Why don't I see NEON instruction from the example project?

Robert W

Mastermind 7100 points

Other Parts Discussed in Thread: AM3358

Hi,

I use AM3358 Starter Kit.

When I run StarterWare example project neonVFPBenchmark, the result in the console window is:

VectorScale function takes 8418366 microseconds to run 100000 times

The ticks clocked by the timer is 202040805

VectorProduct function takes 2071820 microseconds to run 100000 times

The ticks clocked by the timer is 49723682

I don't know whether the above number is correct or not.

When I step in the disassembly code trying to see what NEON instructions are used, I do not find NEON instruction in the for loop. Please see the picture:

The compiler and linker options are coming from the StarterWare(I do not modify it at all).

Do you see the NEON instructions from this project?

Here is the compiler options:

-mv7A8 --code_state=32 --abi=eabi -me -O2 --opt_for_speed=2 --include_path="C:/ti/ccsv5/tools/compiler/arm_5.1.1/include" --include_path="../../../../../../../include" --include_path="../../../../../../../include/hw" --include_path="../../../../../../../include/armv7a/am335x" --include_path="../../../../../../../include/armv7a" --gcc --diag_warning=225 --display_error_number --diag_wrap=off --neon

This is the linker options:

-mv7A8 --code_state=32 --abi=eabi -me -O2 --opt_for_speed=2 --gcc --diag_warning=225 --display_error_number --diag_wrap=off --neon -z -m"neonVFPBenchmark.map" -i"C:/ti/ccsv5/tools/compiler/arm_5.1.1/lib" -i"C:/ti/ccsv5/tools/compiler/arm_5.1.1/include" --reread_libs --warn_sections --display_error_number --diag_wrap=off --xml_link_info="neonVFPBenchmark_linkInfo.xml" --ram_model

over 9 years ago

0 Robert W over 9 years ago

Mastermind 7100 points

I use cgt compiler. Inside init.s, there is already the enable instruction as below, but I am not sure about whether it is executed. Is there any register can be check for the enable effectiveness?
Thanks,

; Enable NEON/VFP CoProcessor
;
MRC p15, #0, r1, c1, c0, #2 ; r1 = Access Control Register
ORR r1, r1, #(0xf << 20) ; enable full access for p10,11
MCR p15, #0, r1, c1, c0, #2 ; Access Control Register = r1
MOV r1, #0
MCR p15, #0, r1, c7, c5, #4 ; flush prefetch buffer
MOV r0,#0x40000000
FMXR FPEXC, r0 ; Set the VFP/Neon Enable bit

0 Robert W over 9 years ago in reply to Robert W

Mastermind 7100 points

On CCS 5.5.0, Windows 7, I do not see NEON/VPF registers in the debug perspective. Can you see NEON registers in CCS 5.5? What name category is it?

I see Core Regirster; USER Register etc....

One past post mentioned to watch the bottom status bar. In my debug session, it is always 'SYS' (privileged mode). Is it correct?
Thanks

0 Biser Gatchev-XID over 9 years ago in reply to Robert W

TI__Guru**** 393215 points

Robert,

This forum supports only Linux.

For Starterware support please ask on the dedicated forum: https://e2e.ti.com/support/embedded/starterware/f/790

For CCS support please ask on the dedicated forum: https://e2e.ti.com/support/development_tools/code_composer_studio/f/81

0 Joe G. over 9 years ago in reply to Biser Gatchev-XID

TI__Mastermind 25816 points

Gotta make sure your code is actually using the Neon (or VFP registers). Eyeballing the above assembly, it looks like RTS functions are being called to do the multiply.

Please see Neon link and (VFP) screenshot here: e2e.ti.com/.../1539592

Processors

Processors forum

Why don't I see NEON instruction from the example project?