[ARM64_DYNAREC] Few fixes and small cosmetic changes to some partial (V)MOV opcodes

8 files changed, 36 insertions, 58 deletions

diff --git a/src/dynarec/arm64/dynarec_arm64_0f.c b/src/dynarec/arm64/dynarec_arm64_0f.c
index 7071d5d2..4cc1dffd 100644
--- a/src/dynarec/arm64/dynarec_arm64_0f.c
+++ b/src/dynarec/arm64/dynarec_arm64_0f.c
@@ -279,12 +279,12 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
         case 0x12:

             nextop = F8;

             if(MODREG) {

-                INST_NAME("MOVHLPS Gx,Ex");

+                INST_NAME("MOVHLPS Gx, Ex");

                 GETGX(v0, 1);

                 v1 = sse_get_reg(dyn, ninst, x1, (nextop&7)+(rex.b<<3), 0);

                 VMOVeD(v0, 0, v1, 1);

             } else {

-                INST_NAME("MOVLPS Gx,Ex");

+                INST_NAME("MOVLPS Gx, Ex");

                 GETGX(v0, 1);

                 SMREAD();

                 addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, NULL, 0, 0, rex, NULL, 0, 0);

@@ -293,11 +293,11 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
             break;

         case 0x13:

             nextop = F8;

-            INST_NAME("MOVLPS Ex,Gx");

+            INST_NAME("MOVLPS Ex, Gx");

             GETGX(v0, 0);

             if(MODREG) {

-                v1 = sse_get_reg(dyn, ninst, x1, (nextop&7)+(rex.b<<3), 1);

-                VMOVeD(v1, 0, v0, 0);

+                DEFAULT;

+                return addr;

             } else {

                 addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, NULL, 0, 0, rex, NULL, 0, 0);

                 VST1_64(v0, 0, ed);  // better to use VST1 than VSTR_64, to avoid NEON->VFPU transfert I assume

@@ -321,12 +321,12 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
         case 0x16:

             nextop = F8;

             if(MODREG) {

-                INST_NAME("MOVLHPS Gx,Ex");

+                INST_NAME("MOVLHPS Gx, Ex");

                 GETGX(v0, 1);

                 v1 = sse_get_reg(dyn, ninst, x1, (nextop&7)+(rex.b<<3), 0);

                 VMOVeD(v0, 1, v1, 0);

             } else {

-                INST_NAME("MOVHPS Gx,Ex");

+                INST_NAME("MOVHPS Gx, Ex");

                 SMREAD();

                 GETGX(v0, 1);

                 addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, NULL, 0, 0, rex, NULL, 0, 0);

@@ -335,7 +335,7 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
             break;

         case 0x17:

             nextop = F8;

-            INST_NAME("MOVHPS Ex,Gx");

+            INST_NAME("MOVHPS Ex, Gx");

             GETGX(v0, 0);

             if(MODREG) {

                 v1 = sse_get_reg(dyn, ninst, x1, (nextop&7)+(rex.b<<3), 1);

diff --git a/src/dynarec/arm64/dynarec_arm64_64.c b/src/dynarec/arm64/dynarec_arm64_64.c
index 616d9d00..f5c46cd2 100644
--- a/src/dynarec/arm64/dynarec_arm64_64.c
+++ b/src/dynarec/arm64/dynarec_arm64_64.c
@@ -264,7 +264,7 @@ 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_660f.c b/src/dynarec/arm64/dynarec_arm64_660f.c
index 6b35db00..88fbe3ac 100644
--- a/src/dynarec/arm64/dynarec_arm64_660f.c
+++ b/src/dynarec/arm64/dynarec_arm64_660f.c
@@ -87,12 +87,12 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
         case 0x12:

             INST_NAME("MOVLPD Gx, Eq");

             nextop = F8;

-            GETGX(v0, 1);

             if(MODREG) {

                 // access register instead of memory is bad opcode!

                 DEFAULT;

                 return addr;

             }

+            GETGX(v0, 1);

             SMREAD();

             addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, NULL, 0, 0, rex, NULL, 0, 0);

             VLD1_64(v0, 0, ed);

@@ -100,12 +100,12 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
         case 0x13:

             INST_NAME("MOVLPD Eq, Gx");

             nextop = F8;

-            GETGX(v0, 0);

             if(MODREG) {

                 // access register instead of memory is bad opcode!

                 DEFAULT;

                 return addr;

             }

+            GETGX(v0, 0);

             addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, NULL, 0, 0, rex, NULL, 0, 0);

             VST1_64(v0, 0, ed);

             SMWRITE2();

diff --git a/src/dynarec/arm64/dynarec_arm64_6664.c b/src/dynarec/arm64/dynarec_arm64_6664.c
index 0b5ff101..465e7fd6 100644
--- a/src/dynarec/arm64/dynarec_arm64_6664.c
+++ b/src/dynarec/arm64/dynarec_arm64_6664.c
@@ -81,8 +81,7 @@ uintptr_t dynarec64_6664(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                         addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, &unscaled, 0xfff<<4, 15, rex, NULL, 0, 0);
                         SMREAD();
                         ADDz_REG(x4, x4, ed);
-                        ed = x4;
-                        VLD128(v0, ed, fixedaddress);
+                        VLD128(v0, x4, fixedaddress);
                     }
                     break;
 
@@ -97,8 +96,7 @@ uintptr_t dynarec64_6664(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                         grab_segdata(dyn, addr, ninst, x4, seg, (MODREG));
                         addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, &unscaled, 0xfff<<4, 15, rex, NULL, 0, 0);
                         ADDz_REG(x4, x4, ed);
-                        ed = x4;
-                        VST128(v0, ed, fixedaddress);
+                        VST128(v0, x4, fixedaddress);
                         SMWRITE2();
                     }
                     break;
diff --git a/src/dynarec/arm64/dynarec_arm64_avx_0f.c b/src/dynarec/arm64/dynarec_arm64_avx_0f.c
index 4deb01a9..e37a9221 100644
--- a/src/dynarec/arm64/dynarec_arm64_avx_0f.c
+++ b/src/dynarec/arm64/dynarec_arm64_avx_0f.c
@@ -141,17 +141,14 @@ uintptr_t dynarec64_AVX_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int
             break;
         case 0x13:
             nextop = F8;
-            INST_NAME("VMOVLPS Ex, Vx, Gx");
+            INST_NAME("VMOVLPS Ex, Gx");
             GETGX(v0, 0);
-            GETVX(v2, 0);
-            if(v1!=v2)
-                VMOVQ(v0, v2);
             if(MODREG) {
-                v1 = sse_get_reg(dyn, ninst, x1, (nextop&7)+(rex.b<<3), 1);
-                if(v0!=v1) VMOVeD(v1, 0, v0, 0);
+                DEFAULT;
+                return addr;
             } else {
                 addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, NULL, 0, 0, rex, NULL, 0, 0);
-                VST1_64(v0, 0, ed);  // better to use VST1 than VSTR_64, to avoid NEON->VFPU transfert I assume
+                VST1_64(v0, 0, ed);
                 SMWRITE2();
             }
             break;
@@ -199,12 +196,14 @@ uintptr_t dynarec64_AVX_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int
             YMM0(gd);
             break;
         case 0x17:
-            INST_NAME("VMOVHPS Ex,Gx");
+            INST_NAME("VMOVHPS Ex, Gx");
             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);
                 VMOVeD(v1, 0, v0, 1);
+                YMM0(ed);
             } else {
                 addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, NULL, 0, 0, rex, NULL, 0, 0);
                 VST1_64(v0, 1, ed);
diff --git a/src/dynarec/arm64/dynarec_arm64_avx_66_0f.c b/src/dynarec/arm64/dynarec_arm64_avx_66_0f.c
index 31da91af..520adbf2 100644
--- a/src/dynarec/arm64/dynarec_arm64_avx_66_0f.c
+++ b/src/dynarec/arm64/dynarec_arm64_avx_66_0f.c
@@ -121,14 +121,11 @@ uintptr_t dynarec64_AVX_66_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip,
                 return addr;
             }
             GETGX_empty_VX(v0, v2);
-            if(v0==v2) {
-                q0 = fpu_get_scratch(dyn, ninst);
-                VMOVQ(q0, v2);
-            }
             SMREAD();
             addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, NULL, 0, 0, rex, NULL, 0, 0);
             VLD1_64(v0, 0, ed);
-            VMOVeD(v0, 1, (v0==v2)?q0:v2, 1);
+            if(v0!=v2)
+                VMOVeD(v0, 1, v2, 1);
             YMM0(gd);
             break;
         case 0x13:
@@ -171,25 +168,22 @@ uintptr_t dynarec64_AVX_66_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip,
                 return addr;
             }
             GETGX_empty_VX(v0, v2);
-            if(v0==v2) {
-                q0 = fpu_get_scratch(dyn, ninst);
-                VMOVQ(q0, v2);
-            }
             SMREAD();
             addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, NULL, 0, 0, rex, NULL, 0, 0);
             VLD1_64(v0, 1, ed);
-            VMOVeD(v0, 0, (v0==v2)?q0:v2, 0);
+            if(v0!=v2)
+                VMOVeD(v0, 0, v2, 0);
             YMM0(gd);
             break;
         case 0x17:
             INST_NAME("VMOVHPD Eq, Gx");
             nextop = F8;
-            GETGX(v0, 0);
             if(MODREG) {
                 // access register instead of memory is bad opcode!
                 DEFAULT;
                 return addr;
             }
+            GETGX(v0, 0);
             addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, NULL, 0, 0, rex, NULL, 0, 0);
             VST1_64(v0, 1, ed);
             SMWRITE2();
@@ -788,8 +782,7 @@ uintptr_t dynarec64_AVX_66_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip,
             INST_NAME("VMOVDQA Gx, Ex");
             nextop = F8;
             if(MODREG) {
-                v1 = sse_get_reg(dyn, ninst, x1, (nextop&7)+(rex.b<<3), 0);
-                GETGX_empty(v0);
+                GETGX_empty_EX(v0, v1, 0);
                 VMOVQ(v0, v1);
                 if(vex.l) {
                     GETGY_empty_EY(v0, v1);
diff --git a/src/dynarec/arm64/dynarec_arm64_avx_f3_0f.c b/src/dynarec/arm64/dynarec_arm64_avx_f3_0f.c
index 181170c9..36ff5f76 100644
--- a/src/dynarec/arm64/dynarec_arm64_avx_f3_0f.c
+++ b/src/dynarec/arm64/dynarec_arm64_avx_f3_0f.c
@@ -438,13 +438,10 @@ 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) {
-                ed = (nextop&7)+(rex.b<<3);
-                v1 = sse_get_reg(dyn, ninst, x1, ed, 0);
-                GETGX_empty(v0);
+                GETGX_empty_EX(v0, v1, 0);
                 VMOVQ(v0, v1);
                 if(vex.l) {
-                    v1 = ymm_get_reg(dyn, ninst, x1, ed, 0, gd, -1, -1);
-                    GETGY_empty(v0, ed, -1, -1);
+                    GETGY_empty_EY(v0, v1);
                     VMOVQ(v0, v1);
                 }
             } else {
@@ -522,19 +519,15 @@ uintptr_t dynarec64_AVX_F3_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip,
                 IF_UNALIGNED(ip) {
                     MESSAGE(LOG_DEBUG, "\tUnaligned path");
                     addr = geted(dyn, addr, ninst, nextop, &wback, x1, &fixedaddress, NULL, 0, 0, rex, NULL, 0, 0);
-                    if(wback!=x1) {
-                        MOVx_REG(x1, wback);
-                        wback = x1;
-                    }
                     for(int i=0; i<16; ++i) {
-                        VST1_8(v0, i, wback);
-                        ADDx_U12(wback, wback, 1);
+                        VST1_8(v0, i, i?x1:wback);
+                        ADDx_U12(x1, i?x1:wback, 1);
                     }
                     if(vex.l) {
                         GETGY(v0, 0, -1, -1, -1);
                         for(int i=0; i<16; ++i) {
-                            VST1_8(v0, i, wback);
-                            ADDx_U12(wback, wback, 1);
+                            VST1_8(v0, i, x1);
+                            ADDx_U12(x1, x1, 1);
                         }
                     }
                 } else {
diff --git a/src/dynarec/arm64/dynarec_arm64_f30f.c b/src/dynarec/arm64/dynarec_arm64_f30f.c
index a0cf44c9..ec9c0cca 100644
--- a/src/dynarec/arm64/dynarec_arm64_f30f.c
+++ b/src/dynarec/arm64/dynarec_arm64_f30f.c
@@ -415,8 +415,7 @@ uintptr_t dynarec64_F30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
             INST_NAME("MOVDQU 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);

-                GETGX_empty(v0);

+                GETGX_empty_EX(v0, v1, 0);

                 VMOVQ(v0, v1);

             } else {

                 GETGX_empty(v0);

@@ -476,13 +475,9 @@ uintptr_t dynarec64_F30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                 IF_UNALIGNED(ip) {

                     MESSAGE(LOG_DEBUG, "\tUnaligned path");

                     addr = geted(dyn, addr, ninst, nextop, &wback, x1, &fixedaddress, NULL, 0, 0, rex, NULL, 0, 0);

-                    if(wback!=x1) {

-                        MOVx_REG(x1, wback);

-                        wback = x1;

-                    }

                     for(int i=0; i<16; ++i) {

-                        VST1_8(v0, i, wback);

-                        ADDx_U12(wback, wback, 1);

+                        VST1_8(v0, i, i?x1:wback);

+                        ADDx_U12(x1, i?x1:wback, 1);

                     }

                 } else {

                     addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, &unscaled, 0xfff<<4, 15, rex, NULL, 0, 0);