After a long afternoon of debugging, Erick and I have managed to fix the last bug preventing us from completing the bootstrap. Tachyon can now compile and run this benchmark once bootstrapped. There are still some issues to fix. Notably, we are not compiling the standard library as of now, because the heap size is limited under 32 bits. Still, this is probably a Champagne-worthy victory ;)
Once the 64 bit bugs are fixed, we will work on compiling the standard library. With some luck, we might be able to run the shell pretty soon.
- Maxime
Afficher les réponses par date
On 2011-03-21, at 17:31 , chevalma@iro.umontreal.ca wrote:
After a long afternoon of debugging, Erick and I have managed to fix the last bug preventing us from completing the bootstrap. Tachyon can now compile and run this benchmark once bootstrapped. There are still some issues to fix. Notably, we are not compiling the standard library as of now, because the heap size is limited under 32 bits. Still, this is probably a Champagne-worthy victory ;)
Congratulations to both of you! This is really exciting news!
Once the 64 bit bugs are fixed, we will work on compiling the standard library. With some luck, we might be able to run the shell pretty soon.
Good luck with the rest.
Bruno
On 2011-03-21, at 5:34 PM, Bruno Dufour wrote:
On 2011-03-21, at 17:31 , chevalma@iro.umontreal.ca wrote:
After a long afternoon of debugging, Erick and I have managed to fix the last bug preventing us from completing the bootstrap. Tachyon can now compile and run this benchmark once bootstrapped. There are still some issues to fix. Notably, we are not compiling the standard library as of now, because the heap size is limited under 32 bits. Still, this is probably a Champagne-worthy victory ;)
Congratulations to both of you! This is really exciting news!
Yes. Actually, congratulations to all of us! And long live Tachyon!
Champagne will be served at 1 PM tomorrow.
Marc
On 2011-03-21, at 5:31 PM, chevalma@iro.umontreal.ca wrote:
Once the 64 bit bugs are fixed, we will work on compiling the standard library. With some luck, we might be able to run the shell pretty soon.
I think I have located the error. It is in the x86_64 encoding of the move with sign-extension instruction (which is used to convert a C string to a JS string). The code generated for the function cStringToBox contains:
00013c 48 0f be 3c 1e movsxbq (%rsi + %rbx),%rdi 000141 66 0f be ff movsxbw %dil,%di
The second x86_64 instruction, which is generated for the LIR instruction "u16 ch_2 = icast_u16 cCh_2;", should be encoded as:
000141 66 40 0f be ff movsxbw %dil,%di
If the "40" is not there I believe it encodes the instruction "movsxbw %bh,%di", and probably bh contains 0, which explains the error during scanning (unknown token: 0).
Erick can you fix it?
Side note, the instruction mnemonic should not contain the "x". Also the operand (%rsi + %rbx) should be printed as (%rsi, %rbx). That way the listing generated is compatible with the GNU assembler (making it easier to cut and paste, and compare code, from other GNU tools).
Marc
Thanks Marc,
I'll fix it.
Erick
Le 11-03-22 11:13 , Marc Feeley a écrit :
On 2011-03-21, at 5:31 PM, chevalma@iro.umontreal.ca wrote:
Once the 64 bit bugs are fixed, we will work on compiling the standard library. With some luck, we might be able to run the shell pretty soon.
I think I have located the error. It is in the x86_64 encoding of the move with sign-extension instruction (which is used to convert a C string to a JS string). The code generated for the function cStringToBox contains:
00013c 48 0f be 3c 1e movsxbq (%rsi + %rbx),%rdi 000141 66 0f be ff movsxbw %dil,%di
The second x86_64 instruction, which is generated for the LIR instruction "u16 ch_2 = icast_u16 cCh_2;", should be encoded as:
000141 66 40 0f be ff movsxbw %dil,%di
If the "40" is not there I believe it encodes the instruction "movsxbw %bh,%di", and probably bh contains 0, which explains the error during scanning (unknown token: 0).
Erick can you fix it?
Side note, the instruction mnemonic should not contain the "x". Also the operand (%rsi + %rbx) should be printed as (%rsi, %rbx). That way the listing generated is compatible with the GNU assembler (making it easier to cut and paste, and compare code, from other GNU tools).
Marc
Tachyon-list mailing list Tachyon-list@iro.umontreal.ca https://webmail.iro.umontreal.ca/mailman/listinfo/tachyon-list
-
Bruno Dufour -
chevalma＠iro.umontreal.ca -
Erick Lavoie -
Marc Feeley