Question on SCL/SCLU instruction operation
ozpropdev
Posts: 2,793
in Propeller 2
Hi All
Can someone confirm the operation of the SCL/SCLU instruction.
It's my understanding the SCLU is basically a MUL instruction followed by a SAR 16 bits.
I assume the same goes for SCL, a MULS followed by a SAR 16 bits.
Is that correct?
If so, here is what I am seeing.
SCL seems to give a result tat is out by 2 bits.
* Test1 SCLU with unsigned values = Ok.
* Test2 SCL with unsigned values = Needs extra SAR 2 to be correct.
* Test3 SCLU with 1 signed value = Fails as expected.
* Test4 SCL with 1 signed value = Needs extra SAR 2 to be correct.
Can someone confirm the operation of the SCL/SCLU instruction.
It's my understanding the SCLU is basically a MUL instruction followed by a SAR 16 bits.
I assume the same goes for SCL, a MULS followed by a SAR 16 bits.
Is that correct?
If so, here is what I am seeing.
SCL seems to give a result tat is out by 2 bits.
* Test1 SCLU with unsigned values = Ok.
* Test2 SCL with unsigned values = Needs extra SAR 2 to be correct.
* Test3 SCLU with 1 signed value = Fails as expected.
* Test4 SCL with 1 signed value = Needs extra SAR 2 to be correct.
Test #1 R0 = $000004D2 0000001234 R1 = $000004D2 0000001234 MUL R0,R1 = $00173C44 0001522756 SAR 16 = $00000017 0000000023 SCLU R0,R1 = $00000017 0000000023 * SAR 2 = $00000005 0000000005 ========================================== Test #2 R0 = $000004D2 0000001234 R1 = $000004D2 0000001234 MULS R0,R1 = $00173C44 0001522756 SAR 16 = $00000017 0000000023 SCL R0,R1 = $0000005C 0000000092 * SAR 2 = $00000017 0000000023 ========================================== Test #3 R0 = $FFFFFB2E -0000001234 R1 = $000004D2 0000001234 MUL R0,R1 = $04BAC3BC 0079348668 SAR 16 = $000004BA 0000001210 SCLU R0,R1 = $000004BA 0000001210 * SAR 2 = $0000012E 0000000302 ========================================== Test #4 R0 = $FFFFFB2E -0000001234 R1 = $000004D2 0000001234 MULS R0,R1 = $FFE8C3BC -0001522756 SAR 16 = $FFFFFFE8 -0000000024 SCL R0,R1 = $FFFFFFA3 -0000000093 * SAR 2 = $FFFFFFE8 -0000000024
Comments
SCLU multiplies D[15:0] by S[15:0] and substitutes product >> 16 into the next instruction's S value. For example:
SCLU A,B 'Add (A[15:0] * B[15:0]) >> 16 into C
ADD C,0
SCL multiplies D[15:0] by S[15:0] and substitutes product ~> 14 into the next instruction's S value (~> is like SAR). For example:
SCL A,B 'Add (A[15:0] * B[15:0]) ~> 14 into C
ADD C,0
This allows fractional multiply with one factor as a signed 16bit signal and the other factor as a coefficient in the range -2.0 .. +1.99994.
So the SAR by 2 you do in the Tests, is normally done by the correct scaling of the coefficient. $4000 correspond to 1.0.
Fractional multiply is important for DSP application where you work mostly in the range -1.0 to +1.0, Chip wanted to have the scaling so that 1.0 is possible. And for FIR filters the coefficients can go bigger than 1.0, so +-2 is a good decision.
The downside is that the resolution of the fractional part is only 14 bits now. For more resolution we would need a better multiplier with 18x18 bits or more.
Andy