Browse Source

Do not use vsub to clear the register values since it doesn't work with non-normal numbers.

tags/v0.2.16.rc1
Yichao Yu 10 years ago
parent
commit
594b9f4c73
33 changed files with 283 additions and 242 deletions
  1. +4
    -4
      kernel/arm/asum_vfp.S
  2. +4
    -4
      kernel/arm/cdot_vfp.S
  3. +4
    -4
      kernel/arm/cgemm_kernel_2x2_vfp.S
  4. +6
    -4
      kernel/arm/cgemm_kernel_2x2_vfpv3.S
  5. +4
    -4
      kernel/arm/cgemv_n_vfp.S
  6. +12
    -12
      kernel/arm/cgemv_t_vfp.S
  7. +22
    -22
      kernel/arm/ctrmm_kernel_2x2_vfp.S
  8. +6
    -4
      kernel/arm/ctrmm_kernel_2x2_vfpv3.S
  9. +2
    -2
      kernel/arm/ddot_vfp.S
  10. +6
    -6
      kernel/arm/dgemm_kernel_4x2_vfp.S
  11. +9
    -9
      kernel/arm/dgemm_kernel_4x4_vfpv3.S
  12. +6
    -6
      kernel/arm/dtrmm_kernel_4x2_vfp.S
  13. +9
    -9
      kernel/arm/dtrmm_kernel_4x4_vfpv3.S
  14. +8
    -8
      kernel/arm/gemv_n_vfp.S
  15. +8
    -8
      kernel/arm/gemv_n_vfpv3.S
  16. +12
    -12
      kernel/arm/gemv_t_vfp.S
  17. +12
    -12
      kernel/arm/gemv_t_vfpv3.S
  18. +2
    -2
      kernel/arm/iamax_vfp.S
  19. +4
    -4
      kernel/arm/nrm2_vfp.S
  20. +7
    -2
      kernel/arm/nrm2_vfpv3.S
  21. +4
    -4
      kernel/arm/sdot_vfp.S
  22. +6
    -6
      kernel/arm/sgemm_kernel_4x2_vfp.S
  23. +32
    -17
      kernel/arm/sgemm_kernel_4x4_vfpv3.S
  24. +6
    -6
      kernel/arm/strmm_kernel_4x2_vfp.S
  25. +33
    -16
      kernel/arm/strmm_kernel_4x4_vfpv3.S
  26. +4
    -4
      kernel/arm/zdot_vfp.S
  27. +4
    -4
      kernel/arm/zgemm_kernel_2x2_vfp.S
  28. +4
    -4
      kernel/arm/zgemm_kernel_2x2_vfpv3.S
  29. +4
    -4
      kernel/arm/zgemv_n_vfp.S
  30. +12
    -12
      kernel/arm/zgemv_t_vfp.S
  31. +22
    -22
      kernel/arm/ztrmm_kernel_2x2_vfp.S
  32. +4
    -4
      kernel/arm/ztrmm_kernel_2x2_vfpv3.S
  33. +1
    -1
      kernel/generic/zgemmkernel_2x2.c

+ 4
- 4
kernel/arm/asum_vfp.S View File

@@ -368,11 +368,11 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
.align 5

#if defined(DOUBLE)
vsub.f64 d0 , d0 , d0
vsub.f64 d1 , d1 , d1
vldr.f64 d0 , =0
vldr.f64 d1 , =0
#else
vsub.f32 s0 , s0 , s0
vsub.f32 s1 , s1 , s1
vldr.f32 s0 , =0
vldr.f32 s1 , =0
#endif

cmp N, #0


+ 4
- 4
kernel/arm/cdot_vfp.S View File

@@ -188,10 +188,10 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
mov Y, OLD_Y
ldr INC_Y, OLD_INC_Y

vsub.f32 s0 , s0 , s0
vsub.f32 s1 , s1 , s1
vsub.f32 s2 , s2 , s2
vsub.f32 s3 , s3 , s3
vldr.f32 s0 , =0
vldr.f32 s1 , =0
vldr.f32 s2 , =0
vldr.f32 s3 , =0

cmp N, #0
ble cdot_kernel_L999


+ 4
- 4
kernel/arm/cgemm_kernel_2x2_vfp.S View File

@@ -138,7 +138,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT2x2

vsub.f32 s8 , s8 , s8
vldr.f32 s8 , =0
vmov.f32 s9 , s8
vmov.f32 s10, s8
vmov.f32 s11, s8
@@ -340,7 +340,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT1x2

vsub.f32 s8 , s8 , s8
vldr.f32 s8 , =0
vmov.f32 s9 , s8
vmov.f32 s12, s8
vmov.f32 s13, s8
@@ -514,7 +514,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT2x1

vsub.f32 s8 , s8 , s8
vldr.f32 s8 , =0
vmov.f32 s9 , s8
vmov.f32 s10, s8
vmov.f32 s11, s8
@@ -681,7 +681,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT1x1

vsub.f32 s8 , s8 , s8
vldr.f32 s8 , =0
vmov.f32 s9 , s8

.endm


+ 6
- 4
kernel/arm/cgemm_kernel_2x2_vfpv3.S View File

@@ -147,7 +147,6 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT2x2

vsub.f32 s16 , s16 , s16
vmov.f32 s17, s16
vmov.f32 s18, s16
vmov.f32 s19, s16
@@ -368,7 +367,6 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT1x2

vsub.f32 s16 , s16 , s16
vmov.f32 s17, s16
vmov.f32 s20, s16
vmov.f32 s21, s16
@@ -550,7 +548,6 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT2x1

vsub.f32 s16 , s16 , s16
vmov.f32 s17, s16
vmov.f32 s18, s16
vmov.f32 s19, s16
@@ -730,7 +727,6 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT1x1

vsub.f32 s16 , s16 , s16
vmov.f32 s17, s16
vmov.f32 s24, s16
vmov.f32 s25, s16
@@ -1008,9 +1004,12 @@ cgemm_kernel_L2_M2_32:

b cgemm_kernel_L2_M2_44

cgemm_f32zero:
.word 0x00000000

cgemm_kernel_L2_M2_40:

vldr.f32 s16 , cgemm_f32zero
INIT2x2


@@ -1044,6 +1043,7 @@ cgemm_kernel_L2_M1_BEGIN:

cgemm_kernel_L2_M1_20:

vldr.f32 s16 , cgemm_f32zero
INIT1x2

mov BO, BC
@@ -1219,6 +1219,7 @@ cgemm_kernel_L1_M2_32:

cgemm_kernel_L1_M2_40:

vldr.f32 s16 , =0
INIT2x1


@@ -1252,6 +1253,7 @@ cgemm_kernel_L1_M1_BEGIN:

cgemm_kernel_L1_M1_20:

vldr.f32 s16 , =0
INIT1x1

mov BO, BC


+ 4
- 4
kernel/arm/cgemv_n_vfp.S View File

@@ -117,7 +117,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
.macro INIT_F4

pld [ YO, #Y_PRE ]
vsub.f32 s8 , s8 , s8
vldr.f32 s8 , =0
vmov.f32 s9 , s8
vmov.f32 s10, s8
vmov.f32 s11, s8
@@ -220,7 +220,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT_F1

vsub.f32 s8 , s8 , s8
vldr.f32 s8 , =0
vmov.f32 s9 , s8

.endm
@@ -267,7 +267,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT_S4

vsub.f32 s8 , s8 , s8
vldr.f32 s8 , =0
vmov.f32 s9 , s8
vmov.f32 s10, s8
vmov.f32 s11, s8
@@ -384,7 +384,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT_S1

vsub.f32 s8 , s8 , s8
vldr.f32 s8 , =0
vmov.f32 s9 , s8

.endm


+ 12
- 12
kernel/arm/cgemv_t_vfp.S View File

@@ -116,10 +116,10 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT_F2

vsub.f32 s12, s12, s12
vsub.f32 s13, s13, s13
vsub.f32 s14, s14, s14
vsub.f32 s15, s15, s15
vldr.f32 s12, =0
vldr.f32 s13, =0
vldr.f32 s14, =0
vldr.f32 s15, =0

.endm

@@ -172,8 +172,8 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT_F1

vsub.f32 s12, s12, s12
vsub.f32 s13, s13, s13
vldr.f32 s12, =0
vldr.f32 s13, =0

.endm

@@ -215,10 +215,10 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT_S2

vsub.f32 s12, s12, s12
vsub.f32 s13, s13, s13
vsub.f32 s14, s14, s14
vsub.f32 s15, s15, s15
vldr.f32 s12, =0
vldr.f32 s13, =0
vldr.f32 s14, =0
vldr.f32 s15, =0

.endm

@@ -281,8 +281,8 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT_S1

vsub.f32 s12, s12, s12
vsub.f32 s13, s13, s13
vldr.f32 s12, =0
vldr.f32 s13, =0

.endm



+ 22
- 22
kernel/arm/ctrmm_kernel_2x2_vfp.S View File

@@ -136,7 +136,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT2x2

vsub.f32 s8 , s8 , s8
vldr.f32 s8 , =0
vmov.f32 s9 , s8
vmov.f32 s10, s8
vmov.f32 s11, s8
@@ -301,10 +301,10 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
flds s0, ALPHA_R
flds s1, ALPHA_I

vsub.f32 s4, s4, s4
vsub.f32 s5, s5, s5
vsub.f32 s6, s6, s6
vsub.f32 s7, s7, s7
vldr.f32 s4, =0
vldr.f32 s5, =0
vldr.f32 s6, =0
vldr.f32 s7, =0

FMAC_R1 s4 , s0 , s8
FMAC_I1 s5 , s0 , s9
@@ -318,10 +318,10 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

fstmias CO1, { s4 - s7 }

vsub.f32 s4, s4, s4
vsub.f32 s5, s5, s5
vsub.f32 s6, s6, s6
vsub.f32 s7, s7, s7
vldr.f32 s4, =0
vldr.f32 s5, =0
vldr.f32 s6, =0
vldr.f32 s7, =0

FMAC_R1 s4 , s0 , s12
FMAC_I1 s5 , s0 , s13
@@ -343,7 +343,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT1x2

vsub.f32 s8 , s8 , s8
vldr.f32 s8 , =0
vmov.f32 s9 , s8
vmov.f32 s12, s8
vmov.f32 s13, s8
@@ -490,8 +490,8 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
flds s0, ALPHA_R
flds s1, ALPHA_I

vsub.f32 s4, s4, s4
vsub.f32 s5, s5, s5
vldr.f32 s4, =0
vldr.f32 s5, =0

FMAC_R1 s4 , s0 , s8
FMAC_I1 s5 , s0 , s9
@@ -500,8 +500,8 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

fstmias CO1, { s4 - s5 }

vsub.f32 s4, s4, s4
vsub.f32 s5, s5, s5
vldr.f32 s4, =0
vldr.f32 s5, =0

FMAC_R1 s4 , s0 , s12
FMAC_I1 s5 , s0 , s13
@@ -519,7 +519,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT2x1

vsub.f32 s8 , s8 , s8
vldr.f32 s8 , =0
vmov.f32 s9 , s8
vmov.f32 s10, s8
vmov.f32 s11, s8
@@ -663,10 +663,10 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
flds s0, ALPHA_R
flds s1, ALPHA_I

vsub.f32 s4, s4, s4
vsub.f32 s5, s5, s5
vsub.f32 s6, s6, s6
vsub.f32 s7, s7, s7
vldr.f32 s4, =0
vldr.f32 s5, =0
vldr.f32 s6, =0
vldr.f32 s7, =0

FMAC_R1 s4 , s0 , s8
FMAC_I1 s5 , s0 , s9
@@ -689,7 +689,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT1x1

vsub.f32 s8 , s8 , s8
vldr.f32 s8 , =0
vmov.f32 s9 , s8

.endm
@@ -795,8 +795,8 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
flds s0, ALPHA_R
flds s1, ALPHA_I

vsub.f32 s4, s4, s4
vsub.f32 s5, s5, s5
vldr.f32 s4, =0
vldr.f32 s5, =0

FMAC_R1 s4 , s0 , s8
FMAC_I1 s5 , s0 , s9


+ 6
- 4
kernel/arm/ctrmm_kernel_2x2_vfpv3.S View File

@@ -134,7 +134,6 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT2x2

vsub.f32 s16 , s16 , s16
vmov.f32 s17, s16
vmov.f32 s18, s16
vmov.f32 s19, s16
@@ -351,7 +350,6 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT1x2

vsub.f32 s16 , s16 , s16
vmov.f32 s17, s16
vmov.f32 s20, s16
vmov.f32 s21, s16
@@ -529,7 +527,6 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT2x1

vsub.f32 s16 , s16 , s16
vmov.f32 s17, s16
vmov.f32 s18, s16
vmov.f32 s19, s16
@@ -706,7 +703,6 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT1x1

vsub.f32 s16 , s16 , s16
vmov.f32 s17, s16
vmov.f32 s24, s16
vmov.f32 s25, s16
@@ -1020,9 +1016,12 @@ _L2_M2_32:

b _L2_M2_44

ctrmm_f32zero:
.word 0x00000000

_L2_M2_40:

vldr.f32 s16 , ctrmm_f32zero
INIT2x2


@@ -1074,6 +1073,7 @@ _L2_M1_BEGIN:

_L2_M1_20:

vldr.f32 s16 , ctrmm_f32zero
INIT1x2

#if (defined(LEFT) && defined(TRANSA)) || \
@@ -1337,6 +1337,7 @@ _L1_M2_32:

_L1_M2_40:

vldr.f32 s16 , =0
INIT2x1


@@ -1389,6 +1390,7 @@ _L1_M1_BEGIN:

_L1_M1_20:

vldr.f32 s16 , =0
INIT1x1

#if (defined(LEFT) && defined(TRANSA)) || \


+ 2
- 2
kernel/arm/ddot_vfp.S View File

@@ -152,8 +152,8 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
mov Y, OLD_Y
ldr INC_Y, OLD_INC_Y

vsub.f64 d0 , d0 , d0
vsub.f64 d1 , d1 , d1
vldr.f64 d0 , =0
vldr.f64 d1 , =0

cmp N, #0
ble ddot_kernel_L999


+ 6
- 6
kernel/arm/dgemm_kernel_4x2_vfp.S View File

@@ -85,7 +85,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT4x2

vsub.f64 d8 , d8 , d8
vldr.f64 d8 , =0
vmov.f64 d9, d8
vmov.f64 d10, d8
vmov.f64 d11, d8
@@ -173,7 +173,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT2x2

vsub.f64 d8 , d8 , d8
vldr.f64 d8 , =0
vmov.f64 d9, d8
vmov.f64 d12, d8
vmov.f64 d13, d8
@@ -233,7 +233,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT1x2

vsub.f64 d8 , d8 , d8
vldr.f64 d8 , =0
vmov.f64 d12, d8

.endm
@@ -283,7 +283,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT4x1

vsub.f64 d8 , d8 , d8
vldr.f64 d8 , =0
vmov.f64 d9, d8
vmov.f64 d10, d8
vmov.f64 d11, d8
@@ -338,7 +338,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT2x1

vsub.f64 d8 , d8 , d8
vldr.f64 d8 , =0
vmov.f64 d9 , d8

.endm
@@ -380,7 +380,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT1x1

vsub.f64 d8 , d8 , d8
vldr.f64 d8 , =0

.endm



+ 9
- 9
kernel/arm/dgemm_kernel_4x4_vfpv3.S View File

@@ -102,7 +102,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT4x4

vsub.f64 d16 , d16 , d16
vldr.f64 d16 , =0
vmov.f64 d17, d16
vmov.f64 d18, d16
vmov.f64 d19, d16
@@ -376,7 +376,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT2x4

vsub.f64 d16 , d16 , d16
vldr.f64 d16 , =0
vmov.f64 d17, d16
vmov.f64 d20, d16
vmov.f64 d21, d16
@@ -470,7 +470,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT1x4

vsub.f64 d16 , d16 , d16
vldr.f64 d16 , =0
vmov.f64 d20, d16
vmov.f64 d24, d16
vmov.f64 d28, d16
@@ -533,7 +533,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT4x2

vsub.f64 d16 , d16 , d16
vldr.f64 d16 , =0
vmov.f64 d17, d16
vmov.f64 d18, d16
vmov.f64 d19, d16
@@ -617,7 +617,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT2x2

vsub.f64 d16 , d16 , d16
vldr.f64 d16 , =0
vmov.f64 d17, d16
vmov.f64 d20, d16
vmov.f64 d21, d16
@@ -678,7 +678,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT1x2

vsub.f64 d16 , d16 , d16
vldr.f64 d16 , =0
vmov.f64 d20, d16

.endm
@@ -723,7 +723,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT4x1

vsub.f64 d16 , d16 , d16
vldr.f64 d16 , =0
vmov.f64 d17, d16
vmov.f64 d18, d16
vmov.f64 d19, d16
@@ -782,7 +782,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT2x1

vsub.f64 d16 , d16 , d16
vldr.f64 d16 , =0
vmov.f64 d17, d16

.endm
@@ -826,7 +826,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT1x1

vsub.f64 d16 , d16 , d16
vldr.f64 d16 , =0

.endm



+ 6
- 6
kernel/arm/dtrmm_kernel_4x2_vfp.S View File

@@ -90,7 +90,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT4x2

vsub.f64 d8 , d8 , d8
vldr.f64 d8 , =0
vmov.f64 d9, d8
vmov.f64 d10, d8
vmov.f64 d11, d8
@@ -165,7 +165,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT2x2

vsub.f64 d8 , d8 , d8
vldr.f64 d8 , =0
vmov.f64 d9, d8
vmov.f64 d12, d8
vmov.f64 d13, d8
@@ -220,7 +220,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT1x2

vsub.f64 d8 , d8 , d8
vldr.f64 d8 , =0
vmov.f64 d12, d8

.endm
@@ -268,7 +268,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT4x1

vsub.f64 d8 , d8 , d8
vldr.f64 d8 , =0
vmov.f64 d9, d8
vmov.f64 d10, d8
vmov.f64 d11, d8
@@ -318,7 +318,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT2x1

vsub.f64 d8 , d8 , d8
vldr.f64 d8 , =0
vmov.f64 d9 , d8

.endm
@@ -357,7 +357,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT1x1

vsub.f64 d8 , d8 , d8
vldr.f64 d8 , =0

.endm



+ 9
- 9
kernel/arm/dtrmm_kernel_4x4_vfpv3.S View File

@@ -89,7 +89,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT4x4

vsub.f64 d16 , d16 , d16
vldr.f64 d16 , =0
vmov.f64 d17, d16
vmov.f64 d18, d16
vmov.f64 d19, d16
@@ -386,7 +386,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT2x4

vsub.f64 d16 , d16 , d16
vldr.f64 d16 , =0
vmov.f64 d17, d16
vmov.f64 d20, d16
vmov.f64 d21, d16
@@ -468,7 +468,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT1x4

vsub.f64 d16 , d16 , d16
vldr.f64 d16 , =0
vmov.f64 d20, d16
vmov.f64 d24, d16
vmov.f64 d28, d16
@@ -527,7 +527,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT4x2

vsub.f64 d16 , d16 , d16
vldr.f64 d16 , =0
vmov.f64 d17, d16
vmov.f64 d18, d16
vmov.f64 d19, d16
@@ -601,7 +601,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT2x2

vsub.f64 d16 , d16 , d16
vldr.f64 d16 , =0
vmov.f64 d17, d16
vmov.f64 d20, d16
vmov.f64 d21, d16
@@ -656,7 +656,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT1x2

vsub.f64 d16 , d16 , d16
vldr.f64 d16 , =0
vmov.f64 d20, d16

.endm
@@ -699,7 +699,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT4x1

vsub.f64 d16 , d16 , d16
vldr.f64 d16 , =0
vmov.f64 d17, d16
vmov.f64 d18, d16
vmov.f64 d19, d16
@@ -753,7 +753,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT2x1

vsub.f64 d16 , d16 , d16
vldr.f64 d16 , =0
vmov.f64 d17, d16

.endm
@@ -794,7 +794,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT1x1

vsub.f64 d16 , d16 , d16
vldr.f64 d16 , =0

.endm



+ 8
- 8
kernel/arm/gemv_n_vfp.S View File

@@ -79,7 +79,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
pld [ YO , #Y_PRE ]
pld [ YO , #Y_PRE+32 ]

vsub.f64 d8 , d8 , d8
vldr.f64 d8 , =0
vmov.f64 d9 , d8
vmov.f64 d10 , d8
vmov.f64 d11 , d8
@@ -158,7 +158,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT_F1

vsub.f64 d12 , d12 , d12
vldr.f64 d12 , =0

.endm

@@ -185,7 +185,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT_S4

vsub.f64 d12 , d12 , d12
vldr.f64 d12 , =0
vmov.f64 d13 , d12
vmov.f64 d14 , d12
vmov.f64 d15 , d12
@@ -245,7 +245,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT_S1

vsub.f64 d12 , d12 , d12
vldr.f64 d12 , =0

.endm

@@ -279,7 +279,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

pld [ YO , #Y_PRE ]

vsub.f32 s8 , s8 , s8
vldr.f32 s8 , =0
vmov.f32 s9 , s8
vmov.f32 s10 , s8
vmov.f32 s11 , s8
@@ -357,7 +357,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT_F1

vsub.f32 s12 , s12 , s12
vldr.f32 s12 , =0

.endm

@@ -384,7 +384,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT_S4

vsub.f32 s12 , s12 , s12
vldr.f32 s12 , =0
vmov.f32 s13 , s12
vmov.f32 s14 , s12
vmov.f32 s15 , s12
@@ -445,7 +445,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT_S1

vsub.f32 s12 , s12 , s12
vldr.f32 s12 , =0

.endm



+ 8
- 8
kernel/arm/gemv_n_vfpv3.S View File

@@ -79,7 +79,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
pld [ YO , #Y_PRE ]
pld [ YO , #Y_PRE+32 ]

vsub.f64 d24 , d24 , d24
vldr.f64 d24 , =0
vmov.f64 d25 , d24
vmov.f64 d26 , d24
vmov.f64 d27 , d24
@@ -147,7 +147,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT_F1

vsub.f64 d24 , d24 , d24
vldr.f64 d24 , =0

.endm

@@ -175,7 +175,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT_S8

vsub.f64 d24 , d24 , d24
vldr.f64 d24 , =0
vmov.f64 d25 , d24
vmov.f64 d26 , d24
vmov.f64 d27 , d24
@@ -269,7 +269,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT_S1

vsub.f64 d24 , d24 , d24
vldr.f64 d24 , =0

.endm

@@ -302,7 +302,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

pld [ YO , #Y_PRE ]

vsub.f32 s24 , s24 , s24
vldr.f32 s24 , =0
vmov.f32 s25 , s24
vmov.f32 s26 , s24
vmov.f32 s27 , s24
@@ -368,7 +368,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT_F1

vsub.f32 s24 , s24 , s24
vldr.f32 s24 , =0

.endm

@@ -396,7 +396,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT_S8

vsub.f32 s24 , s24 , s24
vldr.f32 s24 , =0
vmov.f32 s25 , s24
vmov.f32 s26 , s24
vmov.f32 s27 , s24
@@ -489,7 +489,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT_S1

vsub.f32 s24 , s24 , s24
vldr.f32 s24 , =0

.endm



+ 12
- 12
kernel/arm/gemv_t_vfp.S View File

@@ -75,8 +75,8 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT_F2

vsub.f64 d2 , d2 , d2
vsub.f64 d3 , d3 , d3
vldr.f64 d2 , =0
vldr.f64 d3 , =0

.endm

@@ -123,7 +123,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT_F1

vsub.f64 d2 , d2 , d2
vldr.f64 d2 , =0

.endm

@@ -160,8 +160,8 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT_S2

vsub.f64 d2 , d2 , d2
vsub.f64 d3 , d3 , d3
vldr.f64 d2 , =0
vldr.f64 d3 , =0

.endm

@@ -224,7 +224,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT_S1

vsub.f64 d2 , d2 , d2
vldr.f64 d2 , =0

.endm

@@ -276,8 +276,8 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT_F2

vsub.f32 s2 , s2 , s2
vsub.f32 s3 , s3 , s3
vldr.f32 s2 , =0
vldr.f32 s3 , =0

.endm

@@ -321,7 +321,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT_F1

vsub.f32 s2 , s2 , s2
vldr.f32 s2 , =0

.endm

@@ -356,8 +356,8 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT_S2

vsub.f32 s2 , s2 , s2
vsub.f32 s3 , s3 , s3
vldr.f32 s2 , =0
vldr.f32 s3 , =0

.endm

@@ -418,7 +418,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT_S1

vsub.f32 s2 , s2 , s2
vldr.f32 s2 , =0

.endm



+ 12
- 12
kernel/arm/gemv_t_vfpv3.S View File

@@ -75,8 +75,8 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT_F2

vsub.f64 d4 , d4 , d4
vsub.f64 d5 , d5 , d5
vldr.f64 d4 , =0
vldr.f64 d5 , =0

.endm

@@ -123,8 +123,8 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT_S2

vsub.f64 d4 , d4 , d4
vsub.f64 d5 , d5 , d5
vldr.f64 d4 , =0
vldr.f64 d5 , =0

.endm

@@ -183,7 +183,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT_F1

vsub.f64 d4 , d4 , d4
vldr.f64 d4 , =0

.endm

@@ -220,7 +220,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT_S1

vsub.f64 d4 , d4 , d4
vldr.f64 d4 , =0

.endm

@@ -268,8 +268,8 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT_F2

vsub.f32 s4 , s4 , s4
vsub.f32 s5 , s5 , s5
vldr.f32 s4 , =0
vldr.f32 s5 , =0

.endm

@@ -313,8 +313,8 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT_S2

vsub.f32 s4 , s4 , s4
vsub.f32 s5 , s5 , s5
vldr.f32 s4 , =0
vldr.f32 s5 , =0

.endm

@@ -371,7 +371,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT_F1

vsub.f32 s4 , s4 , s4
vldr.f32 s4 , =0

.endm

@@ -406,7 +406,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT_S1

vsub.f32 s4 , s4 , s4
vldr.f32 s4 , =0

.endm



+ 2
- 2
kernel/arm/iamax_vfp.S View File

@@ -342,9 +342,9 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
push {r4}

#if defined(DOUBLE)
vsub.f64 d0 , d0 , d0
vldr.f64 d0 , =0
#else
vsub.f32 s0 , s0 , s0
vldr.f32 s0 , =0
#endif
mov INDEX, #0



+ 4
- 4
kernel/arm/nrm2_vfp.S View File

@@ -446,12 +446,12 @@ nrm2_begin:
#if defined(COMPLEX)

#if defined(DOUBLE)
vsub.f64 d0 , d0 , d0 // scale=0.0
vldr.f64 d0 , =0 // scale=0.0
vldr.64 d1 , znrm2_one // ssq=1.0
vmov.f64 d7 , d1 // value 1.0
vmov.f64 d6 , d0 // value 0.0
#else
vsub.f32 s0 , s0 , s0 // scale=0.0
vldr.f32 s0 , =0 // scale=0.0
vldr.32 s1 , cnrm2_one // ssq=1.0
vmov.f32 s7 , s1 // value 1.0
vmov.f32 s6 , s0 // value 0.0
@@ -460,12 +460,12 @@ nrm2_begin:
#else

#if defined(DOUBLE)
vsub.f64 d0 , d0 , d0 // scale=0.0
vldr.f64 d0 , =0 // scale=0.0
vldr.64 d1 , dnrm2_one // ssq=1.0
vmov.f64 d7 , d1 // value 1.0
vmov.f64 d6 , d0 // value 0.0
#else
vsub.f32 s0 , s0 , s0 // scale=0.0
vldr.f32 s0 , =0 // scale=0.0
vldr.32 s1 , snrm2_one // ssq=1.0
vmov.f32 s7 , s1 // value 1.0
vmov.f32 s6 , s0 // value 0.0


+ 7
- 2
kernel/arm/nrm2_vfpv3.S View File

@@ -400,17 +400,22 @@ KERNEL_S1_END_\@:
* End of macro definitions
**************************************************************************************/

nrm2_zeros:
.align 5
.word 0x00000000
.word 0x00000000

PROLOGUE

.align 5

#if defined(DOUBLE)
vsub.f64 d0 , d0 , d0 // scale=0.0
vldr.f64 d0 , nrm2_zeros // scale=0.0
vmov.f64 d1 , #1.0 // ssq=1.0
vmov.f64 d7 , d1 // value 1.0
vmov.f64 d6 , d0 // value 0.0
#else
vsub.f32 s0 , s0 , s0 // scale=0.0
vldr.f32 s0 , nrm2_zeros // scale=0.0
vmov.f32 s1 , #1.0 // ssq=1.0
vmov.f32 s7 , s1 // value 1.0
vmov.f32 s6 , s0 // value 0.0


+ 4
- 4
kernel/arm/sdot_vfp.S View File

@@ -242,13 +242,13 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

#if defined(DSDOT)

vsub.f64 d0 , d0 , d0
vsub.f64 d1 , d1 , d1
vldr.f64 d0 , =0
vldr.f64 d1 , =0

#else

vsub.f32 s0 , s0 , s0
vsub.f32 s1 , s1 , s1
vldr.f32 s0 , =0
vldr.f32 s1 , =0

#endif



+ 6
- 6
kernel/arm/sgemm_kernel_4x2_vfp.S View File

@@ -85,7 +85,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT4x2

vsub.f32 s8 , s8 , s8
vldr.f32 s8 , =0
vmov.f32 s9, s8
vmov.f32 s10, s8
vmov.f32 s11, s8
@@ -161,7 +161,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT2x2

vsub.f32 s8 , s8 , s8
vldr.f32 s8 , =0
vmov.f32 s9, s8
vmov.f32 s12, s8
vmov.f32 s13, s8
@@ -221,7 +221,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT1x2

vsub.f32 s8 , s8 , s8
vldr.f32 s8 , =0
vmov.f32 s12, s8

.endm
@@ -271,7 +271,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT4x1

vsub.f32 s8 , s8 , s8
vldr.f32 s8 , =0
vmov.f32 s9, s8
vmov.f32 s10, s8
vmov.f32 s11, s8
@@ -326,7 +326,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT2x1

vsub.f32 s8 , s8 , s8
vldr.f32 s8 , =0
vmov.f32 s9 , s8

.endm
@@ -368,7 +368,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT1x1

vsub.f32 s8 , s8 , s8
vldr.f32 s8 , =0

.endm



+ 32
- 17
kernel/arm/sgemm_kernel_4x4_vfpv3.S View File

@@ -102,7 +102,6 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT4x4

vsub.f32 s16 , s16 , s16
vmov.f32 s17, s16
vmov.f32 s18, s16
vmov.f32 s19, s16
@@ -349,7 +348,6 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT2x4

vsub.f32 s16 , s16 , s16
vmov.f32 s17, s16
vmov.f32 s20, s16
vmov.f32 s21, s16
@@ -443,7 +441,6 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT1x4

vsub.f32 s16 , s16 , s16
vmov.f32 s20, s16
vmov.f32 s24, s16
vmov.f32 s28, s16
@@ -506,7 +503,6 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT4x2

vsub.f32 s16 , s16 , s16
vmov.f32 s17, s16
vmov.f32 s18, s16
vmov.f32 s19, s16
@@ -590,7 +586,6 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT2x2

vsub.f32 s16 , s16 , s16
vmov.f32 s17, s16
vmov.f32 s20, s16
vmov.f32 s21, s16
@@ -651,7 +646,6 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT1x2

vsub.f32 s16 , s16 , s16
vmov.f32 s20, s16

.endm
@@ -696,7 +690,6 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT4x1

vsub.f32 s16 , s16 , s16
vmov.f32 s17, s16
vmov.f32 s18, s16
vmov.f32 s19, s16
@@ -755,7 +748,6 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT2x1

vsub.f32 s16 , s16 , s16
vmov.f32 s17, s16

.endm
@@ -797,13 +789,6 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

/******************************************************************************/

.macro INIT1x1

vsub.f32 s16 , s16 , s16

.endm



.macro KERNEL1x1_SUB

@@ -930,9 +915,24 @@ sgemm_kernel_L4_M4_32:

b sgemm_kernel_L4_M4_44

/* Note on loading zero into a fp register
* vsub doesn't work since it cannot handle NaN and infinity
* vmov.Fn doesn't work with 0
* vmov.In and veor are neon
* vldr , =0 doesn't work since the function is larger than 2KB
* and the assembler is not able to insert constant pool inside
* the function body.
*
* Therefore, the best way I've found is to manually create multiple
* copies of the zero constant and `vldr` from different ones depending
* on where the instruction is.
*/
sgemm_f32zero:
.word 0x00000000

sgemm_kernel_L4_M4_40:

vldr.f32 s16 , sgemm_f32zero
INIT4x4


@@ -969,6 +969,7 @@ sgemm_kernel_L4_M2_BEGIN:

sgemm_kernel_L4_M2_20:

vldr.f32 s16 , sgemm_f32zero
INIT2x4

mov BO, BC
@@ -1014,9 +1015,14 @@ sgemm_kernel_L4_M1_BEGIN:

tst I, #1 // I = I % 2
ble sgemm_kernel_L4_END
b sgemm_kernel_L4_M1_20

sgemm_f32zero4:
.word 0x00000000

sgemm_kernel_L4_M1_20:

vldr.f32 s16 , sgemm_f32zero4
INIT1x4

mov BO, BC
@@ -1100,6 +1106,7 @@ sgemm_kernel_L2_M4_BEGIN:

sgemm_kernel_L2_M4_20:

vldr.f32 s16 , sgemm_f32zero3
INIT4x2

mov BO, BC
@@ -1121,7 +1128,6 @@ sgemm_kernel_L2_M4_22:
subs L, L, #1
bgt sgemm_kernel_L2_M4_22


sgemm_kernel_L2_M4_40:

ands L , K1, #7 // L = L % 8
@@ -1142,7 +1148,10 @@ sgemm_kernel_L2_M4_END:

subs I, I, #1
bgt sgemm_kernel_L2_M4_20
b sgemm_kernel_L2_M2_BEGIN

sgemm_f32zero3:
.word 0x00000000

sgemm_kernel_L2_M2_BEGIN:

@@ -1155,6 +1164,7 @@ sgemm_kernel_L2_M2_BEGIN:

sgemm_kernel_L2_M2_20:

vldr.f32 s16 , sgemm_f32zero3
INIT2x2

mov BO, BC
@@ -1203,6 +1213,7 @@ sgemm_kernel_L2_M1_BEGIN:

sgemm_kernel_L2_M1_20:

vldr.f32 s16 , sgemm_f32zero3
INIT1x2

mov BO, BC
@@ -1278,6 +1289,7 @@ sgemm_kernel_L1_M4_BEGIN:

sgemm_kernel_L1_M4_20:

vldr.f32 s16 , sgemm_f32zero3
INIT4x1

mov BO, BC
@@ -1333,6 +1345,7 @@ sgemm_kernel_L1_M2_BEGIN:

sgemm_kernel_L1_M2_20:

vldr.f32 s16 , sgemm_f32zero2
INIT2x1

mov BO, BC
@@ -1381,7 +1394,7 @@ sgemm_kernel_L1_M1_BEGIN:

sgemm_kernel_L1_M1_20:

INIT1x1
vldr.f32 s16 , sgemm_f32zero2

mov BO, BC
asrs L , K1, #3 // L = L / 8
@@ -1434,3 +1447,5 @@ sgemm_kernel_L999:

EPILOGUE

sgemm_f32zero2:
.word 0x00000000

+ 6
- 6
kernel/arm/strmm_kernel_4x2_vfp.S View File

@@ -90,7 +90,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT4x2

vsub.f32 s8 , s8 , s8
vldr.f32 s8 , =0
vmov.f32 s9, s8
vmov.f32 s10, s8
vmov.f32 s11, s8
@@ -156,7 +156,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT2x2

vsub.f32 s8 , s8 , s8
vldr.f32 s8 , =0
vmov.f32 s9, s8
vmov.f32 s12, s8
vmov.f32 s13, s8
@@ -211,7 +211,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT1x2

vsub.f32 s8 , s8 , s8
vldr.f32 s8 , =0
vmov.f32 s12, s8

.endm
@@ -259,7 +259,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT4x1

vsub.f32 s8 , s8 , s8
vldr.f32 s8 , =0
vmov.f32 s9, s8
vmov.f32 s10, s8
vmov.f32 s11, s8
@@ -309,7 +309,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT2x1

vsub.f32 s8 , s8 , s8
vldr.f32 s8 , =0
vmov.f32 s9 , s8

.endm
@@ -348,7 +348,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT1x1

vsub.f32 s8 , s8 , s8
vldr.f32 s8 , =0

.endm



+ 33
- 16
kernel/arm/strmm_kernel_4x4_vfpv3.S View File

@@ -88,7 +88,6 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT4x4

vsub.f32 s16 , s16 , s16
vmov.f32 s17, s16
vmov.f32 s18, s16
vmov.f32 s19, s16
@@ -322,7 +321,6 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT2x4

vsub.f32 s16 , s16 , s16
vmov.f32 s17, s16
vmov.f32 s20, s16
vmov.f32 s21, s16
@@ -405,7 +403,6 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT1x4

vsub.f32 s16 , s16 , s16
vmov.f32 s20, s16
vmov.f32 s24, s16
vmov.f32 s28, s16
@@ -464,7 +461,6 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT4x2

vsub.f32 s16 , s16 , s16
vmov.f32 s17, s16
vmov.f32 s18, s16
vmov.f32 s19, s16
@@ -538,7 +534,6 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT2x2

vsub.f32 s16 , s16 , s16
vmov.f32 s17, s16
vmov.f32 s20, s16
vmov.f32 s21, s16
@@ -593,7 +588,6 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT1x2

vsub.f32 s16 , s16 , s16
vmov.f32 s20, s16

.endm
@@ -636,7 +630,6 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT4x1

vsub.f32 s16 , s16 , s16
vmov.f32 s17, s16
vmov.f32 s18, s16
vmov.f32 s19, s16
@@ -690,7 +683,6 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT2x1

vsub.f32 s16 , s16 , s16
vmov.f32 s17, s16

.endm
@@ -729,13 +721,6 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

/******************************************************************************/

.macro INIT1x1

vsub.f32 s16 , s16 , s16

.endm



.macro KERNEL1x1_SUB

@@ -955,9 +940,24 @@ _L4_M4_32:

b _L4_M4_44

/* Note on loading zero into a fp register
* vsub doesn't work since it cannot handle NaN and infinity
* vmov.Fn doesn't work with 0
* vmov.In and veor are neon
* vldr , =0 doesn't work since the function is larger than 2KB
* and the assembler is not able to insert constant pool inside
* the function body.
*
* Therefore, the best way I've found is to manually create multiple
* copies of the zero constant and `vldr` from different ones depending
* on where the instruction is.
*/
strmm_f32zero:
.word 0x00000000

_L4_M4_40:

vldr.f32 s16 , strmm_f32zero
INIT4x4


@@ -1014,6 +1014,7 @@ _L4_M2_BEGIN:

_L4_M2_20:

vldr.f32 s16 , strmm_f32zero
INIT2x4

#if (defined(LEFT) && defined(TRANSA)) || \
@@ -1112,6 +1113,7 @@ _L4_M1_BEGIN:

_L4_M1_20:

vldr.f32 s16 , strmm_f32zero3
INIT1x4

#if (defined(LEFT) && defined(TRANSA)) || \
@@ -1252,9 +1254,14 @@ _L2_M4_BEGIN:
ldr I, M
asrs I, I, #2 // I = I / 4
ble _L2_M2_BEGIN
b _L2_M4_20

strmm_f32zero3:
.word 0x00000000

_L2_M4_20:

vldr.f32 s16 , strmm_f32zero3
INIT4x2

#if (defined(LEFT) && defined(TRANSA)) || \
@@ -1360,6 +1367,7 @@ _L2_M2_BEGIN:

_L2_M2_20:

vldr.f32 s16 , strmm_f32zero3
INIT2x2

#if (defined(LEFT) && defined(TRANSA)) || \
@@ -1458,6 +1466,7 @@ _L2_M1_BEGIN:

_L2_M1_20:

vldr.f32 s16 , strmm_f32zero4
INIT1x2

#if (defined(LEFT) && defined(TRANSA)) || \
@@ -1591,9 +1600,14 @@ _L1_M4_BEGIN:
ldr I, M
asrs I, I, #2 // I = I / 4
ble _L1_M2_BEGIN
b _L1_M4_20

strmm_f32zero4:
.word 0x00000000

_L1_M4_20:

vldr.f32 s16 , strmm_f32zero4
INIT4x1

#if (defined(LEFT) && defined(TRANSA)) || \
@@ -1700,6 +1714,7 @@ _L1_M2_BEGIN:

_L1_M2_20:

vldr.f32 s16 , strmm_f32zero2
INIT2x1

#if (defined(LEFT) && defined(TRANSA)) || \
@@ -1798,7 +1813,7 @@ _L1_M1_BEGIN:

_L1_M1_20:

INIT1x1
vldr.f32 s16 , strmm_f32zero2

#if (defined(LEFT) && defined(TRANSA)) || \
(!defined(LEFT) && !defined(TRANSA))
@@ -1882,3 +1897,5 @@ _L999:

EPILOGUE

strmm_f32zero2:
.word 0x00000000

+ 4
- 4
kernel/arm/zdot_vfp.S View File

@@ -190,10 +190,10 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
mov Y, OLD_Y
ldr INC_Y, OLD_INC_Y

vsub.f64 d0 , d0 , d0
vsub.f64 d1 , d1 , d1
vsub.f64 d2 , d2 , d2
vsub.f64 d3 , d3 , d3
vldr.f64 d0 , =0
vldr.f64 d1 , =0
vldr.f64 d2 , =0
vldr.f64 d3 , =0

cmp N, #0
ble zdot_kernel_L999


+ 4
- 4
kernel/arm/zgemm_kernel_2x2_vfp.S View File

@@ -131,7 +131,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT2x2

vsub.f64 d8 , d8 , d8
vldr.f64 d8 , =0
vmov.f64 d9 , d8
vmov.f64 d10, d8
vmov.f64 d11, d8
@@ -383,7 +383,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT1x2

vsub.f64 d8 , d8 , d8
vldr.f64 d8 , =0
vmov.f64 d9 , d8
vmov.f64 d12, d8
vmov.f64 d13, d8
@@ -557,7 +557,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT2x1

vsub.f64 d8 , d8 , d8
vldr.f64 d8 , =0
vmov.f64 d9 , d8
vmov.f64 d10, d8
vmov.f64 d11, d8
@@ -724,7 +724,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT1x1

vsub.f64 d8 , d8 , d8
vldr.f64 d8 , =0
vmov.f64 d9 , d8

.endm


+ 4
- 4
kernel/arm/zgemm_kernel_2x2_vfpv3.S View File

@@ -147,7 +147,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT2x2

vsub.f64 d16 , d16 , d16
vldr.f64 d16 , =0
vmov.f64 d17, d16
vmov.f64 d18, d16
vmov.f64 d19, d16
@@ -404,7 +404,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT1x2

vsub.f64 d16 , d16 , d16
vldr.f64 d16 , =0
vmov.f64 d17, d16
vmov.f64 d20, d16
vmov.f64 d21, d16
@@ -586,7 +586,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT2x1

vsub.f64 d16 , d16 , d16
vldr.f64 d16 , =0
vmov.f64 d17, d16
vmov.f64 d18, d16
vmov.f64 d19, d16
@@ -766,7 +766,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT1x1

vsub.f64 d16 , d16 , d16
vldr.f64 d16 , =0
vmov.f64 d17, d16
vmov.f64 d24, d16
vmov.f64 d25, d16


+ 4
- 4
kernel/arm/zgemv_n_vfp.S View File

@@ -117,7 +117,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
.macro INIT_F4

pld [ YO, #Y_PRE ]
vsub.f64 d8 , d8 , d8
vldr.f64 d8 , =0
vmov.f64 d9 , d8
vmov.f64 d10, d8
vmov.f64 d11, d8
@@ -222,7 +222,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT_F1

vsub.f64 d8 , d8 , d8
vldr.f64 d8 , =0
vmov.f64 d9 , d8

.endm
@@ -269,7 +269,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT_S4

vsub.f64 d8 , d8 , d8
vldr.f64 d8 , =0
vmov.f64 d9 , d8
vmov.f64 d10, d8
vmov.f64 d11, d8
@@ -386,7 +386,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT_S1

vsub.f64 d8 , d8 , d8
vldr.f64 d8 , =0
vmov.f64 d9 , d8

.endm


+ 12
- 12
kernel/arm/zgemv_t_vfp.S View File

@@ -117,10 +117,10 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT_F2

vsub.f64 d12, d12, d12
vsub.f64 d13, d13, d13
vsub.f64 d14, d14, d14
vsub.f64 d15, d15, d15
vldr.f64 d12, =0
vldr.f64 d13, =0
vldr.f64 d14, =0
vldr.f64 d15, =0

.endm

@@ -173,8 +173,8 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT_F1

vsub.f64 d12, d12, d12
vsub.f64 d13, d13, d13
vldr.f64 d12, =0
vldr.f64 d13, =0

.endm

@@ -216,10 +216,10 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT_S2

vsub.f64 d12, d12, d12
vsub.f64 d13, d13, d13
vsub.f64 d14, d14, d14
vsub.f64 d15, d15, d15
vldr.f64 d12, =0
vldr.f64 d13, =0
vldr.f64 d14, =0
vldr.f64 d15, =0

.endm

@@ -282,8 +282,8 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT_S1

vsub.f64 d12, d12, d12
vsub.f64 d13, d13, d13
vldr.f64 d12, =0
vldr.f64 d13, =0

.endm



+ 22
- 22
kernel/arm/ztrmm_kernel_2x2_vfp.S View File

@@ -140,7 +140,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT2x2

vsub.f64 d8 , d8 , d8
vldr.f64 d8 , =0
vmov.f64 d9 , d8
vmov.f64 d10, d8
vmov.f64 d11, d8
@@ -356,10 +356,10 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
fldd d0, ALPHA_R
fldd d1, ALPHA_I

vsub.f64 d4, d4 , d4
vsub.f64 d5, d5 , d5
vsub.f64 d6, d6 , d6
vsub.f64 d7, d7 , d7
vldr.f64 d4, =0
vldr.f64 d5, =0
vldr.f64 d6, =0
vldr.f64 d7, =0

FMAC_R1 d4 , d0 , d8
FMAC_I1 d5 , d0 , d9
@@ -373,10 +373,10 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

fstmiad CO1, { d4 - d7 }

vsub.f64 d4, d4 , d4
vsub.f64 d5, d5 , d5
vsub.f64 d6, d6 , d6
vsub.f64 d7, d7 , d7
vldr.f64 d4, =0
vldr.f64 d5, =0
vldr.f64 d6, =0
vldr.f64 d7, =0

FMAC_R1 d4 , d0 , d12
FMAC_I1 d5 , d0 , d13
@@ -398,7 +398,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT1x2

vsub.f64 d8 , d8 , d8
vldr.f64 d8 , =0
vmov.f64 d9 , d8
vmov.f64 d12, d8
vmov.f64 d13, d8
@@ -545,8 +545,8 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
fldd d0, ALPHA_R
fldd d1, ALPHA_I

vsub.f64 d4, d4 , d4
vsub.f64 d5, d5 , d5
vldr.f64 d4, =0
vldr.f64 d5, =0

FMAC_R1 d4 , d0 , d8
FMAC_I1 d5 , d0 , d9
@@ -555,8 +555,8 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

fstmiad CO1, { d4 - d5 }

vsub.f64 d4, d4 , d4
vsub.f64 d5, d5 , d5
vldr.f64 d4, =0
vldr.f64 d5, =0

FMAC_R1 d4 , d0 , d12
FMAC_I1 d5 , d0 , d13
@@ -574,7 +574,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT2x1

vsub.f64 d8 , d8 , d8
vldr.f64 d8 , =0
vmov.f64 d9 , d8
vmov.f64 d10, d8
vmov.f64 d11, d8
@@ -718,10 +718,10 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
fldd d0, ALPHA_R
fldd d1, ALPHA_I

vsub.f64 d4, d4 , d4
vsub.f64 d5, d5 , d5
vsub.f64 d6, d6 , d6
vsub.f64 d7, d7 , d7
vldr.f64 d4, =0
vldr.f64 d5, =0
vldr.f64 d6, =0
vldr.f64 d7, =0

FMAC_R1 d4 , d0 , d8
FMAC_I1 d5 , d0 , d9
@@ -744,7 +744,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT1x1

vsub.f64 d8 , d8 , d8
vldr.f64 d8 , =0
vmov.f64 d9 , d8

.endm
@@ -850,8 +850,8 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
fldd d0, ALPHA_R
fldd d1, ALPHA_I

vsub.f64 d4, d4 , d4
vsub.f64 d5, d5 , d5
vldr.f64 d4, =0
vldr.f64 d5, =0

FMAC_R1 d4 , d0 , d8
FMAC_I1 d5 , d0 , d9


+ 4
- 4
kernel/arm/ztrmm_kernel_2x2_vfpv3.S View File

@@ -134,7 +134,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT2x2

vsub.f64 d16 , d16 , d16
vldr.f64 d16 , =0
vmov.f64 d17, d16
vmov.f64 d18, d16
vmov.f64 d19, d16
@@ -388,7 +388,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT1x2

vsub.f64 d16 , d16 , d16
vldr.f64 d16 , =0
vmov.f64 d17, d16
vmov.f64 d20, d16
vmov.f64 d21, d16
@@ -566,7 +566,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT2x1

vsub.f64 d16 , d16 , d16
vldr.f64 d16 , =0
vmov.f64 d17, d16
vmov.f64 d18, d16
vmov.f64 d19, d16
@@ -743,7 +743,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

.macro INIT1x1

vsub.f64 d16 , d16 , d16
vldr.f64 d16 , =0
vmov.f64 d17, d16
vmov.f64 d24, d16
vmov.f64 d25, d16


+ 1
- 1
kernel/generic/zgemmkernel_2x2.c View File

@@ -797,7 +797,7 @@ int CNAME(BLASLONG bm,BLASLONG bn,BLASLONG bk,FLOAT alphar,FLOAT alphai,FLOAT* b
res1 = res1-load0*load3;
#endif
#if defined(RN) || defined(RT) || defined(CN) || defined(CT)
load0 = ptrba[2*0+0];
load0 = ptrba[2*0+0];
load1 = ptrbb[2*0+0];
res0 = res0+load0*load1;
load2 = ptrba[2*0+1];


Loading…
Cancel
Save