diff --git a/src/dynarec/arm64/dynarec_arm64_functions.c b/src/dynarec/arm64/dynarec_arm64_functions.c
index e61a44452c..64fbeb4c08 100644
--- a/src/dynarec/arm64/dynarec_arm64_functions.c
+++ b/src/dynarec/arm64/dynarec_arm64_functions.c
@@ -1590,6 +1590,8 @@ void addSSEPreload(dynarec_arm_t* dyn, int ninst, int i2, uint32_t preload)
             has_host_call = 1;
     }
     uint16_t ymm_mask = 0;
+    uint16_t ymm_preload = 0;
+    uint16_t xmm_mask = 0;
     for(int i=0; i<16; ++i) {
         if(preload&(1<<i)) {
             int v = xmm_preload_reg(dyn, i2?(i2-1):0, ninst, i);
@@ -1603,14 +1605,15 @@ void addSSEPreload(dynarec_arm_t* dyn, int ninst, int i2, uint32_t preload)
                     dyn->insts[i3].n.ssecache[i].reg = v;
                     dyn->insts[i3].n.neoncache[v].t = NEON_CACHE_XMMR;
                     dyn->insts[i3].n.neoncache[v].n = i;
+                    if (i3 == i2) xmm_mask |= 1 << i;
                     ++i3;
                 }
             }
         }
         if(preload&(1<<(16+i))) {
             int v = ymm_preload_reg(dyn, i2?(i2-1):0, ninst, i);
-            if(v!=-1) {
-                ymm_mask |= 1<<i;
+            if (v != -1) {
+                ymm_preload |= 1 << i;
                 // propagate until not needed
                 int i3 = i2;
                 while(i3<ninst) {
@@ -1619,16 +1622,17 @@ void addSSEPreload(dynarec_arm_t* dyn, int ninst, int i2, uint32_t preload)
                     else {
                         dyn->insts[i3].n.neoncache[v].t = NEON_CACHE_YMMR;
                         dyn->insts[i3].n.neoncache[v].n = i;
+                        if (i3 == i2) ymm_mask |= 1 << i;
                         ++i3;
                     }
                 }
             }
         }
     }
-    preload = (preload&0xffff) | (ymm_mask<<16);    // update ymm_mask
+    preload = (preload & 0xffff) | (ymm_preload << 16); // update ymm_mask
     dyn->insts[i2].preload_xmmymm = preload;
     dyn->insts[i2].preload_from = ninst;
     dyn->insts[i2].preload_host_call = has_host_call;
-    dyn->insts[i2].n.xmm_unneeded |= preload&0xffff;
+    dyn->insts[i2].n.xmm_unneeded |= xmm_mask;
     dyn->insts[i2].n.ymm_unneeded |= ymm_mask;
 }
diff --git a/src/dynarec/arm64/dynarec_arm64_helper.c b/src/dynarec/arm64/dynarec_arm64_helper.c
index 3ed5deb797..ba27d97067 100644
--- a/src/dynarec/arm64/dynarec_arm64_helper.c
+++ b/src/dynarec/arm64/dynarec_arm64_helper.c
@@ -2815,7 +2815,7 @@ int fpu_get_reg_ymm(dynarec_arm_t* dyn, int ninst, int t, int ymm, int k1, int k
         if(!dyn->n.fpuused[i+j]) {
             // should a test be done to check if ymm is already in the purge list?
             if(!is_ymm_to_keep(dyn, i+j, k1, k2, k3) && (dyn->n.neoncache[i+j].t==NEON_CACHE_YMMR)) {
-                dyn->n.neoncache[i+j].v = 0;
+                fpu_free_reg(dyn, i+j);
                 int ret = internal_mark_ymm(dyn, t, ymm, i+j);
                 if(ret>=0) return ret;
             }
@@ -2824,7 +2824,7 @@ int fpu_get_reg_ymm(dynarec_arm_t* dyn, int ninst, int t, int ymm, int k1, int k
     i = SCRATCH0;
     for(int j=dyn->n.fpu_scratch; j<8; ++j) {
         if(!is_ymm_to_keep(dyn, i+j, k1, k2, k3) && (dyn->n.neoncache[i+j].t==NEON_CACHE_YMMR)) {
-            dyn->n.neoncache[i+j].v = 0;
+            fpu_free_reg(dyn, i+j);
             int ret = internal_mark_ymm(dyn, t, ymm, i+j);
             if(ret>=0) return ret;
         }
@@ -2836,7 +2836,7 @@ int fpu_get_reg_ymm(dynarec_arm_t* dyn, int ninst, int t, int ymm, int k1, int k
                 // Save the reg and recycle it
                 dyn->n.ymm_used|=(1<<dyn->n.neoncache[i+j].n);
                 VSTR128_U12(i+j, xEmu, offsetof(x64emu_t, ymm[dyn->n.neoncache[i+j].n]));
-                dyn->n.neoncache[i+j].v = 0;
+                fpu_free_reg(dyn, i+j);
                 int ret = internal_mark_ymm(dyn, t, ymm, i+j);
                 if(ret>=0) return ret;
             }
@@ -2849,7 +2849,7 @@ int fpu_get_reg_ymm(dynarec_arm_t* dyn, int ninst, int t, int ymm, int k1, int k
             if((dyn->n.neoncache[i+j].t==NEON_CACHE_YMMW) && !is_ymm_to_keep(dyn, i+j, k1, k2, k3)) {
                 dyn->n.ymm_used|=(1<<dyn->n.neoncache[i+j].n);
                 VSTR128_U12(i+j, xEmu, offsetof(x64emu_t, ymm[dyn->n.neoncache[i+j].n]));
-                dyn->n.neoncache[i+j].v = 0;
+                fpu_free_reg(dyn, i+j);
                 int ret = internal_mark_ymm(dyn, t, ymm, i+j);
                 if(ret>=0) return ret;
             }