-- rdpmc_all(uint64_t[nfixed+ngeneral] *dst)                                                                                          
--                                                                                                                                    
-- Generate code to read performance monitoring counter values into dst.                                                              
--                                                                                                                                    
-- The generated code is specialized based on the number of fixed- and                                                                
-- general-purpose performance counters that are available in                                                                         
-- hardware. These numbers are determined separately via the CPUID instruction.                                                       
--                                                                                                                                    
-- In practice on Sandy Bridge - Skylake CPUs the number of                                                                           
-- fixed-purpose registers is three and the number of general-purpose                                                                 
-- registers is either four (hyperthreads enabled) or eight                                                                           
-- (hyperthreads disabled).                                                                                                           
function rdpmc_multi (Dst, nfixed, ngeneral)
   local offset = 0
   -- Read a PMC register value into the next slot of the destination buffer                                                          
   local function rdpmc (isfixed, index)
      local arg = (isfixed and 0x40000000 or 0) + index
      |  mov ecx, arg
      |  rdpmc
      |  mov [edi+offset], eax
      |  mov [edi+offset+4], edx
      offset = offset + 8
   end
   for i = 0, nfixed-1   do rdpmc(true, i)  end
   for i = 0, ngeneral-1 do rdpmc(false, i) end
   |  ret
end