@ -121,11 +121,13 @@ __b_evaluation:
ldr r11 , [ r12 , #o f f W 7 ] @ R11=W7
mul r5 , r10 , r7 @ R5=W5*ROWr16[1]=b2 (ROWr16[1] must be the second arg, to have the possibility to save 1 cycle)
mul r7 , r11 , r7 @ R7=W7*ROWr16[1]=b3 (ROWr16[1] must be the second arg, to have the possibility to save 1 cycle)
teq r2 , #0 @ if null avoid muls
mlane r0 , r9 , r2 , r0 @ R0+=W3*ROWr16[3]=b0 (ROWr16[3] must be the second arg, to have the possibility to save 1 cycle)
teq r2 , #0 @ if null avoid muls
itttt n e
mlane r0 , r9 , r2 , r0 @ R0+=W3*ROWr16[3]=b0 (ROWr16[3] must be the second arg, to have the possibility to save 1 cycle)
rsbne r2 , r2 , #0 @ R2=-ROWr16[3]
mlane r1 , r11 , r2 , r1 @ R1-=W7*ROWr16[3]=b1 (ROWr16[3] must be the second arg, to have the possibility to save 1 cycle)
mlane r5 , r8 , r2 , r5 @ R5-=W1*ROWr16[3]=b2 (ROWr16[3] must be the second arg, to have the possibility to save 1 cycle)
it n e
mlane r7 , r10 , r2 , r7 @ R7-=W5*ROWr16[3]=b3 (ROWr16[3] must be the second arg, to have the possibility to save 1 cycle)
@@ at this point, R0=b0, R1=b1, R2 (free), R3=ROWr32[2], R4=ROWr32[3],
@ -148,19 +150,23 @@ __b_evaluation:
@@ MAC16(b3, -W1, row[7]);
@@ MAC16(b1, -W5, row[7]);
mov r3 , r3 , a s r #16 @ R3=ROWr16[5]
teq r3 , #0 @ if null avoid muls
teq r3 , #0 @ if null avoid muls
it n e
mlane r0 , r10 , r3 , r0 @ R0+=W5*ROWr16[5]=b0
mov r4 , r4 , a s r #16 @ R4=ROWr16[7]
itttt n e
mlane r5 , r11 , r3 , r5 @ R5+=W7*ROWr16[5]=b2
mlane r7 , r9 , r3 , r7 @ R7+=W3*ROWr16[5]=b3
rsbne r3 , r3 , #0 @ R3=-ROWr16[5]
mlane r1 , r8 , r3 , r1 @ R7-=W1*ROWr16[5]=b1
@@ R3 is free now
teq r4 , #0 @ if null avoid muls
teq r4 , #0 @ if null avoid muls
itttt n e
mlane r0 , r11 , r4 , r0 @ R0+=W7*ROWr16[7]=b0
mlane r5 , r9 , r4 , r5 @ R5+=W3*ROWr16[7]=b2
rsbne r4 , r4 , #0 @ R4=-ROWr16[7]
mlane r7 , r8 , r4 , r7 @ R7-=W1*ROWr16[7]=b3
it n e
mlane r1 , r10 , r4 , r1 @ R1-=W5*ROWr16[7]=b1
@@ R4 is free now
__end_b_evaluation :
@ -204,16 +210,19 @@ __a_evaluation:
@@ a2 -= W4*row[4]
@@ a3 += W4*row[4]
ldrsh r11 , [ r14 , #8 ] @ R11=ROWr16[4]
teq r11 , #0 @ if null avoid muls
teq r11 , #0 @ if null avoid muls
it n e
mulne r11 , r9 , r11 @ R11=W4*ROWr16[4]
@@ R9 is free now
ldrsh r9 , [ r14 , #12 ] @ R9=ROWr16[6]
itttt n e
addne r6 , r6 , r11 @ R6+=W4*ROWr16[4] (a0)
subne r2 , r2 , r11 @ R2-=W4*ROWr16[4] (a1)
subne r3 , r3 , r11 @ R3-=W4*ROWr16[4] (a2)
addne r4 , r4 , r11 @ R4+=W4*ROWr16[4] (a3)
@@ W6 alone is no more useful, save W2*ROWr16[6] in it instead
teq r9 , #0 @ if null avoid muls
teq r9 , #0 @ if null avoid muls
itttt n e
mulne r11 , r10 , r9 @ R11=W6*ROWr16[6]
addne r6 , r6 , r11 @ R6+=W6*ROWr16[6] (a0)
mulne r10 , r8 , r9 @ R10=W2*ROWr16[6]
@ -222,6 +231,7 @@ __a_evaluation:
@@ a1 -= W2*row[6];
@@ a2 += W2*row[6];
subne r4 , r4 , r11 @ R4-=W6*ROWr16[6] (a3)
itt n e
subne r2 , r2 , r10 @ R2-=W2*ROWr16[6] (a1)
addne r3 , r3 , r10 @ R3+=W2*ROWr16[6] (a2)
@ -323,10 +333,12 @@ __b_evaluation2:
ldrsh r2 , [ r14 , #48 ]
mul r7 , r11 , r7 @ R7=W7*ROWr16[1]=b3 (ROWr16[1] must be the second arg, to have the possibility to save 1 cycle)
teq r2 , #0 @ if 0, then avoid muls
itttt n e
mlane r0 , r9 , r2 , r0 @ R0+=W3*ROWr16[3]=b0 (ROWr16[3] must be the second arg, to have the possibility to save 1 cycle)
rsbne r2 , r2 , #0 @ R2=-ROWr16[3]
mlane r1 , r11 , r2 , r1 @ R1-=W7*ROWr16[3]=b1 (ROWr16[3] must be the second arg, to have the possibility to save 1 cycle)
mlane r5 , r8 , r2 , r5 @ R5-=W1*ROWr16[3]=b2 (ROWr16[3] must be the second arg, to have the possibility to save 1 cycle)
it n e
mlane r7 , r10 , r2 , r7 @ R7-=W5*ROWr16[3]=b3 (ROWr16[3] must be the second arg, to have the possibility to save 1 cycle)
@@ at this point, R0=b0, R1=b1, R2 (free), R3 (free), R4 (free),
@ -342,18 +354,22 @@ __b_evaluation2:
@@ MAC16(b1, -W5, col[7x8]);
ldrsh r3 , [ r14 , #80 ] @ R3=COLr16[5x8]
teq r3 , #0 @ if 0 then avoid muls
itttt n e
mlane r0 , r10 , r3 , r0 @ R0+=W5*ROWr16[5x8]=b0
mlane r5 , r11 , r3 , r5 @ R5+=W7*ROWr16[5x8]=b2
mlane r7 , r9 , r3 , r7 @ R7+=W3*ROWr16[5x8]=b3
rsbne r3 , r3 , #0 @ R3=-ROWr16[5x8]
ldrsh r4 , [ r14 , #112 ] @ R4=COLr16[7x8]
it n e
mlane r1 , r8 , r3 , r1 @ R7-=W1*ROWr16[5x8]=b1
@@ R3 is free now
teq r4 , #0 @ if 0 then avoid muls
itttt n e
mlane r0 , r11 , r4 , r0 @ R0+=W7*ROWr16[7x8]=b0
mlane r5 , r9 , r4 , r5 @ R5+=W3*ROWr16[7x8]=b2
rsbne r4 , r4 , #0 @ R4=-ROWr16[7x8]
mlane r7 , r8 , r4 , r7 @ R7-=W1*ROWr16[7x8]=b3
it n e
mlane r1 , r10 , r4 , r1 @ R1-=W5*ROWr16[7x8]=b1
@@ R4 is free now
__end_b_evaluation2 :
@ -390,15 +406,18 @@ __a_evaluation2:
@@ a3 += W4*row[4]
ldrsh r11 , [ r14 , #64 ] @ R11=ROWr16[4]
teq r11 , #0 @ if null avoid muls
itttt n e
mulne r11 , r9 , r11 @ R11=W4*ROWr16[4]
@@ R9 is free now
addne r6 , r6 , r11 @ R6+=W4*ROWr16[4] (a0)
subne r2 , r2 , r11 @ R2-=W4*ROWr16[4] (a1)
subne r3 , r3 , r11 @ R3-=W4*ROWr16[4] (a2)
ldrsh r9 , [ r14 , #96 ] @ R9=ROWr16[6]
it n e
addne r4 , r4 , r11 @ R4+=W4*ROWr16[4] (a3)
@@ W6 alone is no more useful, save W2*ROWr16[6] in it instead
teq r9 , #0 @ if null avoid muls
itttt n e
mulne r11 , r10 , r9 @ R11=W6*ROWr16[6]
addne r6 , r6 , r11 @ R6+=W6*ROWr16[6] (a0)
mulne r10 , r8 , r9 @ R10=W2*ROWr16[6]
@ -407,6 +426,7 @@ __a_evaluation2:
@@ a1 -= W2*row[6];
@@ a2 += W2*row[6];
subne r4 , r4 , r11 @ R4-=W6*ROWr16[6] (a3)
itt n e
subne r2 , r2 , r10 @ R2-=W2*ROWr16[6] (a1)
addne r3 , r3 , r10 @ R3+=W2*ROWr16[6] (a2)
__end_a_evaluation2 :