diff --git a/target/i386/cpu.h b/target/i386/cpu.h
index 6bb6653e76..5be90e72a9 100644
--- a/target/i386/cpu.h
+++ b/target/i386/cpu.h
@@ -1233,18 +1233,34 @@ typedef struct SegmentCache {
     uint32_t flags;
 } SegmentCache;
 
-#define MMREG_UNION(n, bits)        \
-    union n {                       \
-        uint8_t  _b_##n[(bits)/8];  \
-        uint16_t _w_##n[(bits)/16]; \
-        uint32_t _l_##n[(bits)/32]; \
-        uint64_t _q_##n[(bits)/64]; \
-        float32  _s_##n[(bits)/32]; \
-        float64  _d_##n[(bits)/64]; \
-    }
+typedef union MMXReg {
+    uint8_t  _b_MMXReg[64 / 8];
+    uint16_t _w_MMXReg[64 / 16];
+    uint32_t _l_MMXReg[64 / 32];
+    uint64_t _q_MMXReg[64 / 64];
+    float32  _s_MMXReg[64 / 32];
+    float64  _d_MMXReg[64 / 64];
+} MMXReg;
 
-typedef MMREG_UNION(ZMMReg, 512) ZMMReg;
-typedef MMREG_UNION(MMXReg, 64)  MMXReg;
+typedef union XMMReg {
+    uint64_t _q_XMMReg[128 / 64];
+} XMMReg;
+
+typedef union YMMReg {
+    uint64_t _q_YMMReg[256 / 64];
+    XMMReg   _x_YMMReg[256 / 128];
+} YMMReg;
+
+typedef union ZMMReg {
+    uint8_t  _b_ZMMReg[512 / 8];
+    uint16_t _w_ZMMReg[512 / 16];
+    uint32_t _l_ZMMReg[512 / 32];
+    uint64_t _q_ZMMReg[512 / 64];
+    float32  _s_ZMMReg[512 / 32];
+    float64  _d_ZMMReg[512 / 64];
+    XMMReg   _x_ZMMReg[512 / 128];
+    YMMReg   _y_ZMMReg[512 / 256];
+} ZMMReg;
 
 typedef struct BNDReg {
     uint64_t lb;
@@ -1267,6 +1283,13 @@ typedef struct BNDCSReg {
 #define ZMM_S(n) _s_ZMMReg[15 - (n)]
 #define ZMM_Q(n) _q_ZMMReg[7 - (n)]
 #define ZMM_D(n) _d_ZMMReg[7 - (n)]
+#define ZMM_X(n) _x_ZMMReg[3 - (n)]
+#define ZMM_Y(n) _y_ZMMReg[1 - (n)]
+
+#define XMM_Q(n) _q_XMMReg[1 - (n)]
+
+#define YMM_Q(n) _q_YMMReg[3 - (n)]
+#define YMM_X(n) _x_YMMReg[1 - (n)]
 
 #define MMX_B(n) _b_MMXReg[7 - (n)]
 #define MMX_W(n) _w_MMXReg[3 - (n)]
@@ -1279,6 +1302,13 @@ typedef struct BNDCSReg {
 #define ZMM_S(n) _s_ZMMReg[n]
 #define ZMM_Q(n) _q_ZMMReg[n]
 #define ZMM_D(n) _d_ZMMReg[n]
+#define ZMM_X(n) _x_ZMMReg[n]
+#define ZMM_Y(n) _y_ZMMReg[n]
+
+#define XMM_Q(n) _q_XMMReg[n]
+
+#define YMM_Q(n) _q_YMMReg[n]
+#define YMM_X(n) _x_YMMReg[n]
 
 #define MMX_B(n) _b_MMXReg[n]
 #define MMX_W(n) _w_MMXReg[n]
@@ -1556,8 +1586,8 @@ typedef struct CPUArchState {
     float_status mmx_status; /* for 3DNow! float ops */
     float_status sse_status;
     uint32_t mxcsr;
-    ZMMReg xmm_regs[CPU_NB_REGS == 8 ? 8 : 32];
-    ZMMReg xmm_t0;
+    ZMMReg xmm_regs[CPU_NB_REGS == 8 ? 8 : 32] QEMU_ALIGNED(16);
+    ZMMReg xmm_t0 QEMU_ALIGNED(16);
     MMXReg mmx_t0;
 
     uint64_t opmask_regs[NB_OPMASK_REGS];