From 43344c5400e8fba315bc7d37bcd425c1b1d983d1 Mon Sep 17 00:00:00 2001
From: Lioncash <mathew1800@gmail.com>
Date: Sun, 29 Jul 2018 19:00:18 -0400
Subject: [PATCH] A64: Implement SMAXV, SMINV, UMAXV, and UMINV

---
 src/frontend/A64/decoder/a64.inc              |  8 +-
 .../A64/translate/impl/simd_across_lanes.cpp  | 79 +++++++++++++++++++
 2 files changed, 83 insertions(+), 4 deletions(-)

diff --git a/src/frontend/A64/decoder/a64.inc b/src/frontend/A64/decoder/a64.inc
index ac2b50f6..acc62190 100644
--- a/src/frontend/A64/decoder/a64.inc
+++ b/src/frontend/A64/decoder/a64.inc
@@ -660,8 +660,8 @@ INST(FRSQRTE_4,              "FRSQRTE",                                   "0Q101
 
 // Data Processing - FP and SIMD - SIMD across lanes
 INST(SADDLV,                 "SADDLV",                                    "0Q001110zz110000001110nnnnnddddd")
-//INST(SMAXV,                  "SMAXV",                                     "0Q001110zz110000101010nnnnnddddd")
-//INST(SMINV,                  "SMINV",                                     "0Q001110zz110001101010nnnnnddddd")
+INST(SMAXV,                  "SMAXV",                                     "0Q001110zz110000101010nnnnnddddd")
+INST(SMINV,                  "SMINV",                                     "0Q001110zz110001101010nnnnnddddd")
 INST(ADDV,                   "ADDV",                                      "0Q001110zz110001101110nnnnnddddd")
 //INST(FMAXNMV_1,              "FMAXNMV",                                   "0Q00111000110000110010nnnnnddddd")
 INST(FMAXNMV_2,              "FMAXNMV",                                   "0Q1011100z110000110010nnnnnddddd")
@@ -672,8 +672,8 @@ INST(FMINNMV_2,              "FMINNMV",                                   "0Q101
 //INST(FMINV_1,                "FMINV",                                     "0Q00111010110000111110nnnnnddddd")
 INST(FMINV_2,                "FMINV",                                     "0Q1011101z110000111110nnnnnddddd")
 INST(UADDLV,                 "UADDLV",                                    "0Q101110zz110000001110nnnnnddddd")
-//INST(UMAXV,                  "UMAXV",                                     "0Q101110zz110000101010nnnnnddddd")
-//INST(UMINV,                  "UMINV",                                     "0Q101110zz110001101010nnnnnddddd")
+INST(UMAXV,                  "UMAXV",                                     "0Q101110zz110000101010nnnnnddddd")
+INST(UMINV,                  "UMINV",                                     "0Q101110zz110001101010nnnnnddddd")
 
 // Data Processing - FP and SIMD - SIMD three different
 INST(SADDL,                  "SADDL, SADDL2",                             "0Q001110zz1mmmmm000000nnnnnddddd")
diff --git a/src/frontend/A64/translate/impl/simd_across_lanes.cpp b/src/frontend/A64/translate/impl/simd_across_lanes.cpp
index 771c1f57..0e8b4d71 100644
--- a/src/frontend/A64/translate/impl/simd_across_lanes.cpp
+++ b/src/frontend/A64/translate/impl/simd_across_lanes.cpp
@@ -103,6 +103,69 @@ bool FPMinMax(TranslatorVisitor& v, bool Q, bool sz, Vec Vn, Vec Vd, MinMaxOpera
     v.V_scalar(esize, Vd, result);
     return true;
 }
+
+enum class ScalarMinMaxOperation {
+    Max,
+    Min,
+};
+
+bool ScalarMinMax(TranslatorVisitor& v, bool Q, Imm<2> size, Vec Vn, Vec Vd,
+                  ScalarMinMaxOperation operation, Signedness sign) {
+    if ((size == 0b10 && !Q) || size == 0b11) {
+        return v.ReservedValue();
+    }
+
+    const size_t esize = 8 << size.ZeroExtend();
+    const size_t datasize = Q ? 128 : 64;
+    const size_t elements = datasize / esize;
+
+    const auto get_element = [&](IR::U128 vec, size_t element) {
+        const auto vec_element = v.ir.VectorGetElement(esize, vec, element);
+
+        if (sign == Signedness::Signed) {
+            return v.ir.SignExtendToWord(vec_element);
+        }
+
+        return v.ir.ZeroExtendToWord(vec_element);
+    };
+    
+    const auto op_func = [&](const auto& a, const auto& b) {
+        switch (operation) {
+        case ScalarMinMaxOperation::Max:
+            if (sign == Signedness::Signed) {
+                return v.ir.MaxSigned(a, b);
+            }
+            return v.ir.MaxUnsigned(a, b);
+
+        case ScalarMinMaxOperation::Min:
+            if (sign == Signedness::Signed) {
+                return v.ir.MinSigned(a, b);
+            }
+            return v.ir.MinUnsigned(a, b);
+
+        default:
+            UNREACHABLE();
+            return IR::U32U64{};
+        }
+    };
+
+    const IR::U128 operand = v.V(datasize, Vn);
+
+    IR::U32 value = get_element(operand, 0);
+    for (size_t i = 1; i < elements; i++) {
+        value = op_func(value, get_element(operand, i));
+    }
+
+    if (size == 0b00) {
+        v.V(datasize, Vd, v.ir.ZeroExtendToQuad(v.ir.LeastSignificantByte(value)));
+    } else if (size == 0b01) {
+        v.V(datasize, Vd, v.ir.ZeroExtendToQuad(v.ir.LeastSignificantHalf(value)));
+    } else {
+        v.V(datasize, Vd, v.ir.ZeroExtendToQuad(value));
+    }
+
+    return true;
+}
 } // Anonymous namespace
 
 bool TranslatorVisitor::ADDV(bool Q, Imm<2> size, Vec Vn, Vec Vd) {
@@ -156,7 +219,23 @@ bool TranslatorVisitor::SADDLV(bool Q, Imm<2> size, Vec Vn, Vec Vd) {
     return LongAdd(*this, Q, size, Vn, Vd, Signedness::Signed);
 }
 
+bool TranslatorVisitor::SMAXV(bool Q, Imm<2> size, Vec Vn, Vec Vd) {
+    return ScalarMinMax(*this, Q, size, Vn, Vd, ScalarMinMaxOperation::Max, Signedness::Signed);
+}
+
+bool TranslatorVisitor::SMINV(bool Q, Imm<2> size, Vec Vn, Vec Vd) {
+    return ScalarMinMax(*this, Q, size, Vn, Vd, ScalarMinMaxOperation::Min, Signedness::Signed);
+}
+
 bool TranslatorVisitor::UADDLV(bool Q, Imm<2> size, Vec Vn, Vec Vd) {
     return LongAdd(*this, Q, size, Vn, Vd, Signedness::Unsigned);
 }
+
+bool TranslatorVisitor::UMAXV(bool Q, Imm<2> size, Vec Vn, Vec Vd) {
+    return ScalarMinMax(*this, Q, size, Vn, Vd, ScalarMinMaxOperation::Max, Signedness::Unsigned);
+}
+
+bool TranslatorVisitor::UMINV(bool Q, Imm<2> size, Vec Vn, Vec Vd) {
+    return ScalarMinMax(*this, Q, size, Vn, Vd, ScalarMinMaxOperation::Min, Signedness::Unsigned);
+}
 } // namespace Dynarmic::A64