Implementation notes: armeabi, cubox, crypto_core/salsa20

Computer: cubox
Architecture: armeabi
CPU ID: unknown CPU ID
SUPERCOP version: 20161026
Operation: crypto_core
Primitive: salsa20

Time	Implementation	Compiler	Benchmark date	SUPERCOP version
1808	`ref`	`gcc -mcpu=marvell-pj4 -O2`	20161208	20161026
1835	`ref`	`gcc -funroll-loops -mcpu=marvell-pj4 -O2`	20161208	20161026
1847	`ref`	`gcc -funroll-loops -mcpu=marvell-pj4 -O3`	20161208	20161026
1916	`ref`	`gcc -mcpu=marvell-pj4 -O3`	20161208	20161026
2092	`ref`	`gcc -mcpu=marvell-pj4 -Os`	20161208	20161026
2140	`ref`	`gcc -funroll-loops -mcpu=marvell-pj4 -Os`	20161208	20161026

Test failure

Implementation: crypto_core/salsa20/armneon2
Compiler: gcc -funroll-loops -mcpu=marvell-pj4 -O2

error 111

Number of similar (compiler,implementation) pairs: 6, namely:

Compiler	Implementations
gcc -funroll-loops -mcpu=marvell-pj4 -O2	armneon2
gcc -funroll-loops -mcpu=marvell-pj4 -O3	armneon2
gcc -funroll-loops -mcpu=marvell-pj4 -Os	armneon2
gcc -mcpu=marvell-pj4 -O2	armneon2
gcc -mcpu=marvell-pj4 -O3	armneon2
gcc -mcpu=marvell-pj4 -Os	armneon2

Compiler output

Implementation: crypto_core/salsa20/armneon
Compiler: gcc -funroll-loops -mcpu=marvell-pj4 -O2

core.c: In file included from core.c:8:0:
core.c: core.c: In function 'crypto_core_salsa20_armneon':
core.c: /usr/lib/gcc/armv7l-unknown-linux-gnueabihf/6.2.1/include/arm_neon.h:7628:1: error: inlining failed in call to always_inline 'vextq_u32': target specific option mismatch
core.c: vextq_u32 (uint32x4_t __a, uint32x4_t __b, const int __c)
core.c: ^~~~~~~~~
core.c: core.c:50:14: note: called from here
core.c: uint32x4_t start3 = vextq_u32(k2k3n3k7,k2k3n3k7,1);
core.c: ^~~~~~
core.c: In file included from core.c:8:0:
core.c: /usr/lib/gcc/armv7l-unknown-linux-gnueabihf/6.2.1/include/arm_neon.h:7628:1: error: inlining failed in call to always_inline 'vextq_u32': target specific option mismatch
core.c: ...
core.c: core.c:111:14: note: called from here
core.c: x12x13x6x7 = vbslq_u32(abab,x12x1x6x11,x8x13x2x7);
core.c: ~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
core.c: In file included from core.c:8:0:
core.c: /usr/lib/gcc/armv7l-unknown-linux-gnueabihf/6.2.1/include/arm_neon.h:7994:1: error: inlining failed in call to always_inline 'vbslq_u32': target specific option mismatch
core.c: vbslq_u32 (uint32x4_t __a, uint32x4_t __b, uint32x4_t __c)
core.c: ^~~~~~~~~
core.c: core.c:110:14: note: called from here
core.c: x0x1x10x11 = vbslq_u32(abab,x0x5x10x15,x12x1x6x11);
core.c: ~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

Number of similar (compiler,implementation) pairs: 6, namely:

Compiler	Implementations
gcc -funroll-loops -mcpu=marvell-pj4 -O2	armneon
gcc -funroll-loops -mcpu=marvell-pj4 -O3	armneon
gcc -funroll-loops -mcpu=marvell-pj4 -Os	armneon
gcc -mcpu=marvell-pj4 -O2	armneon
gcc -mcpu=marvell-pj4 -O3	armneon
gcc -mcpu=marvell-pj4 -Os	armneon