Re: [Tarantool-patches] [PATCH v1 luajit 13/41] perf: adjust mandelbrot-bit in LuaJIT-benches

[Sergey Kaplun via Tarantool-patches <tarantool-patches@dev.tarantool.org>]

Hi, Sergey!
Thanks for the review!
Fixed your comments and answered your questions below.

On 17.11.25, Sergey Bronnikov wrote:
> Hi, Sergey!
> 
> thanks for the patch! See my comments.
> 
> Sergey
> 
> On 10/24/25 13:50, Sergey Kaplun wrote:
> > This patch adjusts the aforementioned test to use the benchmark
> > framework introduced before. The default arguments are adjusted
> > according to the <PARAM_x86.txt> file. The arguments to the script still
> > can be provided in the command line run.
> >
> > The output is redirected to /dev/null. The check is skipped since it is
> > very inconvenient to check the binary output, especially since it may be
> > configured by the parameter.
> > ---
> >   perf/LuaJIT-benches/mandelbrot-bit.lua | 86 +++++++++++++++++---------
> >   1 file changed, 57 insertions(+), 29 deletions(-)
> >
> > diff --git a/perf/LuaJIT-benches/mandelbrot-bit.lua b/perf/LuaJIT-benches/mandelbrot-bit.lua
> > index 91d96975..a6b5e1f8 100644
> > --- a/perf/LuaJIT-benches/mandelbrot-bit.lua
> > +++ b/perf/LuaJIT-benches/mandelbrot-bit.lua

Added the comment with the short benchmark description, as we
discussed offline:

===================================================================
diff --git a/perf/LuaJIT-benches/mandelbrot-bit.lua b/perf/LuaJIT-benches/mandelbrot-bit.lua
index 88df480e..53c3ad4e 100644
--- a/perf/LuaJIT-benches/mandelbrot-bit.lua
+++ b/perf/LuaJIT-benches/mandelbrot-bit.lua
@@ -1,3 +1,10 @@
+-- The benchmark to check the performance of multiple inner loops
+-- with arithmetic operations. Bit variation. Calculates the
+-- Mandelbrot Set on a bitmap and dumps output in the portable
+-- bitmap format.
+-- For the details see:
+-- https://benchmarksgame-team.pages.debian.net/benchmarksgame/description/mandelbrot.html
+
 local bit = require("bit")
 
 local bench = require("bench").new(arg)
===================================================================

> > @@ -1,33 +1,61 @@
> > -
> >   local bit = require("bit")

<snipped>

> > +local bench = require("bench").new(arg)
> > +
> > +local N = tonumber(arg and arg[1]) or 5000
> > +
> > +local function payload()
> > +  -- These functions must not be an upvalue but the stack slot.
> please add here details about performance impact

I suppose this is related to the open upvalue fetching overhead, but I'm
not sure. I suggest to postponing this research out of scope of this
issue.

> > +  local N = N
> > +  local bor, band = bit.bor, bit.band
> > +  local shl, shr, rol = bit.lshift, bit.rshift, bit.rol
> > +  local write, char, unpack = io.write, string.char, unpack
> > +
> > +  local M, buf = 2/N, {}
> > +  write("P4\n", N, " ", N, "\n")
> > +  for y=0,N-1 do
> please add spaces here and below

Fixed the formating as you suggested:

===================================================================
diff --git a/perf/LuaJIT-benches/mandelbrot-bit.lua b/perf/LuaJIT-benches/mandelbrot-bit.lua
index a6b5e1f8..88df480e 100644
--- a/perf/LuaJIT-benches/mandelbrot-bit.lua
+++ b/perf/LuaJIT-benches/mandelbrot-bit.lua
@@ -11,30 +11,30 @@ local function payload()
   local shl, shr, rol = bit.lshift, bit.rshift, bit.rol
   local write, char, unpack = io.write, string.char, unpack
 
-  local M, buf = 2/N, {}
+  local M, buf = 2 / N, {}
   write("P4\n", N, " ", N, "\n")
-  for y=0,N-1 do
-    local Ci, b, p = y*M-1, -16777216, 0
-    local Ciq = Ci*Ci
-    for x=0,N-1,2 do
-      local Cr, Cr2 = x*M-1.5, (x+1)*M-1.5
-      local Zr, Zi, Zrq, Ziq = Cr, Ci, Cr*Cr, Ciq
-      local Zr2, Zi2, Zrq2, Ziq2 = Cr2, Ci, Cr2*Cr2, Ciq
+  for y = 0, N - 1 do
+    local Ci, b, p = y * M - 1, -16777216, 0
+    local Ciq = Ci * Ci
+    for x = 0, N - 1, 2 do
+      local Cr, Cr2 = x * M - 1.5, (x + 1) * M - 1.5
+      local Zr, Zi, Zrq, Ziq = Cr, Ci, Cr * Cr, Ciq
+      local Zr2, Zi2, Zrq2, Ziq2 = Cr2, Ci, Cr2 * Cr2, Ciq
       b = rol(b, 2)
-      for i=1,49 do
-        Zi = Zr*Zi*2 + Ci; Zi2 = Zr2*Zi2*2 + Ci
-        Zr = Zrq-Ziq + Cr; Zr2 = Zrq2-Ziq2 + Cr2
-        Ziq = Zi*Zi; Ziq2 = Zi2*Zi2
-        Zrq = Zr*Zr; Zrq2 = Zr2*Zr2
-        if band(b, 2) ~= 0 and Zrq+Ziq > 4.0 then b = band(b, -3) end
-        if band(b, 1) ~= 0 and Zrq2+Ziq2 > 4.0 then b = band(b, -2) end
+      for i = 1, 49 do
+        Zi = Zr * Zi * 2 + Ci; Zi2 = Zr2 * Zi2 * 2 + Ci
+        Zr = Zrq - Ziq + Cr; Zr2 = Zrq2 - Ziq2 + Cr2
+        Ziq = Zi * Zi; Ziq2 = Zi2 * Zi2
+        Zrq = Zr * Zr; Zrq2 = Zr2 * Zr2
+        if band(b, 2) ~= 0 and Zrq + Ziq > 4.0 then b = band(b, -3) end
+        if band(b, 1) ~= 0 and Zrq2 + Ziq2 > 4.0 then b = band(b, -2) end
         if band(b, 3) == 0 then break end
       end
       if b >= 0 then p = p + 1; buf[p] = b; b = -16777216; end
     end
     if b ~= -16777216 then
       if band(N, 1) ~= 0 then b = shr(b, 1) end
-      p = p + 1; buf[p] = shl(b, 8-band(N, 7))
+      p = p + 1; buf[p] = shl(b, 8 - band(N, 7))
     end
     write(char(unpack(buf, 1, p)))
   end
===================================================================

> > +    local Ci, b, p = y*M-1, -16777216, 0
> > +    local Ciq = Ci*Ci
> > +    for x=0,N-1,2 do
> > +      local Cr, Cr2 = x*M-1.5, (x+1)*M-1.5
> > +      local Zr, Zi, Zrq, Ziq = Cr, Ci, Cr*Cr, Ciq
> > +      local Zr2, Zi2, Zrq2, Ziq2 = Cr2, Ci, Cr2*Cr2, Ciq
> > +      b = rol(b, 2)
> > +      for i=1,49 do
> > +        Zi = Zr*Zi*2 + Ci; Zi2 = Zr2*Zi2*2 + Ci
> > +        Zr = Zrq-Ziq + Cr; Zr2 = Zrq2-Ziq2 + Cr2
> > +        Ziq = Zi*Zi; Ziq2 = Zi2*Zi2
> > +        Zrq = Zr*Zr; Zrq2 = Zr2*Zr2
> > +        if band(b, 2) ~= 0 and Zrq+Ziq > 4.0 then b = band(b, -3) end
> > +        if band(b, 1) ~= 0 and Zrq2+Ziq2 > 4.0 then b = band(b, -2) end
> > +        if band(b, 3) == 0 then break end
> > +      end
> > +      if b >= 0 then p = p + 1; buf[p] = b; b = -16777216; end
> >       end
> > -    if b >= 0 then p = p + 1; buf[p] = b; b = -16777216; end
> > -  end
> > -  if b ~= -16777216 then
> > -    if band(N, 1) ~= 0 then b = shr(b, 1) end
> > -    p = p + 1; buf[p] = shl(b, 8-band(N, 7))
> > +    if b ~= -16777216 then
> > +      if band(N, 1) ~= 0 then b = shr(b, 1) end
> > +      p = p + 1; buf[p] = shl(b, 8-band(N, 7))
> > +    end
> > +    write(char(unpack(buf, 1, p)))
> >     end
> > -  write(char(unpack(buf, 1, p)))
> >   end

<snipped>

-- 
Best regards,
Sergey Kaplun