SPA 5.0:
{@{!}Pg} IMADSP.saext.sbext.scext Rd{.CC},Ra,Sb,Sc {&req_6} {&rdN} {&wrN} {?sched} ;
.saext : { .U32, .S32, .U24. .S24, .U16H0, .S16H0, .U16H1, .S16H1 } // operand extract modifier
.sbext : { .U24, .S24, .U16H0, .S16H0 } // operand extract modifier
.scext : { .U32, .S32, .U24, .S24, .U16H0, .S16H0 } // operand extract modifier
IMADSP allows the following sources Sb,Sc:
Sb(register), Sc(register)
Sb(constant with immediate address), Sc(register)
Sb(#Imm20), Sc(register)
Sb(register), Sc(constant with immediate address)
Multiply and add sources (smaller than or equal to 32 bits) to produce single precision (32 bits) result.
IMADSP operates as follows. First the operands are extracted/sign extended based on the saext/sbext/scext modifiers. These ext modifiers define if the data container is the whole 32 bits or 24 lsb bits or top 16 bits or bottom 16 bits.
saext/sbext/scext are the operand extract modifiers to extract 16/24/32 bit part selects out of A/B/C operands (similar to VMAD). In case of saext/sbext, these modifiers also define the signed/unsigned behavior of the operands.
After extracting the correct bits, the operands are extended (based on signed or unsigned) to full 32 bits. The sign of operand C is signed if any of the Ra/Sb is signed. Thus only following multiplication combinations are possible (post extention)
tmp = (RaExtracted * SbExtracted) + RcEtracted: tmp = (U32 * U32) + U32 // intermediate unsigned; final unsigned tmp = (U32 * S32) + S32 // intermediate signed; final signed tmp = (S32 * U32) + S32 // intermediate signed; final signed tmp = (S32 * S32) + S32 // intermediate signed; final signed Note: The intermediate sign of the multiply (Ra*Sb) will be unsigned if Both Ra and Sb are unsigned. Otherwise, the intermediate sign will be signed.
The sign of source C and final destination will be same as the intermediate sign.
IMADSP.U24.U24.U16H0 R0,R1,R2,R3; IMADSP.S24.S24.S16H0 R0,R1,R2,R3;