From 4e2efab94ef6302cce5f4f5b6eaec3312cd4eb00 Mon Sep 17 00:00:00 2001
From: MITSUNARI Shigeo <herumi@nifty.com>
Date: Sat, 19 Oct 2024 08:54:22 +0900
Subject: The encoding of vmovd(w) with REG-to-XMM operands is not affected by
 the encoding flag.

---
 test/misc.cpp | 44 ++++++++++++++++++++++++++++++++++++++++++++
 xbyak/xbyak.h |  9 +++++----
 2 files changed, 49 insertions(+), 4 deletions(-)

diff --git a/test/misc.cpp b/test/misc.cpp
index bc5083b..b4874a7 100644
--- a/test/misc.cpp
+++ b/test/misc.cpp
@@ -2284,4 +2284,48 @@ CYBOZU_TEST_AUTO(avx_vnni_int)
 	CYBOZU_TEST_EQUAL_ARRAY(c.getCode(), tbl, n);
 }
 
+CYBOZU_TEST_AUTO(vmovd)
+{
+	struct Code : Xbyak::CodeGenerator {
+		Code()
+		{
+			setDefaultEncodingAVX10(PreAVX10v2Encoding);
+			vmovd(eax, xm1); // always AVX10.1
+			vmovd(xm3, xm1); // always AVX10.2
+			// AVX-512 (AVX10.1)
+			vmovd(ptr[rax+128], xm1);
+			vmovd(xm1, ptr[rax+128]);
+			vmovd(ptr[rax+128], xm30);
+			vmovd(xm30, ptr[rax+128]);
+
+			setDefaultEncodingAVX10(AVX10v2Encoding);
+			vmovd(eax, xm1); // always AVX10.1
+			vmovd(xm3, xm1); // always AVX10.2
+			// AVX10.2
+			vmovd(ptr[rax+128], xm1);
+			vmovd(xm1, ptr[rax+128]);
+			vmovd(ptr[rax+128], xm30);
+			vmovd(xm30, ptr[rax+128]);
+		}
+	} c;
+	const uint8_t tbl[] = {
+		0xc5, 0xf9, 0x7e, 0xc8, // avx10.1
+		0x62, 0xf1, 0x7e, 0x08, 0x7e, 0xd9, // avx10.2
+		0xc5, 0xf9, 0x7e, 0x88, 0x80, 0x00, 0x00, 0x00, // avx
+		0xc5, 0xf9, 0x6e, 0x88, 0x80, 0x00, 0x00, 0x00, // avx
+		0x62, 0x61, 0x7d, 0x08, 0x7e, 0x70, 0x20, // avx10.1
+		0x62, 0x61, 0x7d, 0x08, 0x6e, 0x70, 0x20, // avx10.1
+
+		0xc5, 0xf9, 0x7e, 0xc8, // avx10.1
+		0x62, 0xf1, 0x7e, 0x08, 0x7e, 0xd9, // avx10.2
+		0x62, 0xf1, 0x7d, 0x08, 0xd6, 0x48, 0x20, // avx10.2
+		0x62, 0xf1, 0x7e, 0x08, 0x7e, 0x48, 0x20, // avx10.2
+		0x62, 0x61, 0x7d, 0x08, 0xd6, 0x70, 0x20, // avx10.2
+		0x62, 0x61, 0x7e, 0x08, 0x7e, 0x70, 0x20, // avx10.2
+	};
+	const size_t n = sizeof(tbl) / sizeof(tbl[0]);
+	CYBOZU_TEST_EQUAL(c.getSize(), n);
+	CYBOZU_TEST_EQUAL_ARRAY(c.getCode(), tbl, n);
+}
+
 #endif
diff --git a/xbyak/xbyak.h b/xbyak/xbyak.h
index b3a4df8..001f5ae 100644
--- a/xbyak/xbyak.h
+++ b/xbyak/xbyak.h
@@ -2809,11 +2809,12 @@ private:
 			std::swap(p1, p2);
 			rev = !rev;
 		}
+		enc = getEncoding(enc, 1);
 		int sel = -1;
-		if (getEncoding(enc, 1) == AVX10v2Encoding) {
-			if ((p1->isXMM() || p1->isMEM()) && p2->isXMM()) sel = 2 + int(rev);
-		} else {
-			if ((p1->isREG(bit) || p1->isMEM()) && p2->isXMM()) sel = int(rev);
+		if (p1->isXMM() || (p1->isMEM() && enc == AVX10v2Encoding)) {
+			sel = 2 + int(rev);
+		} else if (p1->isREG(bit) || p1->isMEM()) {
+			sel = int(rev);
 		}
 		if (sel == -1) XBYAK_THROW(ERR_BAD_COMBINATION)
 		opAVX_X_X_XM(*static_cast<const Xmm*>(p2), xm0, *p1, typeTbl[sel], codeTbl[sel]);
-- 
cgit v1.2.3