yuzu/xbyak
1
0
Fork 0
mirror of https://github.com/herumi/xbyak synced 2024-11-20 16:06:14 -07:00
Code Issues Projects Releases Packages Wiki Activity Actions

add test of vmovw

Browse source
This commit is contained in:
MITSUNARI Shigeo 2024-10-19 09:43:24 +09:00
parent 6d48afc6c0
commit ab9653f483
1 changed files with 52 additions and 0 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

52
test/misc.cpp
View file

@ -2291,6 +2291,7 @@ CYBOZU_TEST_AUTO(vmovd)
{ {
setDefaultEncodingAVX10(PreAVX10v2Encoding); setDefaultEncodingAVX10(PreAVX10v2Encoding);
vmovd(eax, xm1); // always AVX10.1 vmovd(eax, xm1); // always AVX10.1
vmovd(xm1, eax); // always AVX10.1
vmovd(xm3, xm1); // always AVX10.2 vmovd(xm3, xm1); // always AVX10.2
// AVX-512 (AVX10.1) // AVX-512 (AVX10.1)
vmovd(ptr[rax+128], xm1); vmovd(ptr[rax+128], xm1);
@ -2300,6 +2301,7 @@ CYBOZU_TEST_AUTO(vmovd)
setDefaultEncodingAVX10(AVX10v2Encoding); setDefaultEncodingAVX10(AVX10v2Encoding);
vmovd(eax, xm1); // always AVX10.1 vmovd(eax, xm1); // always AVX10.1
vmovd(xm1, eax); // always AVX10.1
vmovd(xm3, xm1); // always AVX10.2 vmovd(xm3, xm1); // always AVX10.2
// AVX10.2 // AVX10.2
vmovd(ptr[rax+128], xm1); vmovd(ptr[rax+128], xm1);
@ -2310,6 +2312,7 @@ CYBOZU_TEST_AUTO(vmovd)
} c; } c;
const uint8_t tbl[] = { const uint8_t tbl[] = {
0xc5, 0xf9, 0x7e, 0xc8, // avx10.1 0xc5, 0xf9, 0x7e, 0xc8, // avx10.1
0xc5, 0xf9, 0x6e, 0xc8, // avx10.1
0x62, 0xf1, 0x7e, 0x08, 0x7e, 0xd9, // avx10.2 0x62, 0xf1, 0x7e, 0x08, 0x7e, 0xd9, // avx10.2
0xc5, 0xf9, 0x7e, 0x88, 0x80, 0x00, 0x00, 0x00, // avx 0xc5, 0xf9, 0x7e, 0x88, 0x80, 0x00, 0x00, 0x00, // avx
0xc5, 0xf9, 0x6e, 0x88, 0x80, 0x00, 0x00, 0x00, // avx 0xc5, 0xf9, 0x6e, 0x88, 0x80, 0x00, 0x00, 0x00, // avx
@ -2317,6 +2320,7 @@ CYBOZU_TEST_AUTO(vmovd)
0x62, 0x61, 0x7d, 0x08, 0x6e, 0x70, 0x20, // avx10.1 0x62, 0x61, 0x7d, 0x08, 0x6e, 0x70, 0x20, // avx10.1
0xc5, 0xf9, 0x7e, 0xc8, // avx10.1 0xc5, 0xf9, 0x7e, 0xc8, // avx10.1
0xc5, 0xf9, 0x6e, 0xc8, // avx10.1
0x62, 0xf1, 0x7e, 0x08, 0x7e, 0xd9, // avx10.2 0x62, 0xf1, 0x7e, 0x08, 0x7e, 0xd9, // avx10.2
0x62, 0xf1, 0x7d, 0x08, 0xd6, 0x48, 0x20, // avx10.2 0x62, 0xf1, 0x7d, 0x08, 0xd6, 0x48, 0x20, // avx10.2
0x62, 0xf1, 0x7e, 0x08, 0x7e, 0x48, 0x20, // avx10.2 0x62, 0xf1, 0x7e, 0x08, 0x7e, 0x48, 0x20, // avx10.2
@ -2328,4 +2332,52 @@ CYBOZU_TEST_AUTO(vmovd)
CYBOZU_TEST_EQUAL_ARRAY(c.getCode(), tbl, n); CYBOZU_TEST_EQUAL_ARRAY(c.getCode(), tbl, n);
} }
CYBOZU_TEST_AUTO(vmovw)
{
struct Code : Xbyak::CodeGenerator {
Code()
{
setDefaultEncodingAVX10(PreAVX10v2Encoding);
vmovw(eax, xm1); // always avx10.1
vmovw(xm1, eax); // always avx10.1
vmovw(xm3, xm1); // always avx10.2
// AVX10.1
vmovw(ptr[rax+128], xm1);
vmovw(xm1, ptr[rax+128]);
vmovw(ptr[rax+128], xm30);
vmovw(xm30, ptr[rax+128]);
setDefaultEncodingAVX10(AVX10v2Encoding);
vmovw(eax, xm1); // always avx10.1
vmovw(xm1, eax); // always avx10.1
vmovw(xm3, xm1); // always avx10.2
// AVX10.2
vmovw(ptr[rax+128], xm1);
vmovw(xm1, ptr[rax+128]);
vmovw(ptr[rax+128], xm30);
vmovw(xm30, ptr[rax+128]);
}
} c;
const uint8_t tbl[] = {
0x62, 0xf5, 0x7d, 0x08, 0x7e, 0xc8,
0x62, 0xf5, 0x7d, 0x08, 0x6e, 0xc8,
0x62, 0xf5, 0x7e, 0x08, 0x6e, 0xd9,
0x62, 0xf5, 0x7d, 0x08, 0x7e, 0x48, 0x40,
0x62, 0xf5, 0x7d, 0x08, 0x6e, 0x48, 0x40,
0x62, 0x65, 0x7d, 0x08, 0x7e, 0x70, 0x40,
0x62, 0x65, 0x7d, 0x08, 0x6e, 0x70, 0x40,
0x62, 0xf5, 0x7d, 0x08, 0x7e, 0xc8,
0x62, 0xf5, 0x7d, 0x08, 0x6e, 0xc8,
0x62, 0xf5, 0x7e, 0x08, 0x6e, 0xd9,
0x62, 0xf5, 0x7e, 0x08, 0x7e, 0x48, 0x40,
0x62, 0xf5, 0x7e, 0x08, 0x6e, 0x48, 0x40,
0x62, 0x65, 0x7e, 0x08, 0x7e, 0x70, 0x40,
0x62, 0x65, 0x7e, 0x08, 0x6e, 0x70, 0x40,
};
const size_t n = sizeof(tbl) / sizeof(tbl[0]);
CYBOZU_TEST_EQUAL(c.getSize(), n);
CYBOZU_TEST_EQUAL_ARRAY(c.getCode(), tbl, n);
}
#endif #endif