Hi,
Genx instructions are not directly inserted into batch buffer.
The batch buffer contains some reference pointer which will finally point to a
GPU buffer which contains GPU instructions for a kernel.
The instruction emit code are located in backend/src/backend/genx_encoder.cpp
To execute a kernel, with only opcode and memory access is not enough.
There is an entire runtime environment need to be set up.
To understand those, I suggest you first study hardware spec at
https://01.org/zh/linuxgraphics/documentation/hardware-specification-prms
Thanks
Zou Nanhai
From: Beignet [mailto:[email protected]] On Behalf Of
Dushyant Behl
Sent: Monday, July 06, 2015 8:49 PM
To: [email protected]
Subject: [Beignet] Gen7 Instructions Opcode Examples
Hello Guys,
I'm extremely sorry if I mailed to the wrong mailing list. Please forgive me if
I've done anything wrong, this is my first message to this mailing list.
I'm a masters student of computer science, I'm trying to understand the beignet
project and how the batch buffers are submitted to the i915 gpu driver.
I have a problem getting hold of the arithmetic instruction formats of the
Intel Gen7 Haswell processors.
Can anyone please help me by pointing me to the location where beignet compiler
backend emits compiled instuctions (byte code) in the batchbuffer.
Also I wanted to ask if is possible to test the execution of one single
instruction like ADD/SUB in the batch buffer by writing the opcodes along with
some src and dest memory locations directly in batchbuffer rather than going
through the beignet compiler.
Can anyone please provide me an example of any such instruction?
Thanks in advance,
Dushyant
_______________________________________________
Beignet mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/beignet
