x86 MUL Instruction from VS 2008/2010

Question

Will modern (2008/2010) incantations of Visual Studio or Visual C++ Express produce x86 MUL instructions (unsigned multiply) in the compiled code? I cannot seem to find or

Answer 1

There's three different types of multiply instructions on x86. The first is MUL reg, which does an unsigned multiply of EAX by reg and puts the (64-bit) result into EDX:EAX. The second is IMUL reg, which does the same with a signed multiply. The third type is either IMUL reg1, reg2 (multiplies reg1 with reg2 and stores the 32-bit result into reg1) or IMUL reg1, reg2, imm (multiplies reg2 by imm and stores the 32-bit result into reg1).

Since in C, multiplies of two 32-bit values produce 32-bit results, compilers normally use the third type (signedness doesn't matter, the low 32 bits agree between signed and unsigned 32x32 multiplies). VC++ will generate the "long multiply" versions of MUL/IMUL if you actually use the full 64-bit results, e.g. here:

unsigned long long prod(unsigned int a, unsigned int b)
{
  return (unsigned long long) a * b;
}

The 2-operand (and 3-operand) versions of IMUL are faster than the one-operand versions simply because they don't produce a full 64-bit result. Wide multipliers are large and slow; it's much easier to build a smaller multiplier and synthesize long multiplies using Microcode if necessary. Also, MUL/IMUL writes two registers, which again is usually resolved by breaking it into multiple instructions internally - it's much easier for the instruction reordering hardware to keep track of two dependent instructions that each write one register (most x86 instructions look like that internally) than it is to keep track of one instruction that writes two.