From 9cf3c25811f3114aad771b356d0e98281adcb4bb Mon Sep 17 00:00:00 2001
From: Lioncash <mathew1800@gmail.com>
Date: Sat, 2 Mar 2019 20:18:59 -0500
Subject: [PATCH] frontend/ir/ir_emitter: Add opcodes for floating point
 reciprocal exponents

---
 src/backend/x64/emit_x64_floating_point.cpp | 17 +++++++++++++++++
 src/frontend/ir/ir_emitter.cpp              |  7 +++++++
 src/frontend/ir/ir_emitter.h                |  1 +
 src/frontend/ir/opcodes.inc                 |  2 ++
 4 files changed, 27 insertions(+)

diff --git a/src/backend/x64/emit_x64_floating_point.cpp b/src/backend/x64/emit_x64_floating_point.cpp
index b6f8a06c..c499e20f 100644
--- a/src/backend/x64/emit_x64_floating_point.cpp
+++ b/src/backend/x64/emit_x64_floating_point.cpp
@@ -719,6 +719,23 @@ void EmitX64::EmitFPRecipEstimate64(EmitContext& ctx, IR::Inst* inst) {
     EmitFPRecipEstimate<u64>(code, ctx, inst);
 }
 
+template <typename FPT>
+static void EmitFPRecipExponent(BlockOfCode& code, EmitContext& ctx, IR::Inst* inst) {
+    auto args = ctx.reg_alloc.GetArgumentInfo(inst);
+    ctx.reg_alloc.HostCall(inst, args[0]);
+    code.mov(code.ABI_PARAM2.cvt32(), ctx.FPCR());
+    code.lea(code.ABI_PARAM3, code.ptr[code.r15 + code.GetJitStateInfo().offsetof_fpsr_exc]);
+    code.CallFunction(&FP::FPRecipExponent<FPT>);
+}
+
+void EmitX64::EmitFPRecipExponent32(EmitContext& ctx, IR::Inst* inst) {
+    EmitFPRecipExponent<u32>(code, ctx, inst);
+}
+
+void EmitX64::EmitFPRecipExponent64(EmitContext& ctx, IR::Inst* inst) {
+    EmitFPRecipExponent<u64>(code, ctx, inst);
+}
+
 template<size_t fsize>
 static void EmitFPRecipStepFused(BlockOfCode& code, EmitContext& ctx, IR::Inst* inst) {
     using FPT = mp::unsigned_integer_of_size<fsize>;
diff --git a/src/frontend/ir/ir_emitter.cpp b/src/frontend/ir/ir_emitter.cpp
index 46b27c0c..6452e7f7 100644
--- a/src/frontend/ir/ir_emitter.cpp
+++ b/src/frontend/ir/ir_emitter.cpp
@@ -1895,6 +1895,13 @@ U32U64 IREmitter::FPRecipEstimate(const U32U64& a) {
     return Inst<U64>(Opcode::FPRecipEstimate64, a);
 }
 
+U32U64 IREmitter::FPRecipExponent(const U32U64& a) {
+    if (a.GetType() == Type::U32) {
+        return Inst<U32>(Opcode::FPRecipExponent32, a);
+    }
+    return Inst<U64>(Opcode::FPRecipExponent64, a);
+}
+
 U32U64 IREmitter::FPRecipStepFused(const U32U64& a, const U32U64& b) {
     if (a.GetType() == Type::U32) {
         return Inst<U32>(Opcode::FPRecipStepFused32, a, b);
diff --git a/src/frontend/ir/ir_emitter.h b/src/frontend/ir/ir_emitter.h
index 5bfb7484..f58c6efa 100644
--- a/src/frontend/ir/ir_emitter.h
+++ b/src/frontend/ir/ir_emitter.h
@@ -305,6 +305,7 @@ public:
     U32U64 FPMulX(const U32U64& a, const U32U64& b);
     U32U64 FPNeg(const U32U64& a);
     U32U64 FPRecipEstimate(const U32U64& a);
+    U32U64 FPRecipExponent(const U32U64& a);
     U32U64 FPRecipStepFused(const U32U64& a, const U32U64& b);
     U32U64 FPRoundInt(const U32U64& a, FP::RoundingMode rounding, bool exact);
     U32U64 FPRSqrtEstimate(const U32U64& a);
diff --git a/src/frontend/ir/opcodes.inc b/src/frontend/ir/opcodes.inc
index 54ea8504..7306e3d0 100644
--- a/src/frontend/ir/opcodes.inc
+++ b/src/frontend/ir/opcodes.inc
@@ -484,6 +484,8 @@ OPCODE(FPNeg32,                                             U32,            U32
 OPCODE(FPNeg64,                                             U64,            U64                                                             )
 OPCODE(FPRecipEstimate32,                                   U32,            U32                                                             )
 OPCODE(FPRecipEstimate64,                                   U64,            U64                                                             )
+OPCODE(FPRecipExponent32,                                   U32,            U32                                                             )
+OPCODE(FPRecipExponent64,                                   U64,            U64                                                             )
 OPCODE(FPRecipStepFused32,                                  U32,            U32,            U32                                             )
 OPCODE(FPRecipStepFused64,                                  U64,            U64,            U64                                             )
 OPCODE(FPRoundInt32,                                        U32,            U32,            U8,             U1                              )