FillChar and StringOfChar under Delphi 10.2 for Win64 Release Target

Question

I have a question about a specific programming problem in Delphi 10.2 Pascal programming language.

The StringOfChar and FillChar don’t work properly under Win64 Release build on CPUs released before year 2012.

• Expected result of FillChar is just plain sequence of just repeating 8-bit characters in a given memory buffer.

• Expected result of StringOfChar is the same, but the result is stored inside a string type.

But, in fact, when I compile our applications that worked in Delphi prior to 10.2 by the 10.2 version of Delphi, our applications compiled for Win64 stop working properly on CPUs released before year 2012.

The StringOfChar and FillChar don’t work properly – they return a string of different characters, although in a repeating pattern – not just a sequence of the same character as they should.

Here is the minimal code enough to demonstrate the issue. Please note that the length of the sequence should be at least 16 characters, and the character should not be nul (#0). The code is below:

procedure TestStringOfChar;
var
  a: AnsiString;
  ac: AnsiChar;
begin
  ac := #1;
  a := StringOfChar(ac, 43);
  if a <> #1#1#1#1#1#1#1#1#1#1#1#1#1#1#1#1#1#1#1#1#1#1#1#1#1#1#1#1#1#1#1#1#1#1#1#1#1#1#1#1#1#1#1 then
  begin
    raise Exception.Create('ANSI StringOfChar Failed!!');
  end;
end;

I know that there are lots of Delphi programmers at StackOverflow. Are you experiencing the same problem? If yes, how you resolve it? What is the solution? By the way, I have contacted the developers of Delphi but they didn’t confirm nor deny the issue so far. I'm using Embarcadero Delphi 10.2 Version 25.0.26309.314.

Update:

If your CPU is manufactured in 2012 or later, additionally include the following lines before calling StringOfChar to reproduce the issue:

const
  ERMSBBit    = 1 shl 9; //$0200
begin
  CPUIDTable[7].EBX := CPUIDTable[7].EBX and not ERMSBBit;

As about the April 2017 RAD Studio 10.2 Hotfix for Toolchain Issues - have tried with it and without it - it didn't help. The issue exists regardless of the Hotfix.

Update #2

Embarcadero has confirmed and resolved this issue on 08/Aug/17 6:03 PM. So, in Delphi 10.2 Tokyo Release 1 (released on August 8, 2017) this bug is fixed.

Answer 1

StringOfChar(A: AnsiChar, count) uses FillChar under the hood.

You can use the following code to fix the issue:

(*******************************************************
 System.FastSystem
 A fast drop-in addition to speed up function in system.pas
 It should compile and run in XE2 and beyond.
 Alpha version 0.5, fully tested in Win64
 (c) Copyright 2016 J. Bontes
   This Source Code Form is subject to the terms of the
   Mozilla Public License, v. 2.0.
   If a copy of the MPL was not distributed with this file,
   You can obtain one at http://mozilla.org/MPL/2.0/.
********************************************************
FillChar code is an altered version FillCharsse2 SynCommons.pas
which is part of Synopse framework by Arnaud Bouchez
********************************************************
Changelog
0.5 Initial version:
********************************************************)

unit FastSystem;

interface

procedure FillChar(var Dest; Count: NativeInt; Value: ansichar); inline; overload;
procedure FillChar(var Dest; Count: NativeInt; Value: Byte); overload;
procedure FillMemory(Destination: Pointer; Length: NativeUInt; Fill: Byte); inline;
{$EXTERNALSYM FillMemory}
procedure ZeroMemory(Destination: Pointer; Length: NativeUInt); inline;
{$EXTERNALSYM ZeroMemory}

implementation

procedure FillChar(var Dest; Count: NativeInt; Value: ansichar); inline; overload;
begin
  FillChar(Dest, Count, byte(Value));
end;

procedure FillMemory(Destination: Pointer; Length: NativeUInt; Fill: Byte);
begin
  FillChar(Destination^, Length, Fill);
end;

procedure ZeroMemory(Destination: Pointer; Length: NativeUInt); inline;
begin
  FillChar(Destination^, Length, 0);
end;

//This code is 3x faster than System.FillChar on x64.

{$ifdef CPUX64}
procedure FillChar(var Dest; Count: NativeInt; Value: Byte);
//rcx = dest
//rdx=count
//r8b=value
asm
              .noframe
              .align 16
              movzx r8,r8b           //There's no need to optimize for count <= 3
              mov rax,$0101010101010101
              mov r9d,edx
              imul rax,r8            //fill rax with value.
              cmp rdx,59             //Use simple code for small blocks.
              jl  @Below32
@Above32:     mov r11,rcx
              mov r8b,7              //code shrink to help alignment.
              lea r9,[rcx+rdx]       //r9=end of array
              sub rdx,8
              rep mov [rcx],rax
              add rcx,8
              and r11,r8             //and 7 See if dest is aligned
              jz @tail
@NotAligned:  xor rcx,r11            //align dest
              lea rdx,[rdx+r11]
@tail:        test r9,r8             //and 7 is tail aligned?
              jz @alignOK
@tailwrite:   mov [r9-8],rax         //no, we need to do a tail write
              and r9,r8              //and 7
              sub rdx,r9             //dec(count, tailcount)
@alignOK:     mov r10,rdx
              and edx,(32+16+8)      //count the partial iterations of the loop
              mov r8b,64             //code shrink to help alignment.
              mov r9,rdx
              jz @Initloop64
@partialloop: shr r9,1              //every instruction is 4 bytes
              lea r11,[rip + @partial +(4*7)] //start at the end of the loop
              sub r11,r9            //step back as needed
              add rcx,rdx            //add the partial loop count to dest
              cmp r10,r8             //do we need to do more loops?
              jmp r11                //do a partial loop
@Initloop64:  shr r10,6              //any work left?
              jz @done               //no, return
              mov rdx,r10
              shr r10,(19-6)         //use non-temporal move for > 512kb
              jnz @InitFillHuge
@Doloop64:    add rcx,r8
              dec edx
              mov [rcx-64+00H],rax
              mov [rcx-64+08H],rax
              mov [rcx-64+10H],rax
              mov [rcx-64+18H],rax
              mov [rcx-64+20H],rax
              mov [rcx-64+28H],rax
              mov [rcx-64+30H],rax
              mov [rcx-64+38H],rax
              jnz @DoLoop64
@done:        rep ret
              //db $66,$66,$0f,$1f,$44,$00,$00 //nop7
@partial:     mov [rcx-64+08H],rax
              mov [rcx-64+10H],rax
              mov [rcx-64+18H],rax
              mov [rcx-64+20H],rax
              mov [rcx-64+28H],rax
              mov [rcx-64+30H],rax
              mov [rcx-64+38H],rax
              jge @Initloop64        //are we done with all loops?
              rep ret
              db $0F,$1F,$40,$00
@InitFillHuge:
@FillHuge:    add rcx,r8
              dec rdx
              db $48,$0F,$C3,$41,$C0 // movnti  [rcx-64+00H],rax
              db $48,$0F,$C3,$41,$C8 // movnti  [rcx-64+08H],rax
              db $48,$0F,$C3,$41,$D0 // movnti  [rcx-64+10H],rax
              db $48,$0F,$C3,$41,$D8 // movnti  [rcx-64+18H],rax
              db $48,$0F,$C3,$41,$E0 // movnti  [rcx-64+20H],rax
              db $48,$0F,$C3,$41,$E8 // movnti  [rcx-64+28H],rax
              db $48,$0F,$C3,$41,$F0 // movnti  [rcx-64+30H],rax
              db $48,$0F,$C3,$41,$F8 // movnti  [rcx-64+38H],rax
              jnz @FillHuge
@donefillhuge:mfence
              rep ret
              db $0F,$1F,$44,$00,$00  //db $0F,$1F,$40,$00
@Below32:     and  r9d,not(3)
              jz @SizeIs3
@FillTail:    sub   edx,4
              lea   r10,[rip + @SmallFill + (15*4)]
              sub   r10,r9
              jmp   r10
@SmallFill:   rep mov [rcx+56], eax
              rep mov [rcx+52], eax
              rep mov [rcx+48], eax
              rep mov [rcx+44], eax
              rep mov [rcx+40], eax
              rep mov [rcx+36], eax
              rep mov [rcx+32], eax
              rep mov [rcx+28], eax
              rep mov [rcx+24], eax
              rep mov [rcx+20], eax
              rep mov [rcx+16], eax
              rep mov [rcx+12], eax
              rep mov [rcx+08], eax
              rep mov [rcx+04], eax
              mov [rcx],eax
@Fallthough:  mov [rcx+rdx],eax  //unaligned write to fix up tail
              rep ret

@SizeIs3:     shl edx,2           //r9 <= 3  r9*4
              lea r10,[rip + @do3 + (4*3)]
              sub r10,rdx
              jmp r10
@do3:         rep mov [rcx+2],al
@do2:         mov [rcx],ax
              ret
@do1:         mov [rcx],al
              rep ret
@do0:         rep ret
end;
{$endif}

The easiest way to fix your issue is to Download Mormot and include SynCommon.pas into your project. This will patch System.FillChar to the above code and include a couple of other performance improvements as well.

Note that you don't need all of Mormot, just SynCommons by itself.

Answer 2

I took the test case from the FastCode Challenge - http://fastcode.sourceforge.net/

I have compiled the FillChar testing tool under Win64, and removed all 32-bit versions of FillChar present in the test.

I have left 2 versions of 64-bit FillChar:

1. FC_TokyoBugfixAVXEx - the one present in Delphi Tokyo 64-bit, with bugs fixed and AVX registers added. There is branching to detect ERMSB, AVX1 and AVX2 CPU capabilities. This branching happens on each FillChar call. There is no entry point patching or function address mapping.
2. FillChar_J_Bontes - another version of FillChar, the function from System.FastSystem that you have posted here.

I didn't test vanilla FillChar from Delphi Tokyo, because it contains a bug described in my initial post, and it improperly handles ERMSB.

Kaby Lake - i7-7700K

First column is the alignment of the function. Next 4 columns are results of various tests, lower is better. There are 4 tests in total. First test operates with smaller block, second with larger, and so on. Last column is a weighted summary of all tests.

The CPU in the first test is Kaby Lake i7-7700K (January 2017). Frequency 4.2 GHz (turbo frequency up to 4.5 GHz), L2 cache 4 × 256 KB, L3 cache 8 MB.

Ivy Bridge - E5-2603 v2

Here are the results of a second test, on a previous microarchitecture: Xeon E5-2603 v2 "Ivy Bridge" (September 2013), frequency 1.8 GHz, L2 Cache 4 × 256 KB, L3 Cache 10 MB, RAM 4 × DDR3-1333.

Ivy Bridge - E5-2643 v2

Here are the test results on a third set of hardware: Intel Xeon E5-2643 v2 (September 2013), frequency 3.5 GHz, L2 Cache 6 × 256 KB, L3 Cache 25 MB, RAM 4 × DDR3-1600.

Intel Core i9 7900X

Here are the test results on a fourth set of hardware: Intel Core i9 7900X (June 2017), frequency 3.3 GHz (turbo frequency up to 4.5 GHz), L2 Cache 10 × 1024 KB, L3 Cache 13.75 MB, RAM 4 × DDR4-2134.