From 2515096f5674a91c1fdf3a6e197af38078c203a9 Mon Sep 17 00:00:00 2001
From: ptitSeb <sebastien.chev@gmail.com>
Date: Wed, 13 Nov 2024 17:34:21 +0100
Subject: [PATCH] [ARM64_DYNAREC] Reworked 8/16/32/64 OR opcodes

---
 src/dynarec/arm64/dynarec_arm64_66.c         |   6 +-
 src/dynarec/arm64/dynarec_arm64_6664.c       |   3 +-
 src/dynarec/arm64/dynarec_arm64_66f0.c       |  18 +++-
 src/dynarec/arm64/dynarec_arm64_67.c         |   3 +-
 src/dynarec/arm64/dynarec_arm64_emit_logic.c | 106 ++++++++-----------
 src/dynarec/arm64/dynarec_arm64_helper.h     |   2 +-
 6 files changed, 62 insertions(+), 76 deletions(-)

diff --git a/src/dynarec/arm64/dynarec_arm64_66.c b/src/dynarec/arm64/dynarec_arm64_66.c
index f9acee68..16701d0f 100644
--- a/src/dynarec/arm64/dynarec_arm64_66.c
+++ b/src/dynarec/arm64/dynarec_arm64_66.c
@@ -119,8 +119,7 @@ uintptr_t dynarec64_66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
             SETFLAGS(X_ALL, SF_SET_PENDING);
             i32 = F16;
             UXTHw(x1, xRAX);
-            MOV32w(x2, i32);
-            emit_or16(dyn, ninst, x1, x2, x3, x4);
+            emit_or16c(dyn, ninst, x1, i32, x3, x4);
             BFIz(xRAX, x1, 0, 16);
             break;
 
@@ -493,8 +492,7 @@ uintptr_t dynarec64_66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                     SETFLAGS(X_ALL, SF_SET_PENDING);
                     GETEW(x1, (opcode==0x81)?2:1);
                     if(opcode==0x81) i16 = F16S; else i16 = F8S;
-                    MOVZw(x5, i16);
-                    emit_or16(dyn, ninst, x1, x5, x2, x4);
+                    emit_or16c(dyn, ninst, x1, i16, x2, x4);
                     EWBACK;
                     break;
                 case 2: //ADC
diff --git a/src/dynarec/arm64/dynarec_arm64_6664.c b/src/dynarec/arm64/dynarec_arm64_6664.c
index 36e4ae0d..ba86644b 100644
--- a/src/dynarec/arm64/dynarec_arm64_6664.c
+++ b/src/dynarec/arm64/dynarec_arm64_6664.c
@@ -151,8 +151,7 @@ uintptr_t dynarec64_6664(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                     grab_segdata(dyn, addr, ninst, x1, seg);
                     GETEWO(x1, (opcode==0x81)?2:1);
                     if(opcode==0x81) i16 = F16S; else i16 = F8S;
-                    MOVZw(x5, i16);
-                    emit_or16(dyn, ninst, x1, x5, x2, x4);
+                    emit_or16c(dyn, ninst, x1, i16, x2, x4);
                     EWBACK;
                     break;
                 case 2: //ADC
diff --git a/src/dynarec/arm64/dynarec_arm64_66f0.c b/src/dynarec/arm64/dynarec_arm64_66f0.c
index 6d8f953b..4092bb71 100644
--- a/src/dynarec/arm64/dynarec_arm64_66f0.c
+++ b/src/dynarec/arm64/dynarec_arm64_66f0.c
@@ -327,25 +327,33 @@ uintptr_t dynarec64_66F0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                     if(MODREG) {
                         if(opcode==0x81) i32 = F16S; else i32 = F8S;
                         ed = xRAX+(nextop&7)+(rex.b<<3);
-                        MOV32w(x5, i32);
                         UXTHw(x6, ed);
-                        emit_or16(dyn, ninst, x6, x5, x3, x4);
+                        emit_or16c(dyn, ninst, x6, i32, x3, x4);
                         BFIx(ed, x6, 0, 16);
                     } else {
                         addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, NULL, 0, 0, rex, LOCK_LOCK, 0, (opcode==0x81)?2:1);
                         if(opcode==0x81) i32 = F16S; else i32 = F8S;
-                        MOV32w(x5, i32);
+                        i64 = convert_bitmask_xw(i32);
+                        if(!i64) {MOV32w(x5, i32);}
                         if(arm64_atomics) {
                             UFLAG_IF {
                                 LDSETALH(x5, x1, wback);
-                                emit_or16(dyn, ninst, x1, x5, x3, x4);
+                                if(i64) {
+                                    emit_or16c(dyn, ninst, x1, i32, x3, x4);
+                                } else {
+                                    emit_or16(dyn, ninst, x1, x5, x3, x4);
+                                }
                             } else {
                                 STSETLH(x5, wback);
                             }
                         } else {
                             MARKLOCK;
                             LDAXRH(x1, wback);
-                            emit_or16(dyn, ninst, x1, x5, x3, x4);
+                            if(i64) {
+                                emit_or16c(dyn, ninst, x1, i32, x3, x4);
+                            } else {
+                                emit_or16(dyn, ninst, x1, x5, x3, x4);
+                            }
                             STLXRH(x3, x1, wback);
                             CBNZx_MARKLOCK(x3);
                         }
diff --git a/src/dynarec/arm64/dynarec_arm64_67.c b/src/dynarec/arm64/dynarec_arm64_67.c
index 4adac1d9..d9f6f70b 100644
--- a/src/dynarec/arm64/dynarec_arm64_67.c
+++ b/src/dynarec/arm64/dynarec_arm64_67.c
@@ -818,8 +818,7 @@ uintptr_t dynarec64_67(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                                 SETFLAGS(X_ALL, SF_SET_PENDING);
                                 GETEW32(x1, (opcode==0x81)?2:1);
                                 if(opcode==0x81) i16 = F16S; else i16 = F8S;
-                                MOVZw(x5, i16);
-                                emit_or16(dyn, ninst, x1, x5, x2, x4);
+                                emit_or16c(dyn, ninst, x1, i16, x2, x4);
                                 EWBACK;
                                 break;
                             case 2: //ADC
diff --git a/src/dynarec/arm64/dynarec_arm64_emit_logic.c b/src/dynarec/arm64/dynarec_arm64_emit_logic.c
index f9921d8e..d3d9635d 100644
--- a/src/dynarec/arm64/dynarec_arm64_emit_logic.c
+++ b/src/dynarec/arm64/dynarec_arm64_emit_logic.c
@@ -73,18 +73,18 @@ void emit_or32(dynarec_arm_t* dyn, int ninst, rex_t rex, int s1, int s2, int s3,
 // emit OR32 instruction, from s1 , constant c, store result in s1 using s3 and s4 as scratch
 void emit_or32c(dynarec_arm_t* dyn, int ninst, rex_t rex, int s1, int64_t c, int s3, int s4)
 {
+    int mask = convert_bitmask_xw(c);
+    if(!mask) {
+        MOV64xw(s3, c);
+        emit_or32(dyn, ninst, rex, s1, s3, s3, s4);
+        return;
+    }
     IFX(X_PEND) {
         SET_DF(s4, rex.w?d_or64:d_or32);
     } else IFX(X_ALL) {
         SET_DFNONE(s4);
     }
-    int mask = convert_bitmask_xw(c);
-    if(mask) {
-        ORRxw_mask(s1, s1, (mask>>12)&1, mask&0x3F, (mask>>6)&0x3F);
-    } else {
-        MOV64xw(s3, c);
-        ORRxw_REG(s1, s1, s3);
-    }
+    ORRxw_mask(s1, s1, (mask>>12)&1, mask&0x3F, (mask>>6)&0x3F);
     IFX(X_PEND) {
         STRxw_U12(s1, xEmu, offsetof(x64emu_t, res));
     }
@@ -331,9 +331,9 @@ void emit_or8(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
 {
     MAYUSE(s2);
     IFX(X_PEND) {
-        SET_DF(s3, d_or8);
+        SET_DF(s4, d_or8);
     } else IFX(X_ALL) {
-        SET_DFNONE(s3);
+        SET_DFNONE(s4);
     }
     ORRw_REG(s1, s1, s2);
     IFX(X_PEND) {
@@ -352,18 +352,18 @@ void emit_or8(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
 // emit OR8 instruction, from s1 , constant c, store result in s1 using s3 and s4 as scratch
 void emit_or8c(dynarec_arm_t* dyn, int ninst, int s1, int32_t c, int s3, int s4)
 {
+    int mask = convert_bitmask_w(c);
+    if(!mask) {
+        MOV32w(s3, c);
+        emit_or8c(dyn, ninst, s1, s3, s3, s4);
+        return;
+    }
     IFX(X_PEND) {
         SET_DF(s4, d_or8);
     } else IFX(X_ALL) {
         SET_DFNONE(s4);
     }
-    int mask = convert_bitmask_w(c);
-    if(mask) {
-        ORRw_mask(s1, s1, mask&0x3F, (mask>>6)&0x3F);
-    } else {
-        MOV32w(s3, c&0xff);
-        ORRw_REG(s1, s1, s3);
-    }
+    ORRw_mask(s1, s1, mask&0x3F, (mask>>6)&0x3F);
     IFX(X_PEND) {
         STRB_U12(s1, xEmu, offsetof(x64emu_t, res));
     }
@@ -518,9 +518,9 @@ void emit_or16(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
 {
     MAYUSE(s2);
     IFX(X_PEND) {
-        SET_DF(s3, d_or16);
+        SET_DF(s4, d_or16);
     } else IFX(X_ALL) {
-        SET_DFNONE(s3);
+        SET_DFNONE(s4);
     }
     ORRw_REG(s1, s1, s2);
     IFX(X_PEND) {
@@ -537,50 +537,32 @@ void emit_or16(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
 }
 
 // emit OR16 instruction, from s1 , constant c, store result in s1 using s3 and s4 as scratch
-//void emit_or16c(dynarec_arm_t* dyn, int ninst, int s1, int32_t c, int s3, int s4)
-//{
-//    IFX(X_PEND) {
-//        MOVW(s3, c);
-//        STR_IMM9(s1, xEmu, offsetof(x64emu_t, op1));
-//        STR_IMM9(s3, xEmu, offsetof(x64emu_t, op2));
-//        SET_DF(s4, d_or16);
-//    } else IFX(X_ALL) {
-//        SET_DFNONE(s4);
-//    }
-//    if(c>=0 && c<256) {
-//        IFX(X_ALL) {
-//            ORRS_IMM8(s1, s1, c, 0);
-//        } else {
-//            ORR_IMM8(s1, s1, c, 0);
-//        }
-//    } else {
-//        IFX(X_PEND) {} else {MOVW(s3, c);}
-//        IFX(X_ALL) {
-//            ORRS_REG_LSL_IMM5(s1, s1, s3, 0);
-//        } else {
-//            ORR_REG_LSL_IMM5(s1, s1, s3, 0);
-//        }
-//    }
-//    IFX(X_PEND) {
-//        STR_IMM9(s1, xEmu, offsetof(x64emu_t, res));
-//    }
-//    IFX(X_CF | X_AF | X_ZF) {
-//        BIC_IMM8(xFlags, xFlags, (1<<F_CF)|(1<<F_AF)|(1<<F_ZF), 0);
-//    }
-//    IFX(X_OF) {
-//        BIC_IMM8(xFlags, xFlags, 0b10, 0x0b);
-//    }
-//    IFX(X_ZF) {
-//        ORR_IMM8_COND(cEQ, xFlags, xFlags, 1<<F_ZF, 0);
-//    }
-//    IFX(X_SF) {
-//        MOV_REG_LSR_IMM5(s3, s1, 15);
-//        BFI(xFlags, s3, F_SF, 1);
-//    }
-//    IFX(X_PF) {
-//        emit_pf(dyn, ninst, s1, s4);
-//    }
-//}
+void emit_or16c(dynarec_arm_t* dyn, int ninst, int s1, int32_t c, int s3, int s4)
+{
+    int mask = convert_bitmask_w(c);
+    if(!mask) {
+        MOV32w(s3, c);
+        emit_or16c(dyn, ninst, s1, s3, s3, s4);
+        return;
+    }
+    IFX(X_PEND) {
+        SET_DF(s4, d_or16);
+    } else IFX(X_ALL) {
+        SET_DFNONE(s4);
+    }
+    ORRw_mask(s1, s1, mask&0x3F, (mask>>6)&0x3F);
+    IFX(X_PEND) {
+        STRH_U12(s1, xEmu, offsetof(x64emu_t, res));
+    }
+    IFX(X_CF | X_AF | X_OF) {
+        MOV32w(s3, (1<<F_CF)|(1<<F_AF)|(1<<F_OF));
+        BICw(xFlags, xFlags, s3);
+    }
+    COMP_ZFSF(s1, 16)
+    IFX(X_PF) {
+        emit_pf(dyn, ninst, s1, s4);
+    }
+}
 
 // emit XOR16 instruction, from s1, s2, store result in s1 using s3 and s4 as scratch, s4 can be same as s2 (and so s2 destroyed)
 void emit_xor16(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
diff --git a/src/dynarec/arm64/dynarec_arm64_helper.h b/src/dynarec/arm64/dynarec_arm64_helper.h
index 3965d5fe..d6554061 100644
--- a/src/dynarec/arm64/dynarec_arm64_helper.h
+++ b/src/dynarec/arm64/dynarec_arm64_helper.h
@@ -1460,7 +1460,7 @@ void emit_add16(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4);
 void emit_sub16(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4);
 //void emit_sub16c(dynarec_arm_t* dyn, int ninst, int s1, int32_t c, int s3, int s4);
 void emit_or16(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4);
-//void emit_or16c(dynarec_arm_t* dyn, int ninst, int s1, int32_t c, int s3, int s4);
+void emit_or16c(dynarec_arm_t* dyn, int ninst, int s1, int32_t c, int s3, int s4);
 void emit_xor16(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4);
 //void emit_xor16c(dynarec_arm_t* dyn, int ninst, int s1, int32_t c, int s3, int s4);
 void emit_and16(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4);