From 40f05769959c6a84b9096719dffdab24053e2e60 Mon Sep 17 00:00:00 2001 From: Lioncash Date: Fri, 7 Sep 2018 17:49:39 -0400 Subject: [PATCH] A64: Implement SQDMULH's by-element vector variant --- src/frontend/A64/decoder/a64.inc | 2 +- src/frontend/A64/translate/impl/impl.h | 2 +- .../impl/simd_vector_x_indexed_element.cpp | 25 +++++++++++++++++++ 3 files changed, 27 insertions(+), 2 deletions(-) diff --git a/src/frontend/A64/decoder/a64.inc b/src/frontend/A64/decoder/a64.inc index 3232b0a0..f4c494a1 100644 --- a/src/frontend/A64/decoder/a64.inc +++ b/src/frontend/A64/decoder/a64.inc @@ -827,7 +827,7 @@ INST(USHLL, "USHLL, USHLL2", "0Q101 INST(MUL_elt, "MUL (by element)", "0Q001111zzLMmmmm1000H0nnnnnddddd") //INST(SMULL_elt, "SMULL, SMULL2 (by element)", "0Q001111zzLMmmmm1010H0nnnnnddddd") //INST(SQDMULL_elt_2, "SQDMULL, SQDMULL2 (by element)", "0Q001111zzLMmmmm1011H0nnnnnddddd") -//INST(SQDMULH_elt_2, "SQDMULH (by element)", "0Q001111zzLMmmmm1100H0nnnnnddddd") +INST(SQDMULH_elt_2, "SQDMULH (by element)", "0Q001111zzLMmmmm1100H0nnnnnddddd") //INST(SQRDMULH_elt_2, "SQRDMULH (by element)", "0Q001111zzLMmmmm1101H0nnnnnddddd") INST(SDOT_elt, "SDOT (by element)", "0Q001111zzLMmmmm1110H0nnnnnddddd") //INST(FMLA_elt_3, "FMLA (by element)", "0Q00111100LMmmmm0001H0nnnnnddddd") diff --git a/src/frontend/A64/translate/impl/impl.h b/src/frontend/A64/translate/impl/impl.h index c44ba3e9..150fb473 100644 --- a/src/frontend/A64/translate/impl/impl.h +++ b/src/frontend/A64/translate/impl/impl.h @@ -897,7 +897,7 @@ struct TranslatorVisitor final { bool MUL_elt(bool Q, Imm<2> size, Imm<1> L, Imm<1> M, Imm<4> Vmlo, Imm<1> H, Vec Vn, Vec Vd); bool SMULL_elt(bool Q, Imm<2> size, bool L, bool M, Vec Vm, bool H, Vec Vn, Vec Vd); bool SQDMULL_elt_2(bool Q, Imm<2> size, bool L, bool M, Vec Vm, bool H, Reg Rn, Vec Vd); - bool SQDMULH_elt_2(bool Q, Imm<2> size, bool L, bool M, Vec Vm, bool H, Vec Vn, Vec Vd); + bool SQDMULH_elt_2(bool Q, Imm<2> size, Imm<1> L, Imm<1> M, Imm<4> Vmlo, Imm<1> H, Vec Vn, Vec Vd); bool SQRDMULH_elt_2(bool Q, Imm<2> size, bool L, bool M, Vec Vm, bool H, Vec Vn, Vec Vd); bool SDOT_elt(bool Q, Imm<2> size, Imm<1> L, Imm<1> M, Imm<4> Vmlo, Imm<1> H, Vec Vn, Vec Vd); bool FMLA_elt_3(bool Q, bool L, bool M, Vec Vm, bool H, Vec Vn, Vec Vd); diff --git a/src/frontend/A64/translate/impl/simd_vector_x_indexed_element.cpp b/src/frontend/A64/translate/impl/simd_vector_x_indexed_element.cpp index aef2a807..015ee0ad 100644 --- a/src/frontend/A64/translate/impl/simd_vector_x_indexed_element.cpp +++ b/src/frontend/A64/translate/impl/simd_vector_x_indexed_element.cpp @@ -147,6 +147,31 @@ bool TranslatorVisitor::FMUL_elt_4(bool Q, bool sz, Imm<1> L, Imm<1> M, Imm<4> V return FPMultiplyByElement(*this, Q, sz, L, M, Vmlo, H, Vn, Vd, ExtraBehavior::None); } +bool TranslatorVisitor::SQDMULH_elt_2(bool Q, Imm<2> size, Imm<1> L, Imm<1> M, Imm<4> Vmlo, Imm<1> H, Vec Vn, Vec Vd) { + if (size == 0b00 || size == 0b11) { + return UnallocatedEncoding(); + } + + const size_t idxsize = H == 1 ? 128 : 64; + const size_t esize = 8 << size.ZeroExtend(); + const size_t datasize = Q ? 128 : 64; + const auto [index, Vmhi] = [=] { + if (size == 0b01) { + return std::make_pair(concatenate(H, L, M).ZeroExtend(), Imm<1>{0}); + } + + return std::make_pair(concatenate(H, L).ZeroExtend(), M); + }(); + + const IR::U128 operand1 = V(datasize, Vn); + const IR::U128 operand2 = V(idxsize, concatenate(Vmhi, Vmlo).ZeroExtend()); + const IR::U128 index_vector = ir.VectorBroadcast(esize, ir.VectorGetElement(esize, operand2, index)); + const IR::U128 result = ir.VectorSignedSaturatedDoublingMultiplyReturnHigh(esize, operand1, index_vector); + + V(datasize, Vd, result); + return true; +} + bool TranslatorVisitor::SDOT_elt(bool Q, Imm<2> size, Imm<1> L, Imm<1> M, Imm<4> Vmlo, Imm<1> H, Vec Vn, Vec Vd) { return DotProduct(*this, Q, size, L, M, Vmlo, H, Vn, Vd, &IREmitter::SignExtendToWord); }