 1:
inptr6 = zerobuff;
inptr6 = zerobuff; MEGDNN_FALLTHRU
case 0:
inptr7 = zerobuff;
break;
@@ -1245,17 +1245,17 @@ static void gemm_s8_8x8_pack_B_n(int8_t* out, const int8_t* in, int ldin,
if (k + 7 >= kmax) {
switch (k + 7 - kmax) {
case 6:
inptr1 = zerobuff;
inptr1 = zerobuff; MEGDNN_FALLTHRU
case 5:
inptr2 = zerobuff;
inptr2 = zerobuff; MEGDNN_FALLTHRU
case 4:
inptr3 = zerobuff;
inptr3 = zerobuff; MEGDNN_FALLTHRU
case 3:
inptr4 = zerobuff;
inptr4 = zerobuff; MEGDNN_FALLTHRU
case 2:
inptr5 = zerobuff;
inptr5 = zerobuff; MEGDNN_FALLTHRU
case 1:
inptr6 = zerobuff;
inptr6 = zerobuff; MEGDNN_FALLTHRU
case 0:
inptr7 = zerobuff;
break;
@@ -1332,9 +1332,9 @@ static void gemm_s8_8x8_transpose_pack_B_n(int8_t* outptr, const int8_t* inptr,
if (y + 3 >= ymax) {
switch (y + 3 - ymax) {
case 2:
inptr1 = zerobuff;
inptr1 = zerobuff; MEGDNN_FALLTHRU
case 1:
inptr2 = zerobuff;
inptr2 = zerobuff; MEGDNN_FALLTHRU
case 0:
inptr3 = zerobuff;
break;
@@ -1351,9 +1351,9 @@ static void gemm_s8_8x8_transpose_pack_B_n(int8_t* outptr, const int8_t* inptr,
if (y + 3 >= ymax) {
switch (y + 3 - ymax) {
case 2:
inptr1 = zerobuff;
inptr1 = zerobuff; MEGDNN_FALLTHRU
case 1:
inptr2 = zerobuff;
inptr2 = zerobuff; MEGDNN_FALLTHRU
case 0:
inptr3 = zerobuff;
break;


+ 4
- 4
dnn/src/aarch64/matrix_mul/int8/kernel_mk4_4x4x16.h View File

@@ -846,9 +846,9 @@ static void gemm_mk4_s8_4x4_pack_B(dt_int8* out, const dt_int8* in, int ldin,
if (k + 3 >= ICB) {
switch (k + 3 - ICB) {
case 2:
inptr1 = zerobuff;
inptr1 = zerobuff; MEGDNN_FALLTHRU
case 1:
inptr2 = zerobuff;
inptr2 = zerobuff; MEGDNN_FALLTHRU
case 0:
inptr3 = zerobuff;
break;
@@ -863,9 +863,9 @@ static void gemm_mk4_s8_4x4_pack_B(dt_int8* out, const dt_int8* in, int ldin,
if (k + 3 >= ICB) {
switch (k + 3 - ICB) {
case 2:
inptr1 = zerobuff;
inptr1 = zerobuff; MEGDNN_FALLTHRU
case 1:
inptr2 = zerobuff;
inptr2 = zerobuff; MEGDNN_FALLTHRU
case 0:
inptr3 = zerobuff;
break;


+ 18
- 18
dnn/src/aarch64/matrix_mul/int8x8x16/kernel_4x4x16.h View File

@@ -292,9 +292,9 @@ static void gemm_s8x8x16_4x4_pack_A_n(dt_int8* outptr, const dt_int8* inptr,
if (y + 3 >= ymax) {
switch (y + 3 - ymax) {
case 2:
inptr1 = zerobuff;
inptr1 = zerobuff; MEGDNN_FALLTHRU
case 1:
inptr2 = zerobuff;
inptr2 = zerobuff; MEGDNN_FALLTHRU
case 0:
inptr3 = zerobuff;
break;
@@ -309,9 +309,9 @@ static void gemm_s8x8x16_4x4_pack_A_n(dt_int8* outptr, const dt_int8* inptr,
if (y + 3 >= ymax) {
switch (y + 3 - ymax) {
case 2:
inptr1 = zerobuff;
inptr1 = zerobuff; MEGDNN_FALLTHRU
case 1:
inptr2 = zerobuff;
inptr2 = zerobuff; MEGDNN_FALLTHRU
case 0:
inptr3 = zerobuff;
break;
@@ -362,19 +362,19 @@ static void gemm_s8x8x16_4x4_pack_B_n(dt_int8* out, const dt_int8* in, int ldin,
if (remain >= 0) {
switch (remain) {
case 7:
inptr0 = zerobuff;
inptr0 = zerobuff; MEGDNN_FALLTHRU
case 6:
inptr1 = zerobuff;
inptr1 = zerobuff; MEGDNN_FALLTHRU
case 5:
inptr2 = zerobuff;
inptr2 = zerobuff; MEGDNN_FALLTHRU
case 4:
inptr3 = zerobuff;
inptr3 = zerobuff; MEGDNN_FALLTHRU
case 3:
inptr4 = zerobuff;
inptr4 = zerobuff; MEGDNN_FALLTHRU
case 2:
inptr5 = zerobuff;
inptr5 = zerobuff; MEGDNN_FALLTHRU
case 1:
inptr6 = zerobuff;
inptr6 = zerobuff; MEGDNN_FALLTHRU
case 0:
inptr7 = zerobuff;
break;
@@ -393,19 +393,19 @@ static void gemm_s8x8x16_4x4_pack_B_n(dt_int8* out, const dt_int8* in, int ldin,
if (remain >= 0) {
switch (remain) {
case 7:
inptr0 = zerobuff;
inptr0 = zerobuff; MEGDNN_FALLTHRU
case 6:
inptr1 = zerobuff;
inptr1 = zerobuff; MEGDNN_FALLTHRU
case 5:
inptr2 = zerobuff;
inptr2 = zerobuff; MEGDNN_FALLTHRU
case 4:
inptr3 = zerobuff;
inptr3 = zerobuff; MEGDNN_FALLTHRU
case 3:
inptr4 = zerobuff;
inptr4 = zerobuff; MEGDNN_FALLTHRU
case 2:
inptr5 = zerobuff;
inptr5 = zerobuff; MEGDNN_FALLTHRU
case 1:
inptr6 = zerobuff;
inptr6 = zerobuff; MEGDNN_FALLTHRU
case 0:
inptr7 = zerobuff;
break;


+ 44
- 44
dnn/src/aarch64/matrix_mul/int8x8x16/kernel_8x8x8.h View File

@@ -918,9 +918,9 @@ static void gemm_s8x8x16_8x8_pack_A_n(dt_int8* outptr, const dt_int8* inptr,
if (y + 3 >= ymax) {
switch (y + 3 - ymax) {
case 2:
inptr1 = zerobuff;
inptr1 = zerobuff; MEGDNN_FALLTHRU
case 1:
inptr2 = zerobuff;
inptr2 = zerobuff; MEGDNN_FALLTHRU
case 0:
inptr3 = zerobuff;
break;
@@ -936,9 +936,9 @@ static void gemm_s8x8x16_8x8_pack_A_n(dt_int8* outptr, const dt_int8* inptr,
if (y + 3 >= ymax) {
switch (y + 3 - ymax) {
case 2:
inptr1 = zerobuff;
inptr1 = zerobuff; MEGDNN_FALLTHRU
case 1:
inptr2 = zerobuff;
inptr2 = zerobuff; MEGDNN_FALLTHRU
case 0:
inptr3 = zerobuff;
break;
@@ -991,17 +991,17 @@ static void gemm_s8x8x16_8x8_transpose_pack_A_n(dt_int8* out, const dt_int8* in,
if (k + 7 >= kmax) {
switch (k + 7 - kmax) {
case 6:
inptr1 = zerobuff;
inptr1 = zerobuff; MEGDNN_FALLTHRU
case 5:
inptr2 = zerobuff;
inptr2 = zerobuff; MEGDNN_FALLTHRU
case 4:
inptr3 = zerobuff;
inptr3 = zerobuff; MEGDNN_FALLTHRU
case 3:
inptr4 = zerobuff;
inptr4 = zerobuff; MEGDNN_FALLTHRU
case 2:
inptr5 = zerobuff;
inptr5 = zerobuff; MEGDNN_FALLTHRU
case 1:
inptr6 = zerobuff;
inptr6 = zerobuff; MEGDNN_FALLTHRU
case 0:
inptr7 = zerobuff;
break;
@@ -1019,17 +1019,17 @@ static void gemm_s8x8x16_8x8_transpose_pack_A_n(dt_int8* out, const dt_int8* in,
if (k + 7 >= kmax) {
switch (k + 7 - kmax) {
case 6:
inptr1 = zerobuff;
inptr1 = zerobuff; MEGDNN_FALLTHRU
case 5:
inptr2 = zerobuff;
inptr2 = zerobuff; MEGDNN_FALLTHRU
case 4:
inptr3 = zerobuff;
inptr3 = zerobuff; MEGDNN_FALLTHRU
case 3:
inptr4 = zerobuff;
inptr4 = zerobuff; MEGDNN_FALLTHRU
case 2:
inptr5 = zerobuff;
inptr5 = zerobuff; MEGDNN_FALLTHRU
case 1:
inptr6 = zerobuff;
inptr6 = zerobuff; MEGDNN_FALLTHRU
case 0:
inptr7 = zerobuff;
break;
@@ -1047,17 +1047,17 @@ static void gemm_s8x8x16_8x8_transpose_pack_A_n(dt_int8* out, const dt_int8* in,
if (k + 7 >= kmax) {
switch (k + 7 - kmax) {
case 6:
inptr1 = zerobuff;
inptr1 = zerobuff; MEGDNN_FALLTHRU
case 5:
inptr2 = zerobuff;
inptr2 = zerobuff; MEGDNN_FALLTHRU
case 4:
inptr3 = zerobuff;
inptr3 = zerobuff; MEGDNN_FALLTHRU
case 3:
inptr4 = zerobuff;
inptr4 = zerobuff; MEGDNN_FALLTHRU
case 2:
inptr5 = zerobuff;
inptr5 = zerobuff; MEGDNN_FALLTHRU
case 1:
inptr6 = zerobuff;
inptr6 = zerobuff; MEGDNN_FALLTHRU
case 0:
inptr7 = zerobuff;
break;
@@ -1113,17 +1113,17 @@ static void gemm_s8x8x16_8x8_pack_B_n(dt_int8* out, const dt_int8* in, int ldin,
if (k + 7 >= kmax) {
switch (k + 7 - kmax) {
case 6:
inptr1 = zerobuff;
inptr1 = zerobuff; MEGDNN_FALLTHRU
case 5:
inptr2 = zerobuff;
inptr2 = zerobuff; MEGDNN_FALLTHRU
case 4:
inptr3 = zerobuff;
inptr3 = zerobuff; MEGDNN_FALLTHRU
case 3:
inptr4 = zerobuff;
inptr4 = zerobuff; MEGDNN_FALLTHRU
case 2:
inptr5 = zerobuff;
inptr5 = zerobuff; MEGDNN_FALLTHRU
case 1:
inptr6 = zerobuff;
inptr6 = zerobuff; MEGDNN_FALLTHRU
case 0:
inptr7 = zerobuff;
break;
@@ -1142,17 +1142,17 @@ static void gemm_s8x8x16_8x8_pack_B_n(dt_int8* out, const dt_int8* in, int ldin,
if (k + 7 >= kmax) {
switch (k + 7 - kmax) {
case 6:
inptr1 = zerobuff;
inptr1 = zerobuff; MEGDNN_FALLTHRU
case 5:
inptr2 = zerobuff;
inptr2 = zerobuff; MEGDNN_FALLTHRU
case 4:
inptr3 = zerobuff;
inptr3 = zerobuff; MEGDNN_FALLTHRU
case 3:
inptr4 = zerobuff;
inptr4 = zerobuff; MEGDNN_FALLTHRU
case 2:
inptr5 = zerobuff;
inptr5 = zerobuff; MEGDNN_FALLTHRU
case 1:
inptr6 = zerobuff;
inptr6 = zerobuff; MEGDNN_FALLTHRU
case 0:
inptr7 = zerobuff;
break;
@@ -1171,17 +1171,17 @@ static void gemm_s8x8x16_8x8_pack_B_n(dt_int8* out, const dt_int8* in, int ldin,
if (k + 7 >= kmax) {
switch (k + 7 - kmax) {
case 6:
inptr1 = zerobuff;
inptr1 = zerobuff; MEGDNN_FALLTHRU
case 5:
inptr2 = zerobuff;
inptr2 = zerobuff; MEGDNN_FALLTHRU
case 4:
inptr3 = zerobuff;
inptr3 = zerobuff; MEGDNN_FALLTHRU
case 3:
inptr4 = zerobuff;
inptr4 = zerobuff; MEGDNN_FALLTHRU
case 2:
inptr5 = zerobuff;
inptr5 = zerobuff; MEGDNN_FALLTHRU
case 1:
inptr6 = zerobuff;
inptr6 = zerobuff; MEGDNN_FALLTHRU
case 0:
inptr7 = zerobuff;
break;
@@ -1259,9 +1259,9 @@ static void gemm_s8x8x16_8x8_transpose_pack_B_n(dt_int8* outptr,
if (y + 3 >= ymax) {
switch (y + 3 - ymax) {
case 2:
inptr1 = zerobuff;
inptr1 = zerobuff; MEGDNN_FALLTHRU
case 1:
inptr2 = zerobuff;
inptr2 = zerobuff; MEGDNN_FALLTHRU
case 0:
inptr3 = zerobuff;
break;
@@ -1278,9 +1278,9 @@ static void gemm_s8x8x16_8x8_transpose_pack_B_n(dt_int8* outptr,
if (y + 3 >= ymax) {
switch (y + 3 - ymax) {
case 2:
inptr1 = zerobuff;
inptr1 = zerobuff; MEGDNN_FALLTHRU
case 1:
inptr2 = zerobuff;
inptr2 = zerobuff; MEGDNN_FALLTHRU
case 0:
inptr3 = zerobuff;
break;


+ 44
- 44
dnn/src/aarch64/matrix_mul/quint8/kernel_8x8x8.h View File

@@ -1011,9 +1011,9 @@ static void gemm_u8_8x8_pack_A_n(dt_uint8* outptr, const dt_uint8* inptr,
if (y + 3 >= ymax) {
switch (y + 3 - ymax) {
case 2:
inptr1 = zerobuff;
inptr1 = zerobuff; MEGDNN_FALLTHRU
case 1:
inptr2 = zerobuff;
inptr2 = zerobuff; MEGDNN_FALLTHRU
case 0:
inptr3 = zerobuff;
break;
@@ -1029,9 +1029,9 @@ static void gemm_u8_8x8_pack_A_n(dt_uint8* outptr, const dt_uint8* inptr,
if (y + 3 >= ymax) {
switch (y + 3 - ymax) {
case 2:
inptr1 = zerobuff;
inptr1 = zerobuff; MEGDNN_FALLTHRU
case 1:
inptr2 = zerobuff;
inptr2 = zerobuff; MEGDNN_FALLTHRU
case 0:
inptr3 = zerobuff;
break;
@@ -1084,17 +1084,17 @@ static void gemm_u8_8x8_transpose_pack_A_n(dt_uint8* out, const dt_uint8* in,
if (k + 7 >= kmax) {
switch (k + 7 - kmax) {
case 6:
inptr1 = zerobuff;
inptr1 = zerobuff; MEGDNN_FALLTHRU
case 5:
inptr2 = zerobuff;
inptr2 = zerobuff; MEGDNN_FALLTHRU
case 4:
inptr3 = zerobuff;
inptr3 = zerobuff; MEGDNN_FALLTHRU
case 3:
inptr4 = zerobuff;
inptr4 = zerobuff; MEGDNN_FALLTHRU
case 2:
inptr5 = zerobuff;
inptr5 = zerobuff; MEGDNN_FALLTHRU
case 1:
inptr6 = zerobuff;
inptr6 = zerobuff; MEGDNN_FALLTHRU
case 0:
inptr7 = zerobuff;
break;
@@ -1112,17 +1112,17 @@ static void gemm_u8_8x8_transpose_pack_A_n(dt_uint8* out, const dt_uint8* in,
if (k + 7 >= kmax) {
switch (k + 7 - kmax) {
case 6:
inptr1 = zerobuff;
inptr1 = zerobuff; MEGDNN_FALLTHRU
case 5:
inptr2 = zerobuff;
inptr2 = zerobuff; MEGDNN_FALLTHRU
case 4:
inptr3 = zerobuff;
inptr3 = zerobuff; MEGDNN_FALLTHRU
case 3:
inptr4 = zerobuff;
inptr4 = zerobuff; MEGDNN_FALLTHRU
case 2:
inptr5 = zerobuff;
inptr5 = zerobuff; MEGDNN_FALLTHRU
case 1:
inptr6 = zerobuff;
inptr6 = zerobuff; MEGDNN_FALLTHRU
case 0:
inptr7 = zerobuff;
break;
@@ -1140,17 +1140,17 @@ static void gemm_u8_8x8_transpose_pack_A_n(dt_uint8* out, const dt_uint8* in,
if (k + 7 >= kmax) {
switch (k + 7 - kmax) {
case 6:
inptr1 = zerobuff;
inptr1 = zerobuff; MEGDNN_FALLTHRU
case 5:
inptr2 = zerobuff;
inptr2 = zerobuff; MEGDNN_FALLTHRU
case 4:
inptr3 = zerobuff;
inptr3 = zerobuff; MEGDNN_FALLTHRU
case 3:
inptr4 = zerobuff;
inptr4 = zerobuff; MEGDNN_FALLTHRU
case 2:
inptr5 = zerobuff;
inptr5 = zerobuff; MEGDNN_FALLTHRU
case 1:
inptr6 = zerobuff;
inptr6 = zerobuff; MEGDNN_FALLTHRU
case 0:
inptr7 = zerobuff;
break;
@@ -1208,17 +1208,17 @@ static void gemm_u8_8x8_pack_B_n(dt_uint8* out, const dt_uint8* in, int ldin,
if (k + 7 >= kmax) {
switch (k + 7 - kmax) {
case 6:
inptr1 = zerobuff;
inptr1 = zerobuff; MEGDNN_FALLTHRU
case 5:
inptr2 = zerobuff;
inptr2 = zerobuff; MEGDNN_FALLTHRU
case 4:
inptr3 = zerobuff;
inptr3 = zerobuff; MEGDNN_FALLTHRU
case 3:
inptr4 = zerobuff;
inptr4 = zerobuff; MEGDNN_FALLTHRU
case 2:
inptr5 = zerobuff;
inptr5 = zerobuff; MEGDNN_FALLTHRU
case 1:
inptr6 = zerobuff;
inptr6 = zerobuff; MEGDNN_FALLTHRU
case 0:
inptr7 = zerobuff;
break;
@@ -1237,17 +1237,17 @@ static void gemm_u8_8x8_pack_B_n(dt_uint8* out, const dt_uint8* in, int ldin,
if (k + 7 >= kmax) {
switch (k + 7 - kmax) {
case 6:
inptr1 = zerobuff;
inptr1 = zerobuff; MEGDNN_FALLTHRU
case 5:
inptr2 = zerobuff;
inptr2 = zerobuff; MEGDNN_FALLTHRU
case 4:
inptr3 = zerobuff;
inptr3 = zerobuff; MEGDNN_FALLTHRU
case 3:
inptr4 = zerobuff;
inptr4 = zerobuff; MEGDNN_FALLTHRU
case 2:
inptr5 = zerobuff;
inptr5 = zerobuff; MEGDNN_FALLTHRU
case 1:
inptr6 = zerobuff;
inptr6 = zerobuff; MEGDNN_FALLTHRU
case 0:
inptr7 = zerobuff;
break;
@@ -1266,17 +1266,17 @@ static void gemm_u8_8x8_pack_B_n(dt_uint8* out, const dt_uint8* in, int ldin,
if (k + 7 >= kmax) {
switch (k + 7 - kmax) {
case 6:
inptr1 = zerobuff;
inptr1 = zerobuff; MEGDNN_FALLTHRU
case 5:
inptr2 = zerobuff;
inptr2 = zerobuff; MEGDNN_FALLTHRU
case 4:
inptr3 = zerobuff;
inptr3 = zerobuff; MEGDNN_FALLTHRU
case 3:
inptr4 = zerobuff;
inptr4 = zerobuff; MEGDNN_FALLTHRU
case 2:
inptr5 = zerobuff;
inptr5 = zerobuff; MEGDNN_FALLTHRU
case 1:
inptr6 = zerobuff;
inptr6 = zerobuff; MEGDNN_FALLTHRU
case 0:
inptr7 = zerobuff;
break;
@@ -1354,9 +1354,9 @@ static void gemm_u8_8x8_transpose_pack_B_n(dt_uint8* outptr,
if (y + 3 >= ymax) {
switch (y + 3 - ymax) {
case 2:
inptr1 = zerobuff;
inptr1 = zerobuff; MEGDNN_FALLTHRU
case 1:
inptr2 = zerobuff;
inptr2 = zerobuff; MEGDNN_FALLTHRU
case 0:
inptr3 = zerobuff;
break;
@@ -1373,9 +1373,9 @@ static void gemm_u8_8x8_transpose_pack_B_n(dt_uint8* outptr,
if (y + 3 >= ymax) {
switch (y + 3 - ymax) {
case 2:
inptr1 = zerobuff;
inptr1 = zerobuff; MEGDNN_FALLTHRU
case 1:
inptr2 = zerobuff;
inptr2 = zerobuff; MEGDNN_FALLTHRU
case 0:
inptr3 = zerobuff;
break;


+ 4
- 4
dnn/src/arm_common/conv_bias/matmul_postprocess.h View File

@@ -27,9 +27,9 @@ namespace arm_common {
vreinterpret_s32_s8(vres), idx / 4); \
break; \
case 3: \
vst1_lane_s8(C + 2, vres, idx + 2); \
vst1_lane_s8(C + 2, vres, idx + 2); MEGDNN_FALLTHRU\
case 2: \
vst1_lane_s8(C + 1, vres, idx + 1); \
vst1_lane_s8(C + 1, vres, idx + 1); MEGDNN_FALLTHRU\
case 1: \
vst1_lane_s8(C + 0, vres, idx + 0); \
break; \
@@ -44,9 +44,9 @@ namespace arm_common {
vreinterpret_s32_u8(vres), idx / 4); \
break; \
case 3: \
vst1_lane_u8(C + 2, vres, idx + 2); \
vst1_lane_u8(C + 2, vres, idx + 2); MEGDNN_FALLTHRU\
case 2: \
vst1_lane_u8(C + 1, vres, idx + 1); \
vst1_lane_u8(C + 1, vres, idx + 1); MEGDNN_FALLTHRU\
case 1: \
vst1_lane_u8(C + 0, vres, idx + 0); \
break; \


Write Preview
Loading…
Cancel
Save