[ARM64_DYNAREC] Various improvment to various SSE/AVX 128bits/256bits mov opcodes

8 files changed, 144 insertions, 89 deletions

diff --git a/src/dynarec/arm64/arm64_emitter.h b/src/dynarec/arm64/arm64_emitter.h
index 1c8f0296..749e7e03 100644
--- a/src/dynarec/arm64/arm64_emitter.h
+++ b/src/dynarec/arm64/arm64_emitter.h
@@ -909,7 +909,7 @@ int convert_bitmask(uint64_t bitmask);
 #define FCSELS(Sd, Sn, Sm, cond)        EMIT(FCSEL_scalar(0b00, Sm, cond, Sn, Sd))
 #define FCSELD(Dd, Dn, Dm, cond)        EMIT(FCSEL_scalar(0b01, Dm, cond, Dn, Dd))
  
-// VLDR
+// VLDR/VSTR
 #define VMEM_gen(size, opc, imm12, Rn, Rt)  ((size)<<30 | 0b111<<27 | 1<<26 | 0b01<<24 | (opc)<<22 | (imm12)<<10 | (Rn)<<5 | (Rt))
 // imm13 must be 1-aligned
 #define VLDR16_U12(Ht, Rn, imm13)           EMIT(VMEM_gen(0b01, 0b01, ((uint32_t)((imm13)>>1))&0xfff, Rn, Ht))
@@ -928,6 +928,15 @@ int convert_bitmask(uint64_t bitmask);
 // (imm13) must be 1-aligned
 #define VSTR16_U12(Ht, Rn, imm13)           EMIT(VMEM_gen(0b01, 0b00, ((uint32_t)((imm13)>>1))&0xfff, Rn, Ht))
 
+//VLDP/VSTP
+#define VMEMP_vector(opc, L, imm7, Rt2, Rn, Rt) ((opc)<<30 | 0b101<<27 | 1<<26 | 0b010<<23 | (L)<<22 | (imm7)<<15 | (Rt2)<<10 | (Rn)<<5 | (Rt))
+#define VLDP32_I7(Rt1, Rt2, Rn, imm9)       EMIT(VMEMP_vector(0b00, 1, (((int64_t)(imm9))>>2)&0x7f, Rt2, Rn, Rt1))
+#define VLDP64_I7(Rt1, Rt2, Rn, imm10)      EMIT(VMEMP_vector(0b01, 1, (((int64_t)(imm10))>>3)&0x7f, Rt2, Rn, Rt1))
+#define VLDP128_I7(Rt1, Rt2, Rn, imm11)     EMIT(VMEMP_vector(0b10, 1, (((int64_t)(imm11))>>4)&0x7f, Rt2, Rn, Rt1))
+#define VSTP32_I7(Rt1, Rt2, Rn, imm9)       EMIT(VMEMP_vector(0b00, 0, (((int64_t)(imm9))>>2)&0x7f, Rt2, Rn, Rt1))
+#define VSTP64_I7(Rt1, Rt2, Rn, imm10)      EMIT(VMEMP_vector(0b01, 0, (((int64_t)(imm10))>>3)&0x7f, Rt2, Rn, Rt1))
+#define VSTP128_I7(Rt1, Rt2, Rn, imm11)     EMIT(VMEMP_vector(0b10, 0, (((int64_t)(imm11))>>4)&0x7f, Rt2, Rn, Rt1))
+
 #define VMEMUR_vector(size, opc, imm9, Rn, Rt)  ((size)<<30 | 0b111<<27 | 1<<26 | (opc)<<22 | (imm9)<<12 | (Rn)<<5 | (Rt))
 // signed offset, no alignement!
 #define VLDR8_I9(Vt, Rn, imm9)              EMIT(VMEMUR_vector(0b00, 0b01, (imm9)&0b111111111, Rn, Vt))
diff --git a/src/dynarec/arm64/arm64_printer.c b/src/dynarec/arm64/arm64_printer.c
index a0818a78..7d730bc9 100644
--- a/src/dynarec/arm64/arm64_printer.c
+++ b/src/dynarec/arm64/arm64_printer.c
@@ -1644,6 +1644,17 @@ const char* arm64_print(uint32_t opcode, uintptr_t addr)
             snprintf(buff, sizeof(buff), "%sR %s%d, [%s, %+d]", a.L?"LD":"ST", Y[sz], Rt, XtSp[Rn], imm);

         return buff;

     }

+    // LDP/STP vector

+    if(isMask(opcode, "ff1011010Liiiiiii22222nnnnnttttt", &a)) {

+        const char* Y[] = {"S", "D", "Q", "?"};

+        int sz = sf;

+        int offset = signExtend(imm, 7)<<(2+sz);

+        if(!offset)

+            snprintf(buff, sizeof(buff), "%sP %s%d, %s%d, [%s]", a.L?"LD":"ST", Y[sz], Rt, Y[sz], Rt2, XtSp[Rn]);

+        else

+            snprintf(buff, sizeof(buff), "%sP %s%d, %s%d, [%s, %s0x%x]", a.L?"LD":"ST", Y[sz], Rt, Y[sz], Rt2, XtSp[Rn], (offset<0)?"-":"", abs(offset));

+        return buff;

+    }

 

     // (S/U)QXT(U)N

     if(isMask(opcode, "0Q101110ff100001001010nnnnnddddd", &a)) {

diff --git a/src/dynarec/arm64/dynarec_arm64_64.c b/src/dynarec/arm64/dynarec_arm64_64.c
index 1f9edc20..616d9d00 100644
--- a/src/dynarec/arm64/dynarec_arm64_64.c
+++ b/src/dynarec/arm64/dynarec_arm64_64.c
@@ -217,7 +217,7 @@ uintptr_t dynarec64_64(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                                 SMREAD();
                                 addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, &unscaled, 0xfff<<4, 15, rex, NULL, 0, 0);
                                 ADDz_REG(x4, x4, ed);
-                                VLD128(v0, ed, fixedaddress);
+                                VLD128(v0, x4, fixedaddress);
                             }
                             break;
                         default:
diff --git a/src/dynarec/arm64/dynarec_arm64_avx_0f.c b/src/dynarec/arm64/dynarec_arm64_avx_0f.c
index 50231d0e..4deb01a9 100644
--- a/src/dynarec/arm64/dynarec_arm64_avx_0f.c
+++ b/src/dynarec/arm64/dynarec_arm64_avx_0f.c
@@ -79,11 +79,13 @@ uintptr_t dynarec64_AVX_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int
             } else {
                 v0 = sse_get_reg_empty(dyn, ninst, x1, gd);
                 SMREAD();
-                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, NULL, 0xffe<<4, 15, rex, NULL, 0, 0);
-                VLDR128_U12(v0, ed, fixedaddress);   // no alignment issue with ARMv8 NEON :)
                 if(vex.l) {
-                    v0 = ymm_get_reg_empty(dyn, ninst, x1, gd, -1, -1, -1);
-                    VLDR128_U12(v0, ed, fixedaddress+16);
+                    v1 = ymm_get_reg_empty(dyn, ninst, x1, gd, -1, -1, -1);
+                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, NULL, 0x3f<<4, 15, rex, NULL, 1, 0);
+                    VLDP128_I7(v0, v1, ed, fixedaddress);
+                } else {
+                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, &unscaled, 0xfff<<4, 15, rex, NULL, 0, 0);
+                    VLD128(v0, ed, fixedaddress);   // no alignment issue with ARMv8 NEON :)
                 }
             }
             if(!vex.l) YMM0(gd);
@@ -101,13 +103,15 @@ uintptr_t dynarec64_AVX_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int
                     v0 = ymm_get_reg(dyn, ninst, x1, gd, 0, ed, -1, -1);
                     v1 = ymm_get_reg_empty(dyn, ninst, x1, ed, gd, -1, -1);
                     VMOVQ(v1, v0);
-                }
+                } else YMM0(ed);
             } else {
-                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, NULL, 0xffe<<4, 15, rex, NULL, 0, 0);
-                VSTR128_U12(v0, ed, fixedaddress);
                 if(vex.l) {
-                    v0 = ymm_get_reg(dyn, ninst, x1, gd, 0, ed, -1, -1);
-                    VSTR128_U12(v0, ed, fixedaddress+16);
+                    v1 = ymm_get_reg(dyn, ninst, x1, gd, 0, ed, -1, -1);
+                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, NULL, 0x3f<<4, 15, rex, NULL, 1, 0);
+                    VSTP128_I7(v0, v1, ed, fixedaddress);
+                } else {
+                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, &unscaled, 0xfff<<4, 15, rex, NULL, 0, 0);
+                    VST128(v0, ed, fixedaddress);
                 }
                 SMWRITE2();
             }
@@ -224,11 +228,13 @@ uintptr_t dynarec64_AVX_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int
             } else {
                 v0 = sse_get_reg_empty(dyn, ninst, x1, gd);
                 SMREAD();
-                addr = geted(dyn, addr, ninst, nextop, &ed, x3, &fixedaddress, NULL, 0xffe<<4, 15, rex, NULL, 0, 0);
-                VLDR128_U12(v0, ed, fixedaddress);
                 if(vex.l) {
-                    GETGY_empty(v0, -1, -1, -1);
-                    VLDR128_U12(v0, ed, fixedaddress+16);
+                    GETGY_empty(v1, -1, -1, -1);
+                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, NULL, 0x3f<<4, 15, rex, NULL, 1, 0);
+                    VLDP128_I7(v0, v1, ed, fixedaddress);
+                } else {
+                    addr = geted(dyn, addr, ninst, nextop, &ed, x3, &fixedaddress, &unscaled, 0xfff<<4, 15, rex, NULL, 0, 0);
+                    VLD128(v0, ed, fixedaddress);
                 }
             }
             if(!vex.l) YMM0(gd);
@@ -245,13 +251,15 @@ uintptr_t dynarec64_AVX_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int
                 if(vex.l) {
                     GETGYEY_empty(v0, v1);
                     VMOVQ(v1, v0);
-                }
+                } else YMM0(ed);
             } else {
-                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, NULL, 0xffe<<4, 15, rex, NULL, 0, 0);
-                VSTR128_U12(v0, ed, fixedaddress);
                 if(vex.l) {
-                    GETGY(v0, 0, -1, -1, -1);
-                    VSTR128_U12(v0, ed, fixedaddress+16);
+                    GETGY(v1, 0, -1, -1, -1);
+                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, NULL, 0x3f<<4, 15, rex, NULL, 1, 0);
+                    VSTP128_I7(v0, v1, ed, fixedaddress);
+                } else {
+                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, &unscaled, 0xfff<<4, 15, rex, NULL, 0, 0);
+                    VST128(v0, ed, fixedaddress);
                 }
                 SMWRITE2();
             }
@@ -269,13 +277,15 @@ uintptr_t dynarec64_AVX_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int
                 if(vex.l) {
                     GETGYEY_empty(v0, v1);
                     VMOVQ(v1, v0);
-                }
+                } else YMM0(ed);
             } else {
-                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, NULL, 0xffe<<4, 15, rex, NULL, 0, 0);
-                VSTR128_U12(v0, ed, fixedaddress);
                 if(vex.l) {
-                    GETGY(v0, 0, -1, -1, -1);
-                    VSTR128_U12(v0, ed, fixedaddress+16);
+                    GETGY(v1, 0, -1, -1, -1);
+                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, NULL, 0x3f<<4, 15, rex, NULL, 1, 0);
+                    VSTP128_I7(v0, v1, ed, fixedaddress);
+                } else {
+                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, &unscaled, 0xfff<<4, 15, rex, NULL, 0, 0);
+                    VST128(v0, ed, fixedaddress);
                 }
             }
             break;
diff --git a/src/dynarec/arm64/dynarec_arm64_avx_66_0f.c b/src/dynarec/arm64/dynarec_arm64_avx_66_0f.c
index 4dabd1fe..31da91af 100644
--- a/src/dynarec/arm64/dynarec_arm64_avx_66_0f.c
+++ b/src/dynarec/arm64/dynarec_arm64_avx_66_0f.c
@@ -76,11 +76,13 @@ uintptr_t dynarec64_AVX_66_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip,
             } else {
                 SMREAD();
                 v0 = sse_get_reg_empty(dyn, ninst, x1, gd);
-                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, NULL, 0xffe<<4, 15, rex, NULL, 0, 0);
-                VLDR128_U12(v0, ed, fixedaddress);
                 if(vex.l) {
-                    GETGY_empty(v0, -1, -1, -1);
-                    VLDR128_U12(v0, ed, fixedaddress+16);
+                    GETGY_empty(v1, -1, -1, -1);
+                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, NULL, 0x3f<<4, 15, rex, NULL, 1, 0);
+                    VLDP128_I7(v0, v1, ed, fixedaddress);
+                } else {
+                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, &unscaled, 0xfff<<4, 15, rex, NULL, 0, 0);
+                    VLD128(v0, ed, fixedaddress);
                 }
             }
             if(!vex.l) YMM0(gd);
@@ -91,18 +93,21 @@ uintptr_t dynarec64_AVX_66_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip,
             GETG;
             v0 = sse_get_reg(dyn, ninst, x1, gd, 0);
             if(MODREG) {
-                v1 = sse_get_reg_empty(dyn, ninst, x1, (nextop&7)+(rex.b<<3));
+                ed = (nextop&7)+(rex.b<<3);
+                v1 = sse_get_reg_empty(dyn, ninst, x1, ed);
                 VMOVQ(v1, v0);
                 if(vex.l) {
                     GETGYEY_empty(v0, v1);
                     VMOVQ(v1, v0);
-                }
+                } else YMM0(ed);
             } else {
-                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, NULL, 0xffe<<4, 15, rex, NULL, 0, 0);
-                VSTR128_U12(v0, ed, fixedaddress);
                 if(vex.l) {
-                    GETGY(v0, 0, -1, -1, -1);
-                    VSTR128_U12(v0, ed, fixedaddress+16);
+                    GETGY(v1, 0, -1, -1, -1);
+                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, NULL, 0x3f<<4, 15, rex, NULL, 1, 0);
+                    VSTP128_I7(v0, v1, ed, fixedaddress);
+                } else {
+                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, &unscaled, 0xfff<<4, 15, rex, NULL, 0, 0);
+                    VST128(v0, ed, fixedaddress);
                 }
                 SMWRITE2();
             }
@@ -205,11 +210,13 @@ uintptr_t dynarec64_AVX_66_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip,
             } else {
                 SMREAD();
                 v0 = sse_get_reg_empty(dyn, ninst, x1, gd);
-                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, NULL, 0xffe<<4, 15, rex, NULL, 0, 0);
-                VLDR128_U12(v0, ed, fixedaddress);
                 if(vex.l) {
-                    GETGY_empty(v0, -1, -1, -1);
-                    VLDR128_U12(v0, ed, fixedaddress+16);
+                    GETGY_empty(v1, -1, -1, -1);
+                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, NULL, 0x3f<<4, 15, rex, NULL, 1, 0);
+                    VLDP128_I7(v0, v1, ed, fixedaddress);
+                } else {
+                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, &unscaled, 0xfff<<4, 15, rex, NULL, 0, 0);
+                    VLD128(v0, ed, fixedaddress);
                 }
             }
             if(!vex.l) YMM0(gd);
@@ -220,18 +227,21 @@ uintptr_t dynarec64_AVX_66_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip,
             GETG;
             v0 = sse_get_reg(dyn, ninst, x1, gd, 0);
             if(MODREG) {
-                v1 = sse_get_reg_empty(dyn, ninst, x1, (nextop&7)+(rex.b<<3));
+                ed = (nextop&7)+(rex.b<<3);
+                v1 = sse_get_reg_empty(dyn, ninst, x1, ed);
                 VMOVQ(v1, v0);
                 if(vex.l) {
                     GETGYEY_empty(v0, v1);
                     VMOVQ(v1, v0);
-                }
+                } else YMM0(ed);
             } else {
-                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, NULL, 0xffe<<4, 15, rex, NULL, 0, 0);
-                VSTR128_U12(v0, ed, fixedaddress);
                 if(vex.l) {
-                    GETGY(v0, 0, -1, -1, -1);
-                    VSTR128_U12(v0, ed, fixedaddress+16);
+                    GETGY(v1, 0, -1, -1, -1);
+                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, NULL, 0x3f<<4, 15, rex, NULL, 1, 0);
+                    VSTP128_I7(v0, v1, ed, fixedaddress);
+                } else {
+                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, &unscaled, 0xfff<<4, 15, rex, NULL, 0, 0);
+                    VST128(v0, ed, fixedaddress);
                 }
                 SMWRITE2();
             }
@@ -243,18 +253,21 @@ uintptr_t dynarec64_AVX_66_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip,
             GETG;
             v0 = sse_get_reg(dyn, ninst, x1, gd, 0);
             if(MODREG) {
-                v1 = sse_get_reg_empty(dyn, ninst, x1, (nextop&7)+(rex.b<<3));
+                ed = (nextop&7) + (rex.b<<3);
+                v1 = sse_get_reg_empty(dyn, ninst, x1, ed);
                 VMOVQ(v1, v0);
                 if(vex.l) {
                     GETGYEY_empty(v0, v1);
                     VMOVQ(v1, v0);
-                }
+                } else YMM0(ed);
             } else {
-                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, NULL, 0xffe<<4, 15, rex, NULL, 0, 0);
-                VSTR128_U12(v0, ed, fixedaddress);
                 if(vex.l) {
-                    GETGY(v0, 0, -1, -1, -1);
-                    VSTR128_U12(v0, ed, fixedaddress+16);
+                    GETGY(v1, 0, -1, -1, -1);
+                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, NULL, 0x3f<<4, 15, rex, NULL, 1, 0);
+                    VSTP128_I7(v0, v1, ed, fixedaddress);
+                } else {
+                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, &unscaled, 0xfff<<4, 15, rex, NULL, 0, 0);
+                    VST128(v0, ed, fixedaddress);
                 }
             }
             break;
@@ -785,11 +798,13 @@ uintptr_t dynarec64_AVX_66_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip,
             } else {
                 GETGX_empty(v0);
                 SMREAD();
-                addr = geted(dyn, addr, ninst, nextop, &ed, x3, &fixedaddress, NULL, 0xffe<<4, 15, rex, NULL, 0, 0);
-                VLDR128_U12(v0, ed, fixedaddress);
                 if(vex.l) {
-                    GETGY_empty(v0, -1, -1, -1);
-                    VLDR128_U12(v0, ed, fixedaddress+16);
+                    GETGY_empty(v1, -1, -1, -1);
+                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, NULL, 0x3f<<4, 15, rex, NULL, 1, 0);
+                    VLDP128_I7(v0, v1, ed, fixedaddress);
+                } else {
+                    addr = geted(dyn, addr, ninst, nextop, &ed, x3, &fixedaddress, &unscaled, 0xfff<<4, 15, rex, NULL, 0, 0);
+                    VLD128(v0, ed, fixedaddress);
                 }
             }
             if(!vex.l) YMM0(gd);
@@ -1206,18 +1221,21 @@ uintptr_t dynarec64_AVX_66_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip,
             nextop = F8;
             GETGX(v0, 0);
             if(MODREG) {
-                v1 = sse_get_reg(dyn, ninst, x1, (nextop&7)+(rex.b<<3), 1);
+                ed = (nextop&7)+(rex.b<<3);
+                v1 = sse_get_reg(dyn, ninst, x1, ed, 1);
                 VMOVQ(v1, v0);
                 if(vex.l) {
                     GETGYEY(v0, v1);
                     VMOVQ(v1, v0);
-                }
+                } else YMM0(ed);
             } else {
-                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, &unscaled, 0xffe<<4, 15, rex, NULL, 0, 0);
-                VST128(v0, ed, fixedaddress);
                 if(vex.l) {
-                    GETGY(v0, 0, -1, -1, -1);
-                    VST128(v0, ed, fixedaddress+16);
+                    GETGY(v1, 0, -1, -1, -1);
+                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, NULL, 0x3f<<4, 15, rex, NULL, 1, 0);
+                    VSTP128_I7(v0, v1, ed, fixedaddress);
+                } else {
+                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, &unscaled, 0xfff<<4, 15, rex, NULL, 0, 0);
+                    VST128(v0, ed, fixedaddress);
                 }
                 SMWRITE2();
             }
@@ -1694,18 +1712,21 @@ uintptr_t dynarec64_AVX_66_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip,
             GETG;
             v0 = sse_get_reg(dyn, ninst, x1, gd, 0);
             if(MODREG) {
-                v1 = sse_get_reg_empty(dyn, ninst, x1, (nextop&7)+(rex.b<<3));
+                ed = (nextop&7)+(rex.b<<3);
+                v1 = sse_get_reg_empty(dyn, ninst, x1, ed);
                 VMOVQ(v1, v0);
                 if(vex.l) {
                     GETGYEY_empty(v0, v1);
                     VMOVQ(v1, v0);
-                }
+                } else YMM0(ed);
             } else {
-                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, NULL, 0xffe<<4, 15, rex, NULL, 0, 0);
-                VSTR128_U12(v0, ed, fixedaddress);
                 if(vex.l) {
-                    GETGY(v0, 0, -1, -1, -1);
-                    VSTR128_U12(v0, ed, fixedaddress+16);
+                    GETGY(v1, 0, -1, -1, -1);
+                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, NULL, 0x3f<<4, 15, rex, NULL, 1, 0);
+                    VSTP128_I7(v0, v1, ed, fixedaddress);
+                } else {
+                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, &unscaled, 0xfff<<4, 15, rex, NULL, 0, 0);
+                    VST128(v0, ed, fixedaddress);
                 }
             }
             break;
diff --git a/src/dynarec/arm64/dynarec_arm64_avx_66_0f38.c b/src/dynarec/arm64/dynarec_arm64_avx_66_0f38.c
index f6c036aa..fe254a73 100644
--- a/src/dynarec/arm64/dynarec_arm64_avx_66_0f38.c
+++ b/src/dynarec/arm64/dynarec_arm64_avx_66_0f38.c
@@ -386,9 +386,8 @@ uintptr_t dynarec64_AVX_66_0F38(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip
                 v1 = ymm_get_reg(dyn, ninst, x3, s0, 0, gd, vex.v, -1);
                 VMOVQ(d1, v1);
             } else {
-                addr = geted(dyn, addr, ninst, nextop, &ed, x3, &fixedaddress, NULL, 0xffe<<4, 15, rex, NULL, 0, 0);
-                VLDR128_U12(d0, ed, fixedaddress);
-                VLDR128_U12(d1, ed, fixedaddress+16);
+                addr = geted(dyn, addr, ninst, nextop, &ed, x3, &fixedaddress, NULL, 0x3f<<4, 15, rex, NULL, 1, 0);
+                VLDP128_I7(d0, d1, ed, fixedaddress);
             }
             MOV32w(x3, 0x03020100);
             VDUPQS(q1, x3);
diff --git a/src/dynarec/arm64/dynarec_arm64_avx_f2_0f.c b/src/dynarec/arm64/dynarec_arm64_avx_f2_0f.c
index 5e79f6a1..a07dcee8 100644
--- a/src/dynarec/arm64/dynarec_arm64_avx_f2_0f.c
+++ b/src/dynarec/arm64/dynarec_arm64_avx_f2_0f.c
@@ -599,11 +599,10 @@ uintptr_t dynarec64_AVX_F2_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip,
                 }
             } else {
                 v0 = sse_get_reg_empty(dyn, ninst, x1, gd);
+                v1 = ymm_get_reg_empty(dyn, ninst, x1, gd, -1, -1, -1);
                 SMREAD();
-                addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, NULL, 0xffe<<4, 15, rex, NULL, 0, 0);
-                VLDR128_U12(v0, ed, fixedaddress);
-                v0 = ymm_get_reg_empty(dyn, ninst, x1, gd, -1, -1, -1);
-                VLDR128_U12(v0, ed, fixedaddress+16);
+                addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, NULL, 0x3f<<4, 15, rex, NULL, 1, 0);
+                VLDP128_I7(v0, v1, ed, fixedaddress);
             }
             if(!vex.l) YMM0(gd);
             break;
diff --git a/src/dynarec/arm64/dynarec_arm64_avx_f3_0f.c b/src/dynarec/arm64/dynarec_arm64_avx_f3_0f.c
index 4cc04f7e..181170c9 100644
--- a/src/dynarec/arm64/dynarec_arm64_avx_f3_0f.c
+++ b/src/dynarec/arm64/dynarec_arm64_avx_f3_0f.c
@@ -438,22 +438,25 @@ uintptr_t dynarec64_AVX_F3_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip,
             INST_NAME("VMOVDQU Gx, Ex");// no alignment constraint on NEON here, so same as MOVDQA
             nextop = F8;
             if(MODREG) {
-                v1 = sse_get_reg(dyn, ninst, x1, (nextop&7)+(rex.b<<3), 0);
+                ed = (nextop&7)+(rex.b<<3);
+                v1 = sse_get_reg(dyn, ninst, x1, ed, 0);
                 GETGX_empty(v0);
                 VMOVQ(v0, v1);
                 if(vex.l) {
-                    v1 = ymm_get_reg(dyn, ninst, x1, (nextop&7)+(rex.b<<3), 0, gd, -12, -1);
-                    GETGY_empty(v0, (nextop&7)+(rex.b<<3), -1, -1);
+                    v1 = ymm_get_reg(dyn, ninst, x1, ed, 0, gd, -1, -1);
+                    GETGY_empty(v0, ed, -1, -1);
                     VMOVQ(v0, v1);
                 }
             } else {
                 GETGX_empty(v0);
                 SMREAD();
-                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, NULL, 0xffe<<4, 15, rex, NULL, 0, 0);
-                VLDR128_U12(v0, ed, fixedaddress);
                 if(vex.l) {
-                    GETGY_empty(v0, -1, -1, -1);
-                    VLDR128_U12(v0, ed, fixedaddress+16);
+                    GETGY_empty(v1, -1, -1, -1);
+                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, NULL, 0x3f<<4, 15, rex, NULL, 1, 0);
+                    VLDP128_I7(v0, v1, ed, fixedaddress);
+                } else {
+                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, &unscaled, 0xfff<<4, 15, rex, NULL, 0, 0);
+                    VLD128(v0, ed, fixedaddress);
                 }
             }
             if(!vex.l) YMM0(gd);
@@ -507,13 +510,14 @@ uintptr_t dynarec64_AVX_F3_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip,
             nextop = F8;
             GETGX(v0, 0);
             if(MODREG) {
-                v1 = sse_get_reg_empty(dyn, ninst, x1, (nextop&7) + (rex.b<<3));
+                ed = (nextop&7) + (rex.b<<3);
+                v1 = sse_get_reg_empty(dyn, ninst, x1, ed);
                 VMOVQ(v1, v0);
                 if(vex.l) {
-                    GETGY(v0, 0, (nextop&7) + (rex.b<<3), -1, -1);
-                    v1 = ymm_get_reg_empty(dyn, ninst, x1, (nextop&7) + (rex.b<<3), gd, -1, -1);
+                    GETGY(v0, 0, ed, -1, -1);
+                    v1 = ymm_get_reg_empty(dyn, ninst, x1, ed, gd, -1, -1);
                     VMOVQ(v1, v0);
-                } // no ymm raz here it seems
+                } else YMM0(ed);
             } else {
                 IF_UNALIGNED(ip) {
                     MESSAGE(LOG_DEBUG, "\tUnaligned path");
@@ -534,11 +538,13 @@ uintptr_t dynarec64_AVX_F3_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip,
                         }
                     }
                 } else {
-                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, NULL, 0xffe<<4, 15, rex, NULL, 0, 0);
-                    VSTR128_U12(v0, ed, fixedaddress);
                     if(vex.l) {
-                        GETGY(v0, 0, -1, -1, -1);
-                        VSTR128_U12(v0, ed, fixedaddress+16);
+                        GETGY(v1, 0, -1, -1, -1);
+                        addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, NULL, 0x3f<<4, 15, rex, NULL, 1, 0);
+                        VSTP128_I7(v0, v1, ed, fixedaddress);
+                    } else {
+                        addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, &unscaled, 0xfff<<4, 15, rex, NULL, 0, 0);
+                        VST128(v0, ed, fixedaddress);
                     }
                 }
                 SMWRITE2();