From c778c7b868b2f233d31c1cbf3e7b5e42961681fa Mon Sep 17 00:00:00 2001 From: Lioncash Date: Sat, 7 Jul 2018 15:38:18 -0400 Subject: [PATCH] A64: Implement FMAX's vector single and double precision variants --- src/frontend/A64/decoder/a64.inc | 2 +- .../A64/translate/impl/simd_three_same.cpp | 45 ++++++++++++++----- 2 files changed, 34 insertions(+), 13 deletions(-) diff --git a/src/frontend/A64/decoder/a64.inc b/src/frontend/A64/decoder/a64.inc index 3b70753d..90d4aee2 100644 --- a/src/frontend/A64/decoder/a64.inc +++ b/src/frontend/A64/decoder/a64.inc @@ -727,7 +727,7 @@ INST(ADDP_vec, "ADDP (vector)", "0Q001 //INST(FMAXNM_2, "FMAXNM (vector)", "0Q0011100z1mmmmm110001nnnnnddddd") INST(FMLA_vec_2, "FMLA (vector)", "0Q0011100z1mmmmm110011nnnnnddddd") INST(FADD_2, "FADD (vector)", "0Q0011100z1mmmmm110101nnnnnddddd") -//INST(FMAX_2, "FMAX (vector)", "0Q0011100z1mmmmm111101nnnnnddddd") +INST(FMAX_2, "FMAX (vector)", "0Q0011100z1mmmmm111101nnnnnddddd") //INST(FMULX_vec_4, "FMULX", "0Q0011100z1mmmmm110111nnnnnddddd") INST(FCMEQ_reg_4, "FCMEQ (register)", "0Q0011100z1mmmmm111001nnnnnddddd") //INST(FMLAL_vec_1, "FMLAL, FMLAL2 (vector)", "0Q0011100z1mmmmm111011nnnnnddddd") diff --git a/src/frontend/A64/translate/impl/simd_three_same.cpp b/src/frontend/A64/translate/impl/simd_three_same.cpp index 162051f6..cde6547d 100644 --- a/src/frontend/A64/translate/impl/simd_three_same.cpp +++ b/src/frontend/A64/translate/impl/simd_three_same.cpp @@ -145,6 +145,34 @@ bool FPCompareRegister(TranslatorVisitor& v, bool Q, bool sz, Vec Vm, Vec Vn, Ve v.V(datasize, Vd, result); return true; } + +enum class MinMaxOperation { + Min, + Max, +}; + +bool FPMinMaxOperation(TranslatorVisitor& v, bool Q, bool sz, Vec Vm, Vec Vn, Vec Vd, MinMaxOperation operation) { + if (sz && !Q) { + return v.ReservedValue(); + } + + const size_t esize = sz ? 64 : 32; + const size_t datasize = Q ? 128 : 64; + + const IR::U128 operand1 = v.V(datasize, Vn); + const IR::U128 operand2 = v.V(datasize, Vm); + const IR::U128 result = [&] { + if (operation == MinMaxOperation::Min) { + return v.ir.FPVectorMin(esize, operand1, operand2); + } + + return v.ir.FPVectorMax(esize, operand1, operand2); + }(); + + v.V(datasize, Vd, result); + return true; +} + } // Anonymous namespace bool TranslatorVisitor::CMGT_reg_2(bool Q, Imm<2> size, Vec Vm, Vec Vn, Vec Vd) { @@ -743,19 +771,12 @@ bool TranslatorVisitor::EOR_asimd(bool Q, Vec Vm, Vec Vn, Vec Vd) { return true; } +bool TranslatorVisitor::FMAX_2(bool Q, bool sz, Vec Vm, Vec Vn, Vec Vd) { + return FPMinMaxOperation(*this, Q, sz, Vm, Vn, Vd, MinMaxOperation::Max); +} + bool TranslatorVisitor::FMIN_2(bool Q, bool sz, Vec Vm, Vec Vn, Vec Vd) { - if (sz && !Q) { - return ReservedValue(); - } - - const size_t esize = sz ? 64 : 32; - const size_t datasize = Q ? 128 : 64; - - const IR::U128 operand1 = V(datasize, Vn); - const IR::U128 operand2 = V(datasize, Vm); - const IR::U128 result = ir.FPVectorMin(esize, operand1, operand2); - V(datasize, Vd, result); - return true; + return FPMinMaxOperation(*this, Q, sz, Vm, Vn, Vd, MinMaxOperation::Min); } bool TranslatorVisitor::FADDP_vec_2(bool Q, bool sz, Vec Vm, Vec Vn, Vec Vd) {