From 82fb41cba06dceffc4cb9f51277814174bde842d Mon Sep 17 00:00:00 2001 From: "Unknown W. Brackets" Date: Sun, 20 Aug 2023 13:50:02 -0700 Subject: [PATCH] irjit: Implement vtfm 4x4 using dots. --- Core/MIPS/IR/IRCompVFPU.cpp | 32 +++++++++++--------------------- 1 file changed, 11 insertions(+), 21 deletions(-) diff --git a/Core/MIPS/IR/IRCompVFPU.cpp b/Core/MIPS/IR/IRCompVFPU.cpp index 0dee1eb5f2..dbb63fd04b 100644 --- a/Core/MIPS/IR/IRCompVFPU.cpp +++ b/Core/MIPS/IR/IRCompVFPU.cpp @@ -1658,28 +1658,18 @@ namespace MIPSComp { } } return; - } else if (msz == M_4x4 && IsMatrixVec4(msz, sregs)) { - // Consecutive, which is harder. - DISABLE; - int s0 = IRVTEMP_0; - int s1 = IRVTEMP_PFX_S; - // Doesn't make complete sense to me why this works.... (because it doesn't.) - ir.Write(IROp::Vec4Scale, s0, sregs[0], tregs[0]); - for (int i = 1; i < 4; i++) { - if (!homogenous || (i != n - 1)) { - ir.Write(IROp::Vec4Scale, s1, sregs[i], tregs[i]); - ir.Write(IROp::Vec4Add, s0, s0, s1); - } else { - ir.Write(IROp::Vec4Add, s0, s0, sregs[i]); - } - } - if (IsVec4(sz, dregs)) { - ir.Write(IROp::Vec4Mov, dregs[0], s0); - } else { - for (int i = 0; i < 4; i++) { - ir.Write(IROp::FMov, dregs[i], s0 + i); - } + } else if (msz == M_4x4 && IsMatrixVec4(msz, sregs) && IsVec4(sz, tregs)) { + IRReg t = tregs[0]; + if (homogenous) { + // This is probably even what the hardware basically does, wiring t[3] to 1.0f. + ir.Write(IROp::Vec4Init, IRVTEMP_PFX_T, (int)Vec4Init::AllONE); + ir.Write(IROp::Vec4Blend, IRVTEMP_PFX_T, t, IRVTEMP_PFX_T, 0x7); + t = IRVTEMP_PFX_T; } + for (int i = 0; i < 4; i++) + ir.Write(IROp::Vec4Dot, IRVTEMP_PFX_D + i, sregs[i * 4], t); + for (int i = 0; i < 4; i++) + ir.Write(IROp::FMov, dregs[i], IRVTEMP_PFX_D + i); return; }