dmd.backend.cgxmm

Compiler implementation of the D programming language.

Members

Functions

cdvecfill
void cdvecfill(ref CodeBuilder cdb, elem* e, regm_t* pretregs)

Generate code for OPvecfill (broadcast). OPvecfill takes the single value in e1 and fills the vector type with it.

cdvecsto
void cdvecsto(ref CodeBuilder cdb, elem* e, regm_t* pretregs)

Generate code for vector "store" operations. The tree e must look like: (op1 OPvecsto (op OPparam op2)) where op is the store instruction STOxxxx.

checkSetVex
void checkSetVex(code* c, tym_t ty)

Determine if operation should be rewritten as a VEX operation; and do so.

checkSetVex3
void checkSetVex3(code* c)

VEX prefixes can be 2 or 3 bytes. If it must be 3 bytes, set the CFvex3 flag.

isXMMstore
bool isXMMstore(opcode_t op)

Is operator a store operator?

orthxmm
void orthxmm(ref CodeBuilder cdb, elem* e, regm_t* pretregs)

Do simple orthogonal operators for XMM registers.

xmmIsAligned
bool xmmIsAligned(elem* e)

Determine if lvalue e is a vector aligned on a 16/32 byte boundary. Assume it to be aligned unless can prove it is not.

xmmcnvt
void xmmcnvt(ref CodeBuilder cdb, elem* e, regm_t* pretregs)

Generate code for conversion using SSE2 instructions.

xmmeq
void xmmeq(ref CodeBuilder cdb, elem* e, opcode_t op, elem* e1, elem* e2, regm_t* pretregs)

Generate code for an assignment using XMM registers.

xmmload
opcode_t xmmload(tym_t tym, bool aligned)

Get correct load operator based on type. It is important to use the right one even if the number of bits moved is the same, as there are performance consequences for using the wrong one.

xmmneg
void xmmneg(ref CodeBuilder cdb, elem* e, regm_t* pretregs)

Negate operator

xmmopass
void xmmopass(ref CodeBuilder cdb, elem* e, regm_t* pretregs)

Generate code for op=

xmmpost
void xmmpost(ref CodeBuilder cdb, elem* e, regm_t* pretregs)

Generate code for post increment and post decrement.

xmmstore
opcode_t xmmstore(tym_t tym, bool aligned)

Get correct store operator based on type.

Meta