From 1a4170952d580e2d16b19b3babc3e364519a3ac3 Mon Sep 17 00:00:00 2001 From: Eric Anholt Date: Mon, 4 Feb 2019 11:18:55 -0800 Subject: [PATCH] v3d: Fix pack/unpack of VFPACK operand unpacks. We want to be able to copy propagate our texture unpacks into the vfpack. --- src/broadcom/qpu/qpu_pack.c | 32 ++++++++++++++++++++++++++++- src/broadcom/qpu/tests/qpu_disasm.c | 2 ++ 2 files changed, 33 insertions(+), 1 deletion(-) diff --git a/src/broadcom/qpu/qpu_pack.c b/src/broadcom/qpu/qpu_pack.c index 70f31d73400..b4d1edf3cef 100644 --- a/src/broadcom/qpu/qpu_pack.c +++ b/src/broadcom/qpu/qpu_pack.c @@ -776,7 +776,11 @@ v3d_qpu_add_unpack(const struct v3d_device_info *devinfo, uint64_t packed_inst, case V3D_QPU_A_FMIN: case V3D_QPU_A_FMAX: case V3D_QPU_A_FCMP: - instr->alu.add.output_pack = (op >> 4) & 0x3; + case V3D_QPU_A_VFPACK: + if (instr->alu.add.op != V3D_QPU_A_VFPACK) + instr->alu.add.output_pack = (op >> 4) & 0x3; + else + instr->alu.add.output_pack = V3D_QPU_PACK_NONE; if (!v3d_qpu_float32_unpack_unpack((op >> 2) & 0x3, &instr->alu.add.a_unpack)) { @@ -1042,6 +1046,32 @@ v3d_qpu_add_pack(const struct v3d_device_info *devinfo, opcode |= a_unpack << 2; opcode |= b_unpack << 0; + + break; + } + + case V3D_QPU_A_VFPACK: { + uint32_t a_unpack; + uint32_t b_unpack; + + if (instr->alu.add.a_unpack == V3D_QPU_UNPACK_ABS || + instr->alu.add.b_unpack == V3D_QPU_UNPACK_ABS) { + return false; + } + + if (!v3d_qpu_float32_unpack_pack(instr->alu.add.a_unpack, + &a_unpack)) { + return false; + } + + if (!v3d_qpu_float32_unpack_pack(instr->alu.add.b_unpack, + &b_unpack)) { + return false; + } + + opcode = (opcode & ~(1 << 2)) | (a_unpack << 2); + opcode = (opcode & ~(1 << 0)) | (b_unpack << 0); + break; } diff --git a/src/broadcom/qpu/tests/qpu_disasm.c b/src/broadcom/qpu/tests/qpu_disasm.c index 2e8d980581a..d9561a3c8d7 100644 --- a/src/broadcom/qpu/tests/qpu_disasm.c +++ b/src/broadcom/qpu/tests/qpu_disasm.c @@ -48,6 +48,8 @@ static const struct { { 33, 0x1c0a0dfde2294000ull, "fcmp.ifna rf61.h, r4.abs, r2.l; vfmul rf55, r2.hh, r1" }, { 33, 0x2011c89b402cc000ull, "fsub.norz rf27, r4.abs, r1.abs; vfmul.ifa rf34, r3.swp, r1" }, + { 33, 0xe01b42ab3bb063c0ull, "vfpack.andnc rf43, rf15.l, r0.h; fmul.ifna rf10.h, r4.l, r5.abs" }, + /* small immediates */ { 33, 0x5de24398bbdc6218ull, "vflb.andnn rf24 ; fmul rf14, -8, rf8.h" }, { 33, 0x25ef83d8b166f00full, "vfmin.pushn rf24, 15.ff, r5; smul24.ifnb rf15, r1, r3" }, -- 2.30.2