Merge "Add boilerplate to decode and process vector FP instructions." into main am: 42638c1568

Original change: https://android-review.googlesource.com/c/platform/frameworks/libs/binary_translation/+/2973380

Change-Id: Ie82563afd32e6802db4af3ada5e2a3831fbfe7db
Signed-off-by: Automerger Merge Worker <android-build-automerger-merge-worker@system.gserviceaccount.com>

3 files changed, 178 insertions, 16 deletions

diff --git a/decoder/include/berberis/decoder/riscv64/decoder.h b/decoder/include/berberis/decoder/riscv64/decoder.h
index 7343987a..39c1ad55 100644
--- a/decoder/include/berberis/decoder/riscv64/decoder.h
+++ b/decoder/include/berberis/decoder/riscv64/decoder.h
@@ -241,6 +241,89 @@ class Decoder {
     kVlm = 0b01011,
   };
 
+  enum class VOpFVfOpcode : uint8_t {
+    kVfaddvf = 0b000000,
+    kVfsubvf = 0b000010,
+    kVfminvf = 0b000100,
+    kVfmaxvf = 0b000110,
+    kVfsgnjvf = 0b001000,
+    kVfsgnjnvf = 0b001001,
+    kVfsgnjxvf = 0b001010,
+    kVfslide1upvf = 0b001110,
+    kVfslide1downvf = 0b001111,
+    kVfmvsf = 0b010000,
+    kVfmergevf = 0b010111,  // Also kVfmv.vf
+    kVmfeqvf = 0b011000,
+    kVmflevf = 0b011001,
+    kVmfltvf = 0b011011,
+    kVmfnevf = 0b011100,
+    kVmfgtvf = 0b011101,
+    kVmfgevf = 0b011111,
+    kVfdivvf = 0b100000,
+    kVfrdivvf = 0b100001,
+    kVfmulvf = 0b100100,
+    kVfrsubvf = 0b100111,
+    kVfmaddvf = 0b101000,
+    kVfnmaddvf = 0b101001,
+    kVfmsubvf = 0b101010,
+    kVfnmsubvf = 0b101011,
+    kVfmaccvf = 0b101100,
+    kVfnmaccvf = 0b101101,
+    kVfmsacvf = 0b101110,
+    kVfnmsacvf = 0b101111,
+    kVfwaddvf = 0b110000,
+    kVfwsubvf = 0b110010,
+    kVfwaddwf = 0b110100,
+    kVfwsubwf = 0b110110,
+    kVfwmulvf = 0b111000,
+    kVfwmaccvf = 0b111100,
+    kVfwnmaccvf = 0b111100,
+    kVfwmsacvf = 0b111100,
+    kVfwnmsacvf = 0b111111,
+  };
+
+  enum class VOpFVvOpcode : uint8_t {
+    kVfaddvv = 0b000000,
+    kVfredusumvv = 0b000001,
+    kVfsubvv = 0b000010,
+    kVfredosumvv = 0b000011,
+    kVfminvv = 0b000100,
+    kVfredminvv = 0b000101,
+    kVfmaxvv = 0b000110,
+    kVfredmaxvv = 0b000111,
+    kVfsgnjvv = 0b001000,
+    kVfsgnjnvv = 0b001001,
+    kVfsgnjxvv = 0b001010,
+    kVfmvfs = 0b010000,
+    kVfcvtXX = 0b010010,
+    kVXXXv = 010011,  // Vfsqrt.v/Vfrsqrt7.v/Vfrec7.v/Vfclass.v
+    kVmfeqvv = 0b011000,
+    kVmflevv = 0b011001,
+    kVmfltvv = 0b011011,
+    kVmfnevv = 0b011100,
+    kVfdivvv = 0b100000,
+    kVfmulvv = 0b100100,
+    kVfmaddvv = 0b101000,
+    kVfnmaddvv = 0b101001,
+    kVfmsubvv = 0b101010,
+    kVfnmsubvv = 0b101011,
+    kVfmaccvv = 0b101100,
+    kVfnmaccvv = 0b101101,
+    kVfmsacvv = 0b101110,
+    kVfnmsacvv = 0b101111,
+    kVfwaddvv = 0b110000,
+    kVfwredusumvv = 0b110001,
+    kVfwsubvv = 0b110010,
+    kVfwredosumvv = 0b110011,
+    kVfwaddwv = 0b110100,
+    kVfwsubwv = 0b110110,
+    kVfwmulvv = 0b111000,
+    kVfwmaccvv = 0b111100,
+    kVfwnmaccvv = 0b111100,
+    kVfwmsacvv = 0b111100,
+    kVfwnmsacvv = 0b111111,
+  };
+
   enum class VOpIViOpcode : uint8_t {
     kVaddvi = 0b000000,
     kVrsubvi = 0b000011,
@@ -252,7 +335,7 @@ class Decoder {
     kVslidedownvi = 0b001111,
     kVadcvi = 0b010000,
     kVmadcvi = 0b010001,
-    kVmergevi = 0b010111,
+    kVmergevi = 0b010111,  // Also kVmv.vi
     kVmseqvi = 0b011000,
     kVmsnevi = 0b011001,
     kVmsleuvi = 0b011100,
@@ -289,7 +372,7 @@ class Decoder {
     kVmadcvv = 0b010001,
     kVsbcvv = 0b010010,
     kVmsbcvv = 0b010011,
-    kVmergevv = 0b010111,
+    kVmergevv = 0b010111,  // Also kVmv.vv
     kVmseqvv = 0b011000,
     kVmsnevv = 0b011001,
     kVmsltuvv = 0b011010,
@@ -366,7 +449,7 @@ class Decoder {
     kVmadcvx = 0b010001,
     kVsbcvx = 0b010010,
     kVmsbcvx = 0b010011,
-    kVmergevx = 0b010111,
+    kVmergevx = 0b010111,  // Also Vmv.vx
     kVmseqvx = 0b011000,
     kVmsnevx = 0b011001,
     kVmsltuvx = 0b011010,
@@ -699,6 +782,22 @@ class Decoder {
     uint8_t src;
   };
 
+  struct VOpFVfArgs {
+    VOpFVfOpcode opcode;
+    bool vm;
+    uint8_t dst;
+    uint8_t src1;
+    uint8_t src2;
+  };
+
+  struct VOpFVvArgs {
+    VOpFVvOpcode opcode;
+    bool vm;
+    uint8_t dst;
+    uint8_t src1;
+    uint8_t src2;
+  };
+
   struct VOpIViArgs {
     VOpIViOpcode opcode;
     bool vm;
@@ -1940,6 +2039,16 @@ class Decoder {
         };
         return insn_consumer_->OpVector(args);
       }
+      case 0b001: {
+        const VOpFVvArgs args = {
+            .opcode = VOpFVvOpcode{opcode},
+            .vm = vm,
+            .dst = dst,
+            .src1 = src1,
+            .src2 = src2,
+        };
+        return insn_consumer_->OpVector(args);
+      }
       case 0b010: {
         const VOpMVvArgs args = {
             .opcode = VOpMVvOpcode{opcode},
@@ -1970,6 +2079,16 @@ class Decoder {
         };
         return insn_consumer_->OpVector(args);
       }
+      case 0b101: {
+        const VOpFVfArgs args = {
+            .opcode = VOpFVfOpcode{opcode},
+            .vm = vm,
+            .dst = dst,
+            .src1 = src1,
+            .src2 = src2,
+        };
+        return insn_consumer_->OpVector(args);
+      }
       case 0b110: {
         const VOpMVxArgs args = {
             .opcode = VOpMVxOpcode{opcode},
diff --git a/decoder/include/berberis/decoder/riscv64/semantics_player.h b/decoder/include/berberis/decoder/riscv64/semantics_player.h
index 44a6a659..2e948f69 100644
--- a/decoder/include/berberis/decoder/riscv64/semantics_player.h
+++ b/decoder/include/berberis/decoder/riscv64/semantics_player.h
@@ -809,6 +809,15 @@ class SemanticsPlayer {
     listener_->OpVector(args, arg2);
   }
 
+  void OpVector(const typename Decoder::VOpFVfArgs& args) {
+    // Note: we don't have information here to chosee between GetFRegAndUnboxNan<Float32> and
+    // GetFRegAndUnboxNan<Float64> because that depends on vtype.
+    FpRegister arg2 = GetFpReg(args.src2);
+    listener_->OpVector(args, arg2);
+  }
+
+  void OpVector(const typename Decoder::VOpFVvArgs& args) { listener_->OpVector(args); }
+
   void OpVector(const typename Decoder::VOpIViArgs& args) { listener_->OpVector(args); }
 
   void OpVector(const typename Decoder::VOpIVvArgs& args) { listener_->OpVector(args); }
diff --git a/interpreter/riscv64/interpreter.cc b/interpreter/riscv64/interpreter.cc
index 83a22716..0325e0e6 100644
--- a/interpreter/riscv64/interpreter.cc
+++ b/interpreter/riscv64/interpreter.cc
@@ -559,19 +559,37 @@ class Interpreter {
         default:
           return Unimplemented();
       }
-    }
-    VectorRegisterGroupMultiplier vlmul = static_cast<VectorRegisterGroupMultiplier>(vtype & 0b111);
-    switch (static_cast<VectorSelectElementWidth>((vtype >> 3) & 0b111)) {
-      case VectorSelectElementWidth::k8bit:
-        return OpVector<UInt8>(args, vlmul, vtype, extra_args...);
-      case VectorSelectElementWidth::k16bit:
-        return OpVector<UInt16>(args, vlmul, vtype, extra_args...);
-      case VectorSelectElementWidth::k32bit:
-        return OpVector<UInt32>(args, vlmul, vtype, extra_args...);
-      case VectorSelectElementWidth::k64bit:
-        return OpVector<UInt64>(args, vlmul, vtype, extra_args...);
-      default:
-        return Unimplemented();
+    } else {
+      VectorRegisterGroupMultiplier vlmul = static_cast<VectorRegisterGroupMultiplier>(vtype & 0x7);
+      if constexpr (std::is_same_v<VOpArgs, Decoder::VOpFVfArgs> ||
+                    std::is_same_v<VOpArgs, Decoder::VOpFVvArgs>) {
+        switch (static_cast<VectorSelectElementWidth>((vtype >> 3) & 0b111)) {
+          case VectorSelectElementWidth::k32bit:
+            return OpVector<Float32>(
+                args,
+                vlmul,
+                vtype,
+                std::get<0>(intrinsics::UnboxNan<Float32>(bit_cast<Float64>(extra_args)))...);
+          case VectorSelectElementWidth::k64bit:
+            // Note: if arguments are 64bit floats then we don't need to do any unboxing.
+            return OpVector<Float64>(args, vlmul, vtype, bit_cast<Float64>(extra_args)...);
+          default:
+            return Unimplemented();
+        }
+      } else {
+        switch (static_cast<VectorSelectElementWidth>((vtype >> 3) & 0b111)) {
+          case VectorSelectElementWidth::k8bit:
+            return OpVector<UInt8>(args, vlmul, vtype, extra_args...);
+          case VectorSelectElementWidth::k16bit:
+            return OpVector<UInt16>(args, vlmul, vtype, extra_args...);
+          case VectorSelectElementWidth::k32bit:
+            return OpVector<UInt32>(args, vlmul, vtype, extra_args...);
+          case VectorSelectElementWidth::k64bit:
+            return OpVector<UInt64>(args, vlmul, vtype, extra_args...);
+          default:
+            return Unimplemented();
+        }
+      }
     }
   }
 
@@ -1155,6 +1173,22 @@ class Interpreter {
   }
 
   template <typename ElementType, VectorRegisterGroupMultiplier vlmul, TailProcessing vta, auto vma>
+  void OpVector(const Decoder::VOpFVfArgs& args, ElementType /*arg2*/) {
+    switch (args.opcode) {
+      default:
+        Unimplemented();
+    }
+  }
+
+  template <typename ElementType, VectorRegisterGroupMultiplier vlmul, TailProcessing vta, auto vma>
+  void OpVector(const Decoder::VOpFVvArgs& args) {
+    switch (args.opcode) {
+      default:
+        Unimplemented();
+    }
+  }
+
+  template <typename ElementType, VectorRegisterGroupMultiplier vlmul, TailProcessing vta, auto vma>
   void OpVector(const Decoder::VOpIViArgs& args) {
     using SignedType = berberis::SignedType<ElementType>;
     using UnsignedType = berberis::UnsignedType<ElementType>;