From 00e4fe3f69e36a98241a14524ee984cb50b4084b Mon Sep 17 00:00:00 2001
From: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Date: Wed, 28 Aug 2024 13:06:48 -0400
Subject: [PATCH] IR: push down AXFLAG lowering

so we can get the new axflag optimizations on billy's x13s.

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
---
 .../Interface/Core/JIT/Arm64/ALUOps.cpp       | 19 +++++++++++++++--
 .../Source/Interface/Core/OpcodeDispatcher.h  | 21 ++++---------------
 FEXCore/Source/Interface/IR/IR.json           |  5 +++--
 3 files changed, 24 insertions(+), 21 deletions(-)

diff --git a/FEXCore/Source/Interface/Core/JIT/Arm64/ALUOps.cpp b/FEXCore/Source/Interface/Core/JIT/Arm64/ALUOps.cpp
index d1283b1460..d47501756f 100644
--- a/FEXCore/Source/Interface/Core/JIT/Arm64/ALUOps.cpp
+++ b/FEXCore/Source/Interface/Core/JIT/Arm64/ALUOps.cpp
@@ -5,6 +5,7 @@ tags: backend|arm64
 $end_info$
 */
 
+#include "CodeEmitter/Emitter.h"
 #include "FEXCore/IR/IR.h"
 #include "Interface/Context/Context.h"
 #include "Interface/Core/JIT/Arm64/JITClass.h"
@@ -296,8 +297,22 @@ DEF_OP(SetSmallNZV) {
 }
 
 DEF_OP(AXFlag) {
-  LOGMAN_THROW_A_FMT(CTX->HostFeatures.SupportsFlagM2, "Unsupported flagm2 op");
-  axflag();
+  if (CTX->HostFeatures.SupportsFlagM2) {
+    axflag();
+  } else {
+    // AXFLAG is defined in the Arm spec as
+    //
+    //   gt: nzCv -> nzCv
+    //   lt: Nzcv -> nzcv  <==>  1 + 0
+    //   eq: nZCv -> nZCv  <==>  1 + (~0)
+    //   un: nzCV -> nZcv  <==>  0 + 0
+    //
+    // For the latter 3 cases, we therefore get the right NZCV by adding V_inv
+    // to (eq ? ~0 : 0). The remaining case is forced with ccmn.
+    auto V_inv = GetReg(IROp->Args[0].ID());
+    csetm(ARMEmitter::Size::i64Bit, TMP1, ARMEmitter::Condition::CC_EQ);
+    ccmn(ARMEmitter::Size::i64Bit, V_inv, TMP1, ARMEmitter::StatusFlags {0x2} /* nzCv */, ARMEmitter::Condition::CC_LE);
+  }
 }
 
 DEF_OP(CondAddNZCV) {
diff --git a/FEXCore/Source/Interface/Core/OpcodeDispatcher.h b/FEXCore/Source/Interface/Core/OpcodeDispatcher.h
index d6dfdc0409..d9f19da1bf 100644
--- a/FEXCore/Source/Interface/Core/OpcodeDispatcher.h
+++ b/FEXCore/Source/Interface/Core/OpcodeDispatcher.h
@@ -2102,22 +2102,9 @@ class OpDispatchBuilder final : public IREmitter {
     // Convert NZCV from the Arm representation to an eXternal representation
     // that's totally not a euphemism for x86, nuh-uh. But maps to exactly we
     // need, what a coincidence!
-    if (CTX->HostFeatures.SupportsFlagM2) {
-      _AXFlag();
-    } else {
-      // AXFLAG is defined in the Arm spec as
-      //
-      //   gt: nzCv -> nzCv
-      //   lt: Nzcv -> nzcv  <==>  1 + 0
-      //   eq: nZCv -> nZCv  <==>  1 + (~0)
-      //   un: nzCV -> nZcv  <==>  0 + 0
-      //
-      // For the latter 3 cases, we therefore get the right NZCV by adding V_inv
-      // to (eq ? ~0 : 0). The remaining case is forced with ccmn.
-      Ref Eq = NZCVSelect(OpSize::i64Bit, {COND_EQ}, _Constant(~0ull), _Constant(0));
-      _CondAddNZCV(OpSize::i64Bit, V_inv, Eq, {COND_FLEU}, 0x2 /* nzCv */);
-    }
-
+    //
+    // Our AXFlag emulation on FlagM2-less systems needs V_inv passed.
+    _AXFlag(CTX->HostFeatures.SupportsFlagM2 ? Invalid() : V_inv);
     PossiblySetNZCVBits = ~0;
     CFInverted = true;
   }
@@ -2131,7 +2118,7 @@ class OpDispatchBuilder final : public IREmitter {
       LOGMAN_THROW_A_FMT(!NZCVDirty, "only expected after fcmp");
 
       // Convert to x86 flags, saves us from or'ing after.
-      _AXFlag();
+      _AXFlag(Invalid());
       PossiblySetNZCVBits = ~0;
       CFInverted = true;
 
diff --git a/FEXCore/Source/Interface/IR/IR.json b/FEXCore/Source/Interface/IR/IR.json
index f6c567f80f..590c64bbf0 100644
--- a/FEXCore/Source/Interface/IR/IR.json
+++ b/FEXCore/Source/Interface/IR/IR.json
@@ -1139,8 +1139,9 @@
         "Desc": ["Invert carry flag in NZCV"],
         "HasSideEffects": true
       },
-      "AXFlag": {
-        "Desc": ["After an FCmp, converts NZCV flags from the Arm format to a mysterious eXternal format"],
+      "AXFlag GPR:$V_inv": {
+        "Desc": ["After an FCmp, converts NZCV flags from the Arm format to a mysterious eXternal format",
+                 "On FlagM2-less platforms, takes the inverted 1/0 overflow flag"],
         "HasSideEffects": true
       },
       "RmifNZCV GPR:$Src, u8:$Rotate, u8:$Mask": {