Skip to content

Commit

Permalink
Merge 2261a94 into 1984cad
Browse files Browse the repository at this point in the history
  • Loading branch information
@JackPumpkins
JackPumpkins committed Mar 19, 2020
2 parents 1984cad + 2261a94 commit bf831ed
Showing 1 changed file with 31 additions and 31 deletions.
62 changes: 31 additions & 31 deletions src/layer/arm/gemm_symm_int8.h
View file
Expand Up @@ -660,11 +660,11 @@ void int8kernel_m1(void* dst, int8_t* sa, int8_t* sb, int, int k, int n, int, fl
" // save \n"
" st1 {v8.s}[0], [%2]\n"
" add %2, %2, #4 \n"
" b m1_loopnd4_finish\n"
" b m1_loopnd4_finish%=\n"
" 7: \n"
" st1 {v8.4s}, [%2], #16 \n"

" m1_loopnd4_finish: \n"
" m1_loopnd4_finish%=: \n"
" subs %w8, %w8, #1 \n"
" mov %0, x8 \n"
" bne 9b \n"
Expand Down Expand Up @@ -696,7 +696,7 @@ void int8kernel_m1(void* dst, int8_t* sa, int8_t* sb, int, int k, int n, int, fl

if (n2 > 0) {
asm volatile(
"m1_nd2_start: \n"
"m1_nd2_start%=: \n"
" eor v8.16b, v8.16b, v8.16b \n"
" eor v9.16b, v9.16b, v9.16b \n"
" eor v10.16b, v10.16b, v10.16b\n"
Expand Down Expand Up @@ -822,11 +822,11 @@ void int8kernel_m1(void* dst, int8_t* sa, int8_t* sb, int, int k, int n, int, fl
" // save \n"
" st1 {v8.h}[0], [%2]\n"
" add %2, %2, #2 \n"
" b m1_loopnd2_finish\n"
" b m1_loopnd2_finish%=\n"
" 7: \n"
" st1 {v8.2s}, [%2], #8 \n"

" m1_loopnd2_finish: \n"
" m1_loopnd2_finish%=: \n"
" mov %0, x8 \n"
: "=r"(pa), // %0
"=r"(pb), // %1
Expand Down Expand Up @@ -857,7 +857,7 @@ void int8kernel_m1(void* dst, int8_t* sa, int8_t* sb, int, int k, int n, int, fl
if (n1 > 0) {

asm volatile (
"m1_nd1_start: \n"
"m1_nd1_start%=: \n"
" eor v8.16b, v8.16b, v8.16b \n"
" eor v9.16b, v9.16b, v9.16b \n"
" eor v10.16b, v10.16b, v10.16b\n"
Expand Down Expand Up @@ -973,11 +973,11 @@ void int8kernel_m1(void* dst, int8_t* sa, int8_t* sb, int, int k, int n, int, fl
" sqxtn v8.8b, v8.8h \n"
" // save \n"
" st1 {v8.b}[0], [%2]\n"
" b m1_finish \n"
" b m1_finish%= \n"

" 7: \n"
" st1 {v8.s}[0], [%2] \n"
" m1_finish: \n"
" m1_finish%=: \n"
" mov x0, #0 \n"
: "=r"(pa), // %0
"=r"(pb), // %1
Expand Down Expand Up @@ -1251,12 +1251,12 @@ void int8kernel_m2(void* dst, int8_t* sa, int8_t* sb, int, int k, int n, int ldc
" add %2, %2, #4 \n"
" st1 {v8.s}[1], [%3] \n"
" add %3, %3, #4 \n"
" b m2_loopnd4_finish \n"
" b m2_loopnd4_finish%= \n"
" 7: \n"
" st1 {v8.4s}, [%2], #16 \n"
" st1 {v9.4s}, [%3], #16 \n"

" m2_loopnd4_finish: \n"
" m2_loopnd4_finish%=: \n"
" subs %w9, %w9, #1 \n"
" mov %0, x8 \n"
" bne 9b \n"
Expand Down Expand Up @@ -1308,7 +1308,7 @@ void int8kernel_m2(void* dst, int8_t* sa, int8_t* sb, int, int k, int n, int ldc
"eor v22.16b, v22.16b, v22.16b \n"
"eor v23.16b, v23.16b, v23.16b \n"

"m2_nd2_start: \n"
"m2_nd2_start%=: \n"
" mov x8, %0 // PanelA \n"

" cmp %w5, #0 \n"
Expand Down Expand Up @@ -1465,15 +1465,15 @@ void int8kernel_m2(void* dst, int8_t* sa, int8_t* sb, int, int k, int n, int ldc
" sqxtn v8.8b, v8.8h \n"
" // save \n"
" st1 {v8.h}[0], [%2] \n"
" add %2, %2, #2 \n"
" add %2, %2, #2 \n"
" st1 {v8.h}[1], [%3] \n"
" add %3, %3, #2 \n"
" b m2_loopnd2_finish \n"
" add %3, %3, #2 \n"
" b m2_loopnd2_finish%= \n"

" 7:"
" st1 {v8.2s}, [%2], #8 \n"
" st1 {v12.2s}, [%3], #8 \n"
" m2_loopnd2_finish: \n"
" m2_loopnd2_finish%=: \n"
" mov %0, x8 \n"
: "=r"(pa), // %0
"=r"(pb), // %1
Expand Down Expand Up @@ -1521,7 +1521,7 @@ void int8kernel_m2(void* dst, int8_t* sa, int8_t* sb, int, int k, int n, int ldc
"eor v22.16b, v22.16b, v22.16b \n"
"eor v23.16b, v23.16b, v23.16b \n"

"m2_nd1_start: \n"
"m2_nd1_start%=: \n"
" cmp %w5, #0 \n"
" beq 1f // k <=7\n"

Expand Down Expand Up @@ -1654,12 +1654,12 @@ void int8kernel_m2(void* dst, int8_t* sa, int8_t* sb, int, int k, int n, int ldc
" // save \n"
" st1 {v8.b}[0], [%2] \n"
" st1 {v8.b}[1], [%3] \n"
" b m2_finish \n"
" b m2_finish%= \n"

" 7: \n"
" st1 {v8.s}[0], [%2] \n"
" st1 {v12.s}[0], [%3] \n"
" m2_finish: \n"
" m2_finish%=: \n"
" mov x0, #0 \n"
: "=r"(pa), // %0
"=r"(pb), // %1
Expand Down Expand Up @@ -2043,15 +2043,15 @@ void int8kernel_m4(void* dst, int8_t* sa, int8_t* sb, int, int k, int n, int ldc
" add %x4, %x4, #4 \n"
" st1 {v9.s}[1], [%5] \n"
" add %x5, %x5, #4 \n"
" b m4_loopnd4_finish \n"
" b m4_loopnd4_finish%= \n"

" 9: \n"
" st1 {v8.4s}, [%x2], #16 \n"
" st1 {v9.4s}, [%x3], #16 \n"
" st1 {v10.4s}, [%x4], #16 \n"
" st1 {v11.4s}, [%x5], #16 \n"

" m4_loopnd4_finish: \n"
" m4_loopnd4_finish%=: \n"
" subs %x11, %x11, #1 \n"
" mov %x0, x8 \n"
" bne 8b \n"
Expand Down Expand Up @@ -2107,11 +2107,11 @@ void int8kernel_m4(void* dst, int8_t* sa, int8_t* sb, int, int k, int n, int ldc
" eor v22.8b, v22.8b, v22.8b \n"
" eor v23.8b, v23.8b, v23.8b \n"

"m4_nd2_start: \n"
"m4_nd2_start%=: \n"
" mov x8, %x0 // PanelA \n"

" cmp %w7, #0 \n"
" beq 1f // k <= 7 \n"
" cmp %w7, #0 \n"
" beq 1f // k <= 7 \n"

" mov w20, %w7 \n"
" cmp %w6, #0 \n"
Expand Down Expand Up @@ -2344,20 +2344,20 @@ void int8kernel_m4(void* dst, int8_t* sa, int8_t* sb, int, int k, int n, int ldc
" st1 {v8.h}[0], [%2] \n"
" add %2, %2, #2 \n"
" st1 {v8.h}[1], [%3] \n"
" add %3, %3, #2 \n"
" add %3, %3, #2 \n"
" st1 {v16.h}[0], [%4] \n"
" add %4, %4, #2 \n"
" add %4, %4, #2 \n"
" st1 {v16.h}[1], [%5] \n"
" add %5, %5, #2 \n"
" b m4_loopnd2_finish \n"
" add %5, %5, #2 \n"
" b m4_loopnd2_finish%= \n"

" 7: \n"
" st1 {v8.2s}, [%2], #8 \n"
" st1 {v12.2s}, [%3], #8 \n"
" st1 {v16.2s}, [%4], #8 \n"
" st1 {v20.2s}, [%5], #8 \n"

" m4_loopnd2_finish: \n"
" m4_loopnd2_finish%=: \n"
" mov %0, x8 \n"
: "=r"(pa), // %0
"=r"(pb), // %1
Expand Down Expand Up @@ -2409,7 +2409,7 @@ void int8kernel_m4(void* dst, int8_t* sa, int8_t* sb, int, int k, int n, int ldc
" eor v22.8b, v22.8b, v22.8b \n"
" eor v23.8b, v23.8b, v23.8b \n"

"m4_n1_start: \n"
"m4_n1_start%=: \n"
" cmp %w7, #0 \n"
" beq 10f \n"

Expand Down Expand Up @@ -2587,7 +2587,7 @@ void int8kernel_m4(void* dst, int8_t* sa, int8_t* sb, int, int k, int n, int ldc
" st1 {v8.b}[1], [%3] \n"
" st1 {v8.b}[2], [%4] \n"
" st1 {v8.b}[3], [%5] \n"
" b m4_finish \n"
" b m4_finish%= \n"

" // no need to add the last output pointer\n"
" 16: \n"
Expand All @@ -2596,7 +2596,7 @@ void int8kernel_m4(void* dst, int8_t* sa, int8_t* sb, int, int k, int n, int ldc
" st1 {v16.s}[0], [%4] \n"
" st1 {v20.s}[0], [%5] \n"

" m4_finish: \n"
" m4_finish%=: \n"
" mov x0, #0 \n"
: "=r"(pa), // %0
"=r"(pb), // %1
Expand Down

0 comments on commit bf831ed

Please sign in to comment.