avx_sig: Adapt avx_sig test for Aarch64

[dchagin]

For now test succeed on Linux, fails on native and in the Linuxulator

[kib]

There are no XMM regs on aarch64.

[dchagin]

In D40309#917796, @kib wrote:

There are no XMM regs on aarch64.

I know, it was a quick fix to test some changes,
I propose to move inlined asm code to a separate file and rename xmm names to more general names.

Btw, under Linuxulator test works, I forgot to chroot before testing,
so only native fails.

[dchagin]

Rebase to D40312

tools/test/avx_sig/Makefile

 
 .include <src.opts.mk>
 
 PROG=           avx_sig
 SRCS=           avx_sig.c
 MAN=
 
 LIBADD=         pthread
 
 .if ${MACHINE_CPUARCH} == "amd64"
 SRCS+=          c2x2c_amd64.S
 .endif
+.if ${MACHINE_CPUARCH} == "aarch64"
+SRCS+=          c2x2c_aarch64.S
+.endif
 
 .include <bsd.prog.mk>

tools/test/avx_sig/avx_sig.c

 
 #ifndef __unused
 #define __unused        __attribute__((__unused__))
 #endif
 #ifndef nitems
 #define nitems(x)       (sizeof((x)) / sizeof((x)[0]))
 #endif
 
+#if defined(__amd64__)
 #define SIMDRNAM        "xmm"
+#define NREGS           16
+#elif defined(__aarch64__)
+#define SIMDRNAM        "q"
+#define NREGS           32
+#endif
 
 struct simdreg {
         uint8_t simd_bytes[16];
 };
 
 struct simd {
-        struct simdreg simdreg[16];
+        struct simdreg simdreg[NREGS];
 };
 
 void cpu_to_simd(struct simd *simd);
 void simd_to_cpu(struct simd *simd);
 
 static atomic_uint sigs;
 
 static void

tools/test/avx_sig/c2x2c_aarch64.S

+/*
+ * This file is in public domain.
+ * Written by Dmitry Chagin <[e-mail]>
+ *
+ * $FreeBSD$
+ */
+
+        .global cpu_to_simd
+        .type cpu_to_simd, @function
+cpu_to_simd:
+        stp  q0,  q1, [x0, #( 0 * 16)]
+        stp  q2,  q3, [x0, #( 2 * 16)]
+        stp  q4,  q5, [x0, #( 4 * 16)]
+        stp  q6,  q7, [x0, #( 6 * 16)]
+        stp  q8,  q9, [x0, #( 8 * 16)]
+        stp q10, q11, [x0, #(10 * 16)]
+        stp q12, q13, [x0, #(12 * 16)]
+        stp q14, q15, [x0, #(14 * 16)]
+        stp q16, q17, [x0, #(16 * 16)]
+        stp q18, q19, [x0, #(18 * 16)]
+        stp q20, q21, [x0, #(20 * 16)]
+        stp q22, q23, [x0, #(22 * 16)]
+        stp q24, q25, [x0, #(24 * 16)]
+        stp q26, q27, [x0, #(26 * 16)]
+        stp q28, q29, [x0, #(28 * 16)]
+        stp q30, q31, [x0, #(30 * 16)]
+        ret
+
+        .size cpu_to_simd, . - cpu_to_simd
+
+
+        .global simd_to_cpu
+        .type simd_to_cpu, @function
+simd_to_cpu:
+        ldp  q0,  q1, [x0, #( 0 * 16)]
+        ldp  q2,  q3, [x0, #( 2 * 16)]
+        ldp  q4,  q5, [x0, #( 4 * 16)]
+        ldp  q6,  q7, [x0, #( 6 * 16)]
+        ldp  q8,  q9, [x0, #( 8 * 16)]
+        ldp q10, q11, [x0, #(10 * 16)]
+        ldp q12, q13, [x0, #(12 * 16)]
+        ldp q14, q15, [x0, #(14 * 16)]
+        ldp q16, q17, [x0, #(16 * 16)]
+        ldp q18, q19, [x0, #(18 * 16)]
+        ldp q20, q21, [x0, #(20 * 16)]
+        ldp q22, q23, [x0, #(22 * 16)]
+        ldp q24, q25, [x0, #(24 * 16)]
+        ldp q26, q27, [x0, #(26 * 16)]
+        ldp q28, q29, [x0, #(28 * 16)]
+        ldp q30, q31, [x0, #(30 * 16)]
+        ret
+
+        .size simd_to_cpu, . - simd_to_cpu
+
+        .section        .note.GNU-stack,"",@progbits