From 347065525f2d3db68e90c850e5c619ff4bd81e23 Mon Sep 17 00:00:00 2001 From: =?utf8?q?Alejandro=20Pi=C3=B1eiro?= Date: Fri, 6 Aug 2021 01:22:31 +0200 Subject: [PATCH] broadcom/qpu: define v3d_qpu_input, use on v3d_qpu_alu_instr MIME-Version: 1.0 Content-Type: text/plain; charset=utf8 Content-Transfer-Encoding: 8bit At this point it just tidy up a little the alu_instr structure. But also serves to prepare the structure for new changes, as 7.x uses raddr instead of mux, and it is just easier to add the raddr to the new input structure. Signed-off-by: Alejandro Piñeiro Signed-off-by: Iago Toral Quiroga Reviewed-by: Iago Toral Quiroga Part-of: --- src/broadcom/compiler/qpu_schedule.c | 65 ++++++++++---------- src/broadcom/compiler/vir.c | 16 ++--- src/broadcom/compiler/vir_dump.c | 8 +-- src/broadcom/compiler/vir_opt_copy_propagate.c | 12 ++-- src/broadcom/compiler/vir_opt_redundant_flags.c | 8 +-- src/broadcom/compiler/vir_to_qpu.c | 30 ++++----- src/broadcom/qpu/qpu_disasm.c | 16 ++--- src/broadcom/qpu/qpu_instr.c | 8 +-- src/broadcom/qpu/qpu_instr.h | 13 ++-- src/broadcom/qpu/qpu_pack.c | 82 ++++++++++++------------- src/broadcom/qpu/tests/qpu_disasm.c | 8 +-- 11 files changed, 134 insertions(+), 132 deletions(-) diff --git a/src/broadcom/compiler/qpu_schedule.c b/src/broadcom/compiler/qpu_schedule.c index a10fa03..455fa38 100644 --- a/src/broadcom/compiler/qpu_schedule.c +++ b/src/broadcom/compiler/qpu_schedule.c @@ -306,14 +306,14 @@ calculate_deps(struct schedule_state *state, struct schedule_node *n) /* XXX: LOAD_IMM */ if (v3d_qpu_add_op_num_src(inst->alu.add.op) > 0) - process_mux_deps(state, n, inst->alu.add.a); + process_mux_deps(state, n, inst->alu.add.a.mux); if (v3d_qpu_add_op_num_src(inst->alu.add.op) > 1) - process_mux_deps(state, n, inst->alu.add.b); + process_mux_deps(state, n, inst->alu.add.b.mux); if (v3d_qpu_mul_op_num_src(inst->alu.mul.op) > 0) - process_mux_deps(state, n, inst->alu.mul.a); + process_mux_deps(state, n, inst->alu.mul.a.mux); if (v3d_qpu_mul_op_num_src(inst->alu.mul.op) > 1) - process_mux_deps(state, n, inst->alu.mul.b); + process_mux_deps(state, n, inst->alu.mul.b.mux); switch (inst->alu.add.op) { case V3D_QPU_A_VPMSETUP: @@ -537,22 +537,22 @@ reads_too_soon_after_write(struct choose_scoreboard *scoreboard, if (inst->alu.add.op != V3D_QPU_A_NOP) { if (v3d_qpu_add_op_num_src(inst->alu.add.op) > 0 && - mux_reads_too_soon(scoreboard, inst, inst->alu.add.a)) { + mux_reads_too_soon(scoreboard, inst, inst->alu.add.a.mux)) { return true; } if (v3d_qpu_add_op_num_src(inst->alu.add.op) > 1 && - mux_reads_too_soon(scoreboard, inst, inst->alu.add.b)) { + mux_reads_too_soon(scoreboard, inst, inst->alu.add.b.mux)) { return true; } } if (inst->alu.mul.op != V3D_QPU_M_NOP) { if (v3d_qpu_mul_op_num_src(inst->alu.mul.op) > 0 && - mux_reads_too_soon(scoreboard, inst, inst->alu.mul.a)) { + mux_reads_too_soon(scoreboard, inst, inst->alu.mul.a.mux)) { return true; } if (v3d_qpu_mul_op_num_src(inst->alu.mul.op) > 1 && - mux_reads_too_soon(scoreboard, inst, inst->alu.mul.b)) { + mux_reads_too_soon(scoreboard, inst, inst->alu.mul.b.mux)) { return true; } } @@ -839,20 +839,20 @@ qpu_merge_raddrs(struct v3d_qpu_instr *result, if (!result->sig.small_imm_b) { if (v3d_qpu_uses_mux(add_instr, V3D_QPU_MUX_B) && raddr_a == add_instr->raddr_b) { - if (add_instr->alu.add.a == V3D_QPU_MUX_B) - result->alu.add.a = V3D_QPU_MUX_A; - if (add_instr->alu.add.b == V3D_QPU_MUX_B && + if (add_instr->alu.add.a.mux == V3D_QPU_MUX_B) + result->alu.add.a.mux = V3D_QPU_MUX_A; + if (add_instr->alu.add.b.mux == V3D_QPU_MUX_B && v3d_qpu_add_op_num_src(add_instr->alu.add.op) > 1) { - result->alu.add.b = V3D_QPU_MUX_A; + result->alu.add.b.mux = V3D_QPU_MUX_A; } } if (v3d_qpu_uses_mux(mul_instr, V3D_QPU_MUX_B) && raddr_a == mul_instr->raddr_b) { - if (mul_instr->alu.mul.a == V3D_QPU_MUX_B) - result->alu.mul.a = V3D_QPU_MUX_A; - if (mul_instr->alu.mul.b == V3D_QPU_MUX_B && + if (mul_instr->alu.mul.a.mux == V3D_QPU_MUX_B) + result->alu.mul.a.mux = V3D_QPU_MUX_A; + if (mul_instr->alu.mul.b.mux == V3D_QPU_MUX_B && v3d_qpu_mul_op_num_src(mul_instr->alu.mul.op) > 1) { - result->alu.mul.b = V3D_QPU_MUX_A; + result->alu.mul.b.mux = V3D_QPU_MUX_A; } } } @@ -863,20 +863,20 @@ qpu_merge_raddrs(struct v3d_qpu_instr *result, result->raddr_b = raddr_b; if (v3d_qpu_uses_mux(add_instr, V3D_QPU_MUX_A) && raddr_b == add_instr->raddr_a) { - if (add_instr->alu.add.a == V3D_QPU_MUX_A) - result->alu.add.a = V3D_QPU_MUX_B; - if (add_instr->alu.add.b == V3D_QPU_MUX_A && + if (add_instr->alu.add.a.mux == V3D_QPU_MUX_A) + result->alu.add.a.mux = V3D_QPU_MUX_B; + if (add_instr->alu.add.b.mux == V3D_QPU_MUX_A && v3d_qpu_add_op_num_src(add_instr->alu.add.op) > 1) { - result->alu.add.b = V3D_QPU_MUX_B; + result->alu.add.b.mux = V3D_QPU_MUX_B; } } if (v3d_qpu_uses_mux(mul_instr, V3D_QPU_MUX_A) && raddr_b == mul_instr->raddr_a) { - if (mul_instr->alu.mul.a == V3D_QPU_MUX_A) - result->alu.mul.a = V3D_QPU_MUX_B; - if (mul_instr->alu.mul.b == V3D_QPU_MUX_A && + if (mul_instr->alu.mul.a.mux == V3D_QPU_MUX_A) + result->alu.mul.a.mux = V3D_QPU_MUX_B; + if (mul_instr->alu.mul.b.mux == V3D_QPU_MUX_A && v3d_qpu_mul_op_num_src(mul_instr->alu.mul.op) > 1) { - result->alu.mul.b = V3D_QPU_MUX_B; + result->alu.mul.b.mux = V3D_QPU_MUX_B; } } @@ -927,11 +927,12 @@ qpu_convert_add_to_mul(struct v3d_qpu_instr *inst) inst->flags.auf = V3D_QPU_UF_NONE; inst->alu.mul.output_pack = inst->alu.add.output_pack; - inst->alu.mul.a_unpack = inst->alu.add.a_unpack; - inst->alu.mul.b_unpack = inst->alu.add.b_unpack; + + inst->alu.mul.a.unpack = inst->alu.add.a.unpack; + inst->alu.mul.b.unpack = inst->alu.add.b.unpack; inst->alu.add.output_pack = V3D_QPU_PACK_NONE; - inst->alu.add.a_unpack = V3D_QPU_UNPACK_NONE; - inst->alu.add.b_unpack = V3D_QPU_UNPACK_NONE; + inst->alu.add.a.unpack = V3D_QPU_UNPACK_NONE; + inst->alu.add.b.unpack = V3D_QPU_UNPACK_NONE; } static bool @@ -2064,12 +2065,12 @@ alu_reads_register(struct v3d_qpu_instr *inst, if (add) { num_src = v3d_qpu_add_op_num_src(inst->alu.add.op); - mux_a = inst->alu.add.a; - mux_b = inst->alu.add.b; + mux_a = inst->alu.add.a.mux; + mux_b = inst->alu.add.b.mux; } else { num_src = v3d_qpu_mul_op_num_src(inst->alu.mul.op); - mux_a = inst->alu.mul.a; - mux_b = inst->alu.mul.b; + mux_a = inst->alu.mul.a.mux; + mux_b = inst->alu.mul.b.mux; } for (int i = 0; i < num_src; i++) { diff --git a/src/broadcom/compiler/vir.c b/src/broadcom/compiler/vir.c index a9c928f..ef91bd7 100644 --- a/src/broadcom/compiler/vir.c +++ b/src/broadcom/compiler/vir.c @@ -113,10 +113,10 @@ vir_is_raw_mov(struct qinst *inst) return false; } - if (inst->qpu.alu.add.a_unpack != V3D_QPU_UNPACK_NONE || - inst->qpu.alu.add.b_unpack != V3D_QPU_UNPACK_NONE || - inst->qpu.alu.mul.a_unpack != V3D_QPU_UNPACK_NONE || - inst->qpu.alu.mul.b_unpack != V3D_QPU_UNPACK_NONE) { + if (inst->qpu.alu.add.a.unpack != V3D_QPU_UNPACK_NONE || + inst->qpu.alu.add.b.unpack != V3D_QPU_UNPACK_NONE || + inst->qpu.alu.mul.a.unpack != V3D_QPU_UNPACK_NONE || + inst->qpu.alu.mul.b.unpack != V3D_QPU_UNPACK_NONE) { return false; } @@ -209,15 +209,15 @@ vir_set_unpack(struct qinst *inst, int src, if (vir_is_add(inst)) { if (src == 0) - inst->qpu.alu.add.a_unpack = unpack; + inst->qpu.alu.add.a.unpack = unpack; else - inst->qpu.alu.add.b_unpack = unpack; + inst->qpu.alu.add.b.unpack = unpack; } else { assert(vir_is_mul(inst)); if (src == 0) - inst->qpu.alu.mul.a_unpack = unpack; + inst->qpu.alu.mul.a.unpack = unpack; else - inst->qpu.alu.mul.b_unpack = unpack; + inst->qpu.alu.mul.b.unpack = unpack; } } diff --git a/src/broadcom/compiler/vir_dump.c b/src/broadcom/compiler/vir_dump.c index 5c47bbd..ab5d404 100644 --- a/src/broadcom/compiler/vir_dump.c +++ b/src/broadcom/compiler/vir_dump.c @@ -270,8 +270,8 @@ vir_dump_alu(struct v3d_compile *c, struct qinst *inst) vir_print_reg(c, inst, inst->dst); fprintf(stderr, "%s", v3d_qpu_pack_name(instr->alu.add.output_pack)); - unpack[0] = instr->alu.add.a_unpack; - unpack[1] = instr->alu.add.b_unpack; + unpack[0] = instr->alu.add.a.unpack; + unpack[1] = instr->alu.add.b.unpack; } else { fprintf(stderr, "%s", v3d_qpu_mul_op_name(instr->alu.mul.op)); fprintf(stderr, "%s", v3d_qpu_cond_name(instr->flags.mc)); @@ -282,8 +282,8 @@ vir_dump_alu(struct v3d_compile *c, struct qinst *inst) vir_print_reg(c, inst, inst->dst); fprintf(stderr, "%s", v3d_qpu_pack_name(instr->alu.mul.output_pack)); - unpack[0] = instr->alu.mul.a_unpack; - unpack[1] = instr->alu.mul.b_unpack; + unpack[0] = instr->alu.mul.a.unpack; + unpack[1] = instr->alu.mul.b.unpack; } for (int i = 0; i < nsrc; i++) { diff --git a/src/broadcom/compiler/vir_opt_copy_propagate.c b/src/broadcom/compiler/vir_opt_copy_propagate.c index da121c2..c4aa725 100644 --- a/src/broadcom/compiler/vir_opt_copy_propagate.c +++ b/src/broadcom/compiler/vir_opt_copy_propagate.c @@ -104,14 +104,14 @@ vir_has_unpack(struct qinst *inst, int chan) if (vir_is_add(inst)) { if (chan == 0) - return inst->qpu.alu.add.a_unpack != V3D_QPU_UNPACK_NONE; + return inst->qpu.alu.add.a.unpack != V3D_QPU_UNPACK_NONE; else - return inst->qpu.alu.add.b_unpack != V3D_QPU_UNPACK_NONE; + return inst->qpu.alu.add.b.unpack != V3D_QPU_UNPACK_NONE; } else { if (chan == 0) - return inst->qpu.alu.mul.a_unpack != V3D_QPU_UNPACK_NONE; + return inst->qpu.alu.mul.a.unpack != V3D_QPU_UNPACK_NONE; else - return inst->qpu.alu.mul.b_unpack != V3D_QPU_UNPACK_NONE; + return inst->qpu.alu.mul.b.unpack != V3D_QPU_UNPACK_NONE; } } @@ -161,7 +161,7 @@ try_copy_prop(struct v3d_compile *c, struct qinst *inst, struct qinst **movs) continue; /* these ops can't represent abs. */ - if (mov->qpu.alu.mul.a_unpack == V3D_QPU_UNPACK_ABS) { + if (mov->qpu.alu.mul.a.unpack == V3D_QPU_UNPACK_ABS) { switch (inst->qpu.alu.add.op) { case V3D_QPU_A_VFPACK: case V3D_QPU_A_FROUND: @@ -189,7 +189,7 @@ try_copy_prop(struct v3d_compile *c, struct qinst *inst, struct qinst **movs) inst->src[i] = mov->src[0]; if (vir_has_unpack(mov, 0)) { - enum v3d_qpu_input_unpack unpack = mov->qpu.alu.mul.a_unpack; + enum v3d_qpu_input_unpack unpack = mov->qpu.alu.mul.a.unpack; vir_set_unpack(inst, i, unpack); } diff --git a/src/broadcom/compiler/vir_opt_redundant_flags.c b/src/broadcom/compiler/vir_opt_redundant_flags.c index c7896d5..6b61ed6 100644 --- a/src/broadcom/compiler/vir_opt_redundant_flags.c +++ b/src/broadcom/compiler/vir_opt_redundant_flags.c @@ -81,11 +81,11 @@ vir_instr_flags_op_equal(struct qinst *a, struct qinst *b) a->qpu.flags.mpf != b->qpu.flags.mpf || a->qpu.alu.add.op != b->qpu.alu.add.op || a->qpu.alu.mul.op != b->qpu.alu.mul.op || - a->qpu.alu.add.a_unpack != b->qpu.alu.add.a_unpack || - a->qpu.alu.add.b_unpack != b->qpu.alu.add.b_unpack || + a->qpu.alu.add.a.unpack != b->qpu.alu.add.a.unpack || + a->qpu.alu.add.b.unpack != b->qpu.alu.add.b.unpack || a->qpu.alu.add.output_pack != b->qpu.alu.add.output_pack || - a->qpu.alu.mul.a_unpack != b->qpu.alu.mul.a_unpack || - a->qpu.alu.mul.b_unpack != b->qpu.alu.mul.b_unpack || + a->qpu.alu.mul.a.unpack != b->qpu.alu.mul.a.unpack || + a->qpu.alu.mul.b.unpack != b->qpu.alu.mul.b.unpack || a->qpu.alu.mul.output_pack != b->qpu.alu.mul.output_pack) { return false; } diff --git a/src/broadcom/compiler/vir_to_qpu.c b/src/broadcom/compiler/vir_to_qpu.c index 15c2e36..c8b6e0a 100644 --- a/src/broadcom/compiler/vir_to_qpu.c +++ b/src/broadcom/compiler/vir_to_qpu.c @@ -106,20 +106,20 @@ set_src(struct v3d_qpu_instr *instr, enum v3d_qpu_mux *mux, struct qpu_reg src) return; } - if (instr->alu.add.a != V3D_QPU_MUX_A && - instr->alu.add.b != V3D_QPU_MUX_A && - instr->alu.mul.a != V3D_QPU_MUX_A && - instr->alu.mul.b != V3D_QPU_MUX_A) { + if (instr->alu.add.a.mux != V3D_QPU_MUX_A && + instr->alu.add.b.mux != V3D_QPU_MUX_A && + instr->alu.mul.a.mux != V3D_QPU_MUX_A && + instr->alu.mul.b.mux != V3D_QPU_MUX_A) { instr->raddr_a = src.index; *mux = V3D_QPU_MUX_A; } else { if (instr->raddr_a == src.index) { *mux = V3D_QPU_MUX_A; } else { - assert(!(instr->alu.add.a == V3D_QPU_MUX_B && - instr->alu.add.b == V3D_QPU_MUX_B && - instr->alu.mul.a == V3D_QPU_MUX_B && - instr->alu.mul.b == V3D_QPU_MUX_B) || + assert(!(instr->alu.add.a.mux == V3D_QPU_MUX_B && + instr->alu.add.b.mux == V3D_QPU_MUX_B && + instr->alu.mul.a.mux == V3D_QPU_MUX_B && + instr->alu.mul.b.mux == V3D_QPU_MUX_B) || src.index == instr->raddr_b); instr->raddr_b = src.index; @@ -147,14 +147,14 @@ is_no_op_mov(struct qinst *qinst) if (waddr < V3D_QPU_WADDR_R0 || waddr > V3D_QPU_WADDR_R4) return false; - if (qinst->qpu.alu.mul.a != + if (qinst->qpu.alu.mul.a.mux != V3D_QPU_MUX_R0 + (waddr - V3D_QPU_WADDR_R0)) { return false; } } else { int raddr; - switch (qinst->qpu.alu.mul.a) { + switch (qinst->qpu.alu.mul.a.mux) { case V3D_QPU_MUX_A: raddr = qinst->qpu.raddr_a; break; @@ -171,7 +171,7 @@ is_no_op_mov(struct qinst *qinst) /* No packing or flags updates, or we need to execute the * instruction. */ - if (qinst->qpu.alu.mul.a_unpack != V3D_QPU_UNPACK_NONE || + if (qinst->qpu.alu.mul.a.unpack != V3D_QPU_UNPACK_NONE || qinst->qpu.alu.mul.output_pack != V3D_QPU_PACK_NONE || qinst->qpu.flags.mc != V3D_QPU_COND_NONE || qinst->qpu.flags.mpf != V3D_QPU_PF_NONE || @@ -302,11 +302,11 @@ v3d_generate_code_block(struct v3d_compile *c, assert(qinst->qpu.alu.mul.op == V3D_QPU_M_NOP); if (nsrc >= 1) { set_src(&qinst->qpu, - &qinst->qpu.alu.add.a, src[0]); + &qinst->qpu.alu.add.a.mux, src[0]); } if (nsrc >= 2) { set_src(&qinst->qpu, - &qinst->qpu.alu.add.b, src[1]); + &qinst->qpu.alu.add.b.mux, src[1]); } qinst->qpu.alu.add.waddr = dst.index; @@ -314,11 +314,11 @@ v3d_generate_code_block(struct v3d_compile *c, } else { if (nsrc >= 1) { set_src(&qinst->qpu, - &qinst->qpu.alu.mul.a, src[0]); + &qinst->qpu.alu.mul.a.mux, src[0]); } if (nsrc >= 2) { set_src(&qinst->qpu, - &qinst->qpu.alu.mul.b, src[1]); + &qinst->qpu.alu.mul.b.mux, src[1]); } qinst->qpu.alu.mul.waddr = dst.index; diff --git a/src/broadcom/qpu/qpu_disasm.c b/src/broadcom/qpu/qpu_disasm.c index 6aca3c2..588a665 100644 --- a/src/broadcom/qpu/qpu_disasm.c +++ b/src/broadcom/qpu/qpu_disasm.c @@ -121,16 +121,16 @@ v3d_qpu_disasm_add(struct disasm_state *disasm, if (num_src >= 1) { if (has_dst) append(disasm, ", "); - v3d_qpu_disasm_raddr(disasm, instr, instr->alu.add.a); + v3d_qpu_disasm_raddr(disasm, instr, instr->alu.add.a.mux); append(disasm, "%s", - v3d_qpu_unpack_name(instr->alu.add.a_unpack)); + v3d_qpu_unpack_name(instr->alu.add.a.unpack)); } if (num_src >= 2) { append(disasm, ", "); - v3d_qpu_disasm_raddr(disasm, instr, instr->alu.add.b); + v3d_qpu_disasm_raddr(disasm, instr, instr->alu.add.b.mux); append(disasm, "%s", - v3d_qpu_unpack_name(instr->alu.add.b_unpack)); + v3d_qpu_unpack_name(instr->alu.add.b.unpack)); } } @@ -164,16 +164,16 @@ v3d_qpu_disasm_mul(struct disasm_state *disasm, if (num_src >= 1) { if (has_dst) append(disasm, ", "); - v3d_qpu_disasm_raddr(disasm, instr, instr->alu.mul.a); + v3d_qpu_disasm_raddr(disasm, instr, instr->alu.mul.a.mux); append(disasm, "%s", - v3d_qpu_unpack_name(instr->alu.mul.a_unpack)); + v3d_qpu_unpack_name(instr->alu.mul.a.unpack)); } if (num_src >= 2) { append(disasm, ", "); - v3d_qpu_disasm_raddr(disasm, instr, instr->alu.mul.b); + v3d_qpu_disasm_raddr(disasm, instr, instr->alu.mul.b.mux); append(disasm, "%s", - v3d_qpu_unpack_name(instr->alu.mul.b_unpack)); + v3d_qpu_unpack_name(instr->alu.mul.b.unpack)); } } diff --git a/src/broadcom/qpu/qpu_instr.c b/src/broadcom/qpu/qpu_instr.c index 7759fb0..7ece8b5 100644 --- a/src/broadcom/qpu/qpu_instr.c +++ b/src/broadcom/qpu/qpu_instr.c @@ -926,10 +926,10 @@ v3d_qpu_uses_mux(const struct v3d_qpu_instr *inst, enum v3d_qpu_mux mux) int add_nsrc = v3d_qpu_add_op_num_src(inst->alu.add.op); int mul_nsrc = v3d_qpu_mul_op_num_src(inst->alu.mul.op); - return ((add_nsrc > 0 && inst->alu.add.a == mux) || - (add_nsrc > 1 && inst->alu.add.b == mux) || - (mul_nsrc > 0 && inst->alu.mul.a == mux) || - (mul_nsrc > 1 && inst->alu.mul.b == mux)); + return ((add_nsrc > 0 && inst->alu.add.a.mux == mux) || + (add_nsrc > 1 && inst->alu.add.b.mux == mux) || + (mul_nsrc > 0 && inst->alu.mul.a.mux == mux) || + (mul_nsrc > 1 && inst->alu.mul.b.mux == mux)); } bool diff --git a/src/broadcom/qpu/qpu_instr.h b/src/broadcom/qpu/qpu_instr.h index 13b3f37..53a51bf 100644 --- a/src/broadcom/qpu/qpu_instr.h +++ b/src/broadcom/qpu/qpu_instr.h @@ -294,25 +294,26 @@ enum v3d_qpu_mux { V3D_QPU_MUX_B, }; +struct v3d_qpu_input { + enum v3d_qpu_mux mux; + enum v3d_qpu_input_unpack unpack; +}; + struct v3d_qpu_alu_instr { struct { enum v3d_qpu_add_op op; - enum v3d_qpu_mux a, b; + struct v3d_qpu_input a, b; uint8_t waddr; bool magic_write; enum v3d_qpu_output_pack output_pack; - enum v3d_qpu_input_unpack a_unpack; - enum v3d_qpu_input_unpack b_unpack; } add; struct { enum v3d_qpu_mul_op op; - enum v3d_qpu_mux a, b; + struct v3d_qpu_input a, b; uint8_t waddr; bool magic_write; enum v3d_qpu_output_pack output_pack; - enum v3d_qpu_input_unpack a_unpack; - enum v3d_qpu_input_unpack b_unpack; } mul; }; diff --git a/src/broadcom/qpu/qpu_pack.c b/src/broadcom/qpu/qpu_pack.c index 2820d9d..6e97579 100644 --- a/src/broadcom/qpu/qpu_pack.c +++ b/src/broadcom/qpu/qpu_pack.c @@ -853,12 +853,12 @@ v3d_qpu_add_unpack(const struct v3d_device_info *devinfo, uint64_t packed_inst, instr->alu.add.output_pack = V3D_QPU_PACK_NONE; if (!v3d_qpu_float32_unpack_unpack((op >> 2) & 0x3, - &instr->alu.add.a_unpack)) { + &instr->alu.add.a.unpack)) { return false; } if (!v3d_qpu_float32_unpack_unpack((op >> 0) & 0x3, - &instr->alu.add.b_unpack)) { + &instr->alu.add.b.unpack)) { return false; } break; @@ -872,7 +872,7 @@ v3d_qpu_add_unpack(const struct v3d_device_info *devinfo, uint64_t packed_inst, instr->alu.add.output_pack = mux_b & 0x3; if (!v3d_qpu_float32_unpack_unpack((op >> 2) & 0x3, - &instr->alu.add.a_unpack)) { + &instr->alu.add.a.unpack)) { return false; } break; @@ -884,7 +884,7 @@ v3d_qpu_add_unpack(const struct v3d_device_info *devinfo, uint64_t packed_inst, instr->alu.add.output_pack = V3D_QPU_PACK_NONE; if (!v3d_qpu_float32_unpack_unpack((op >> 2) & 0x3, - &instr->alu.add.a_unpack)) { + &instr->alu.add.a.unpack)) { return false; } break; @@ -892,23 +892,23 @@ v3d_qpu_add_unpack(const struct v3d_device_info *devinfo, uint64_t packed_inst, case V3D_QPU_A_VFMIN: case V3D_QPU_A_VFMAX: if (!v3d_qpu_float16_unpack_unpack(op & 0x7, - &instr->alu.add.a_unpack)) { + &instr->alu.add.a.unpack)) { return false; } instr->alu.add.output_pack = V3D_QPU_PACK_NONE; - instr->alu.add.b_unpack = V3D_QPU_UNPACK_NONE; + instr->alu.add.b.unpack = V3D_QPU_UNPACK_NONE; break; default: instr->alu.add.output_pack = V3D_QPU_PACK_NONE; - instr->alu.add.a_unpack = V3D_QPU_UNPACK_NONE; - instr->alu.add.b_unpack = V3D_QPU_UNPACK_NONE; + instr->alu.add.a.unpack = V3D_QPU_UNPACK_NONE; + instr->alu.add.b.unpack = V3D_QPU_UNPACK_NONE; break; } - instr->alu.add.a = mux_a; - instr->alu.add.b = mux_b; + instr->alu.add.a.mux = mux_a; + instr->alu.add.b.mux = mux_b; instr->alu.add.waddr = QPU_GET_FIELD(packed_inst, V3D_QPU_WADDR_A); instr->alu.add.magic_write = false; @@ -956,12 +956,12 @@ v3d_qpu_mul_unpack(const struct v3d_device_info *devinfo, uint64_t packed_inst, instr->alu.mul.output_pack = ((op >> 4) & 0x3) - 1; if (!v3d_qpu_float32_unpack_unpack((op >> 2) & 0x3, - &instr->alu.mul.a_unpack)) { + &instr->alu.mul.a.unpack)) { return false; } if (!v3d_qpu_float32_unpack_unpack((op >> 0) & 0x3, - &instr->alu.mul.b_unpack)) { + &instr->alu.mul.b.unpack)) { return false; } @@ -972,7 +972,7 @@ v3d_qpu_mul_unpack(const struct v3d_device_info *devinfo, uint64_t packed_inst, ((mux_b >> 2) & 1)); if (!v3d_qpu_float32_unpack_unpack(mux_b & 0x3, - &instr->alu.mul.a_unpack)) { + &instr->alu.mul.a.unpack)) { return false; } @@ -982,23 +982,23 @@ v3d_qpu_mul_unpack(const struct v3d_device_info *devinfo, uint64_t packed_inst, instr->alu.mul.output_pack = V3D_QPU_PACK_NONE; if (!v3d_qpu_float16_unpack_unpack(((op & 0x7) - 4) & 7, - &instr->alu.mul.a_unpack)) { + &instr->alu.mul.a.unpack)) { return false; } - instr->alu.mul.b_unpack = V3D_QPU_UNPACK_NONE; + instr->alu.mul.b.unpack = V3D_QPU_UNPACK_NONE; break; default: instr->alu.mul.output_pack = V3D_QPU_PACK_NONE; - instr->alu.mul.a_unpack = V3D_QPU_UNPACK_NONE; - instr->alu.mul.b_unpack = V3D_QPU_UNPACK_NONE; + instr->alu.mul.a.unpack = V3D_QPU_UNPACK_NONE; + instr->alu.mul.b.unpack = V3D_QPU_UNPACK_NONE; break; } - instr->alu.mul.a = mux_a; - instr->alu.mul.b = mux_b; + instr->alu.mul.a.mux = mux_a; + instr->alu.mul.b.mux = mux_b; instr->alu.mul.waddr = QPU_GET_FIELD(packed_inst, V3D_QPU_WADDR_M); instr->alu.mul.magic_write = packed_inst & V3D_QPU_MM; @@ -1030,8 +1030,8 @@ v3d_qpu_add_pack(const struct v3d_device_info *devinfo, const struct v3d_qpu_instr *instr, uint64_t *packed_instr) { uint32_t waddr = instr->alu.add.waddr; - uint32_t mux_a = instr->alu.add.a; - uint32_t mux_b = instr->alu.add.b; + uint32_t mux_a = instr->alu.add.a.mux; + uint32_t mux_b = instr->alu.add.b.mux; int nsrc = v3d_qpu_add_op_num_src(instr->alu.add.op); const struct opcode_desc *desc = lookup_opcode_from_instr(devinfo, add_ops, ARRAY_SIZE(add_ops), @@ -1102,12 +1102,12 @@ v3d_qpu_add_pack(const struct v3d_device_info *devinfo, } opcode |= output_pack << 4; - if (!v3d_qpu_float32_unpack_pack(instr->alu.add.a_unpack, + 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, + if (!v3d_qpu_float32_unpack_pack(instr->alu.add.b.unpack, &b_unpack)) { return false; } @@ -1141,17 +1141,17 @@ v3d_qpu_add_pack(const struct v3d_device_info *devinfo, 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) { + 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, + 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, + if (!v3d_qpu_float32_unpack_pack(instr->alu.add.b.unpack, &b_unpack)) { return false; } @@ -1176,7 +1176,7 @@ v3d_qpu_add_pack(const struct v3d_device_info *devinfo, } mux_b |= packed; - if (!v3d_qpu_float32_unpack_pack(instr->alu.add.a_unpack, + if (!v3d_qpu_float32_unpack_pack(instr->alu.add.a.unpack, &packed)) { return false; } @@ -1194,7 +1194,7 @@ v3d_qpu_add_pack(const struct v3d_device_info *devinfo, return false; uint32_t packed; - if (!v3d_qpu_float32_unpack_pack(instr->alu.add.a_unpack, + if (!v3d_qpu_float32_unpack_pack(instr->alu.add.a.unpack, &packed)) { return false; } @@ -1207,11 +1207,11 @@ v3d_qpu_add_pack(const struct v3d_device_info *devinfo, case V3D_QPU_A_VFMIN: case V3D_QPU_A_VFMAX: if (instr->alu.add.output_pack != V3D_QPU_PACK_NONE || - instr->alu.add.b_unpack != V3D_QPU_UNPACK_NONE) { + instr->alu.add.b.unpack != V3D_QPU_UNPACK_NONE) { return false; } - if (!v3d_qpu_float16_unpack_pack(instr->alu.add.a_unpack, + if (!v3d_qpu_float16_unpack_pack(instr->alu.add.a.unpack, &packed)) { return false; } @@ -1221,8 +1221,8 @@ v3d_qpu_add_pack(const struct v3d_device_info *devinfo, default: if (instr->alu.add.op != V3D_QPU_A_NOP && (instr->alu.add.output_pack != V3D_QPU_PACK_NONE || - instr->alu.add.a_unpack != V3D_QPU_UNPACK_NONE || - instr->alu.add.b_unpack != V3D_QPU_UNPACK_NONE)) { + instr->alu.add.a.unpack != V3D_QPU_UNPACK_NONE || + instr->alu.add.b.unpack != V3D_QPU_UNPACK_NONE)) { return false; } break; @@ -1242,8 +1242,8 @@ static bool v3d_qpu_mul_pack(const struct v3d_device_info *devinfo, const struct v3d_qpu_instr *instr, uint64_t *packed_instr) { - uint32_t mux_a = instr->alu.mul.a; - uint32_t mux_b = instr->alu.mul.b; + uint32_t mux_a = instr->alu.mul.a.mux; + uint32_t mux_b = instr->alu.mul.b.mux; int nsrc = v3d_qpu_mul_op_num_src(instr->alu.mul.op); const struct opcode_desc *desc = @@ -1277,13 +1277,13 @@ v3d_qpu_mul_pack(const struct v3d_device_info *devinfo, */ opcode += packed << 4; - if (!v3d_qpu_float32_unpack_pack(instr->alu.mul.a_unpack, + if (!v3d_qpu_float32_unpack_pack(instr->alu.mul.a.unpack, &packed)) { return false; } opcode |= packed << 2; - if (!v3d_qpu_float32_unpack_pack(instr->alu.mul.b_unpack, + if (!v3d_qpu_float32_unpack_pack(instr->alu.mul.b.unpack, &packed)) { return false; } @@ -1301,7 +1301,7 @@ v3d_qpu_mul_pack(const struct v3d_device_info *devinfo, opcode |= (packed >> 1) & 1; mux_b = (packed & 1) << 2; - if (!v3d_qpu_float32_unpack_pack(instr->alu.mul.a_unpack, + if (!v3d_qpu_float32_unpack_pack(instr->alu.mul.a.unpack, &packed)) { return false; } @@ -1315,16 +1315,16 @@ v3d_qpu_mul_pack(const struct v3d_device_info *devinfo, if (instr->alu.mul.output_pack != V3D_QPU_PACK_NONE) return false; - if (!v3d_qpu_float16_unpack_pack(instr->alu.mul.a_unpack, + if (!v3d_qpu_float16_unpack_pack(instr->alu.mul.a.unpack, &packed)) { return false; } - if (instr->alu.mul.a_unpack == V3D_QPU_UNPACK_SWAP_16) + if (instr->alu.mul.a.unpack == V3D_QPU_UNPACK_SWAP_16) opcode = 8; else opcode |= (packed + 4) & 7; - if (instr->alu.mul.b_unpack != V3D_QPU_UNPACK_NONE) + if (instr->alu.mul.b.unpack != V3D_QPU_UNPACK_NONE) return false; break; diff --git a/src/broadcom/qpu/tests/qpu_disasm.c b/src/broadcom/qpu/tests/qpu_disasm.c index 2f8e19c..be7b78d 100644 --- a/src/broadcom/qpu/tests/qpu_disasm.c +++ b/src/broadcom/qpu/tests/qpu_disasm.c @@ -160,10 +160,10 @@ main(int argc, char **argv) /* Swap the operands to be sure that we test * how the QPUs distinguish between these ops. */ - swap_mux(&instr.alu.add.a, - &instr.alu.add.b); - swap_pack(&instr.alu.add.a_unpack, - &instr.alu.add.b_unpack); + swap_mux(&instr.alu.add.a.mux, + &instr.alu.add.b.mux); + swap_pack(&instr.alu.add.a.unpack, + &instr.alu.add.b.unpack); break; default: break; -- 2.7.4