+static void
+mir_lower_inverts(midgard_instruction *ins)
+{
+ bool inv[3] = {
+ ins->src_invert[0],
+ ins->src_invert[1],
+ ins->src_invert[2]
+ };
+
+ switch (ins->op) {
+ case midgard_alu_op_iand:
+ /* a & ~b = iandnot(a, b) */
+ /* ~a & ~b = ~(a | b) = inor(a, b) */
+
+ if (inv[0] && inv[1])
+ ins->op = midgard_alu_op_inor;
+ else if (inv[1])
+ ins->op = midgard_alu_op_iandnot;
+
+ break;
+ case midgard_alu_op_ior:
+ /* a | ~b = iornot(a, b) */
+ /* ~a | ~b = ~(a & b) = inand(a, b) */
+
+ if (inv[0] && inv[1])
+ ins->op = midgard_alu_op_inand;
+ else if (inv[1])
+ ins->op = midgard_alu_op_iornot;
+
+ break;
+
+ case midgard_alu_op_ixor:
+ /* ~a ^ b = a ^ ~b = ~(a ^ b) = inxor(a, b) */
+ /* ~a ^ ~b = a ^ b */
+
+ if (inv[0] ^ inv[1])
+ ins->op = midgard_alu_op_inxor;
+
+ break;
+
+ default:
+ break;
+ }
+}
+
+/* Opcodes with ROUNDS are the base (rte/0) type so we can just add */
+
+static void
+mir_lower_roundmode(midgard_instruction *ins)
+{
+ if (alu_opcode_props[ins->op].props & MIDGARD_ROUNDS) {
+ assert(ins->roundmode <= 0x3);
+ ins->op += ins->roundmode;
+ }
+}
+
+static midgard_load_store_word
+load_store_from_instr(midgard_instruction *ins)
+{
+ midgard_load_store_word ldst = ins->load_store;
+ ldst.op = ins->op;
+
+ if (OP_IS_STORE(ldst.op)) {
+ ldst.reg = SSA_REG_FROM_FIXED(ins->src[0]) & 1;
+ } else {
+ ldst.reg = SSA_REG_FROM_FIXED(ins->dest);
+ }
+
+ if (ins->src[1] != ~0) {
+ unsigned src = SSA_REG_FROM_FIXED(ins->src[1]);
+ ldst.arg_1 |= midgard_ldst_reg(src, ins->swizzle[1][0]);
+ }
+
+ if (ins->src[2] != ~0) {
+ unsigned src = SSA_REG_FROM_FIXED(ins->src[2]);
+ ldst.arg_2 |= midgard_ldst_reg(src, ins->swizzle[2][0]);
+ }
+
+ return ldst;
+}
+
+static midgard_texture_word
+texture_word_from_instr(midgard_instruction *ins)
+{
+ midgard_texture_word tex = ins->texture;
+ tex.op = ins->op;
+
+ unsigned src1 = ins->src[1] == ~0 ? REGISTER_UNUSED : SSA_REG_FROM_FIXED(ins->src[1]);
+ tex.in_reg_select = src1 & 1;
+
+ unsigned dest = ins->dest == ~0 ? REGISTER_UNUSED : SSA_REG_FROM_FIXED(ins->dest);
+ tex.out_reg_select = dest & 1;
+
+ if (ins->src[2] != ~0) {
+ midgard_tex_register_select sel = {
+ .select = SSA_REG_FROM_FIXED(ins->src[2]) & 1,
+ .full = 1,
+ .component = ins->swizzle[2][0]
+ };
+ uint8_t packed;
+ memcpy(&packed, &sel, sizeof(packed));
+ tex.bias = packed;
+ }
+
+ if (ins->src[3] != ~0) {
+ unsigned x = ins->swizzle[3][0];
+ unsigned y = x + 1;
+ unsigned z = x + 2;
+
+ /* Check range, TODO: half-registers */
+ assert(z < 4);
+
+ unsigned offset_reg = SSA_REG_FROM_FIXED(ins->src[3]);
+ tex.offset =
+ (1) | /* full */
+ (offset_reg & 1) << 1 | /* select */
+ (0 << 2) | /* upper */
+ (x << 3) | /* swizzle */
+ (y << 5) | /* swizzle */
+ (z << 7); /* swizzle */
+ }
+
+ return tex;
+}
+
+static midgard_vector_alu
+vector_alu_from_instr(midgard_instruction *ins)
+{
+ midgard_vector_alu alu = {
+ .op = ins->op,
+ .outmod = ins->outmod,
+ .reg_mode = reg_mode_for_bitsize(max_bitsize_for_alu(ins))
+ };
+
+ if (ins->has_inline_constant) {
+ /* Encode inline 16-bit constant. See disassembler for
+ * where the algorithm is from */
+
+ int lower_11 = ins->inline_constant & ((1 << 12) - 1);
+ uint16_t imm = ((lower_11 >> 8) & 0x7) |
+ ((lower_11 & 0xFF) << 3);
+
+ alu.src2 = imm << 2;
+ }
+
+ return alu;
+}
+
+static midgard_branch_extended
+midgard_create_branch_extended( midgard_condition cond,
+ midgard_jmp_writeout_op op,
+ unsigned dest_tag,
+ signed quadword_offset)
+{
+ /* The condition code is actually a LUT describing a function to
+ * combine multiple condition codes. However, we only support a single
+ * condition code at the moment, so we just duplicate over a bunch of
+ * times. */
+
+ uint16_t duplicated_cond =
+ (cond << 14) |
+ (cond << 12) |
+ (cond << 10) |
+ (cond << 8) |
+ (cond << 6) |
+ (cond << 4) |
+ (cond << 2) |
+ (cond << 0);
+
+ midgard_branch_extended branch = {
+ .op = op,
+ .dest_tag = dest_tag,
+ .offset = quadword_offset,
+ .cond = duplicated_cond
+ };
+
+ return branch;
+}
+
+static void
+emit_branch(midgard_instruction *ins,
+ compiler_context *ctx,
+ midgard_block *block,
+ midgard_bundle *bundle,
+ struct util_dynarray *emission)
+{
+ /* Parse some basic branch info */
+ bool is_compact = ins->unit == ALU_ENAB_BR_COMPACT;
+ bool is_conditional = ins->branch.conditional;
+ bool is_inverted = ins->branch.invert_conditional;
+ bool is_discard = ins->branch.target_type == TARGET_DISCARD;
+ bool is_tilebuf_wait = ins->branch.target_type == TARGET_TILEBUF_WAIT;
+ bool is_special = is_discard || is_tilebuf_wait;
+ bool is_writeout = ins->writeout;
+
+ /* Determine the block we're jumping to */
+ int target_number = ins->branch.target_block;
+
+ /* Report the destination tag */
+ int dest_tag = is_discard ? 0 :
+ is_tilebuf_wait ? bundle->tag :
+ midgard_get_first_tag_from_block(ctx, target_number);
+
+ /* Count up the number of quadwords we're
+ * jumping over = number of quadwords until
+ * (br_block_idx, target_number) */
+
+ int quadword_offset = 0;
+
+ if (is_discard) {
+ /* Ignored */
+ } else if (is_tilebuf_wait) {
+ quadword_offset = -1;
+ } else if (target_number > block->base.name) {
+ /* Jump forward */
+
+ for (int idx = block->base.name+1; idx < target_number; ++idx) {
+ midgard_block *blk = mir_get_block(ctx, idx);
+ assert(blk);
+
+ quadword_offset += blk->quadword_count;
+ }
+ } else {
+ /* Jump backwards */
+
+ for (int idx = block->base.name; idx >= target_number; --idx) {
+ midgard_block *blk = mir_get_block(ctx, idx);
+ assert(blk);
+
+ quadword_offset -= blk->quadword_count;
+ }
+ }
+
+ /* Unconditional extended branches (far jumps)
+ * have issues, so we always use a conditional
+ * branch, setting the condition to always for
+ * unconditional. For compact unconditional
+ * branches, cond isn't used so it doesn't
+ * matter what we pick. */
+
+ midgard_condition cond =
+ !is_conditional ? midgard_condition_always :
+ is_inverted ? midgard_condition_false :
+ midgard_condition_true;
+
+ midgard_jmp_writeout_op op =
+ is_discard ? midgard_jmp_writeout_op_discard :
+ is_tilebuf_wait ? midgard_jmp_writeout_op_tilebuffer_pending :
+ is_writeout ? midgard_jmp_writeout_op_writeout :
+ (is_compact && !is_conditional) ?
+ midgard_jmp_writeout_op_branch_uncond :
+ midgard_jmp_writeout_op_branch_cond;
+
+ if (is_compact) {
+ unsigned size = sizeof(midgard_branch_cond);
+
+ if (is_conditional || is_special) {
+ midgard_branch_cond branch = {
+ .op = op,
+ .dest_tag = dest_tag,
+ .offset = quadword_offset,
+ .cond = cond
+ };
+ memcpy(util_dynarray_grow_bytes(emission, size, 1), &branch, size);
+ } else {
+ assert(op == midgard_jmp_writeout_op_branch_uncond);
+ midgard_branch_uncond branch = {
+ .op = op,
+ .dest_tag = dest_tag,
+ .offset = quadword_offset,
+ .unknown = 1
+ };
+ assert(branch.offset == quadword_offset);
+ memcpy(util_dynarray_grow_bytes(emission, size, 1), &branch, size);
+ }
+ } else { /* `ins->compact_branch`, misnomer */
+ unsigned size = sizeof(midgard_branch_extended);
+
+ midgard_branch_extended branch =
+ midgard_create_branch_extended(
+ cond, op,
+ dest_tag,
+ quadword_offset);
+
+ memcpy(util_dynarray_grow_bytes(emission, size, 1), &branch, size);
+ }
+}
+