i965/fs: Don't use instruction list after calculating the cfg.

The only trick is changing a break into a return true in register
coalescing, since the macro is actually a double loop, and break will do
something different than you expect. (Wish I'd realized that earlier!)

Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>

diff --git a/src/mesa/drivers/dri/i965/brw_fs.cpp b/src/mesa/drivers/dri/i965/brw_fs.cpp
index 5c78b7ae3c705dc24c1bd94572aa41bbf729fd8f..e3bb807f9cb43932db9574b6688596f37e0ad966 100644 (file)
--- a/src/mesa/drivers/dri/i965/brw_fs.cpp
+++ b/src/mesa/drivers/dri/i965/brw_fs.cpp
@@ -1478,7 +1478,7 @@ fs_visitor::assign_curb_setup()
    prog_data->curb_read_length = ALIGN(stage_prog_data->nr_params, 8) / 8;
 
    /* Map the offsets in the UNIFORM file to fixed HW regs. */
-   foreach_in_list(fs_inst, inst, &instructions) {
+   foreach_block_and_inst(block, fs_inst, inst, cfg) {
       for (unsigned int i = 0; i < inst->sources; i++) {
         if (inst->src[i].file == UNIFORM) {
             int uniform_nr = inst->src[i].reg + inst->src[i].reg_offset;
@@ -1607,7 +1607,7 @@ fs_visitor::assign_urb_setup()
    /* Offset all the urb_setup[] index by the actual position of the
     * setup regs, now that the location of the constants has been chosen.
     */
-   foreach_in_list(fs_inst, inst, &instructions) {
+   foreach_block_and_inst(block, fs_inst, inst, cfg) {
       if (inst->opcode == FS_OPCODE_LINTERP) {
         assert(inst->src[2].file == HW_REG);
         inst->src[2].fixed_hw_reg.nr += urb_start;
@@ -1668,7 +1668,7 @@ fs_visitor::split_virtual_grfs()
          false;
    }
 
-   foreach_in_list(fs_inst, inst, &instructions) {
+   foreach_block_and_inst(block, fs_inst, inst, cfg) {
       /* If there's a SEND message that requires contiguous destination
        * registers, no splitting is allowed.
        */
@@ -1703,7 +1703,7 @@ fs_visitor::split_virtual_grfs()
       }
    }
 
-   foreach_in_list(fs_inst, inst, &instructions) {
+   foreach_block_and_inst(block, fs_inst, inst, cfg) {
       if (inst->dst.file == GRF &&
          split_grf[inst->dst.reg] &&
          inst->dst.reg_offset != 0) {
@@ -1743,7 +1743,7 @@ fs_visitor::compact_virtual_grfs()
    int remap_table[this->virtual_grf_count];
    memset(remap_table, -1, sizeof(remap_table));
 
-   foreach_in_list(const fs_inst, inst, &instructions) {
+   foreach_block_and_inst(block, const fs_inst, inst, cfg) {
       if (inst->dst.file == GRF)
          remap_table[inst->dst.reg] = 0;
 
@@ -1767,7 +1767,7 @@ fs_visitor::compact_virtual_grfs()
    this->virtual_grf_count = new_index;
 
    /* Patch all the instructions to use the newly renumbered registers */
-   foreach_in_list(fs_inst, inst, &instructions) {
+   foreach_block_and_inst(block, fs_inst, inst, cfg) {
       if (inst->dst.file == GRF)
          inst->dst.reg = remap_table[inst->dst.reg];
 
@@ -1831,7 +1831,7 @@ fs_visitor::move_uniform_array_access_to_pull_constants()
     * Note that we don't move constant-indexed accesses to arrays.  No
     * testing has been done of the performance impact of this choice.
     */
-   foreach_in_list_safe(fs_inst, inst, &instructions) {
+   foreach_block_and_inst_safe(block, fs_inst, inst, cfg) {
       for (int i = 0 ; i < inst->sources; i++) {
          if (inst->src[i].file != UNIFORM || !inst->src[i].reladdr)
             continue;
@@ -1879,7 +1879,7 @@ fs_visitor::assign_constant_locations()
       is_live[i] = false;
    }
 
-   foreach_in_list(fs_inst, inst, &instructions) {
+   foreach_block_and_inst(block, fs_inst, inst, cfg) {
       for (int i = 0; i < inst->sources; i++) {
          if (inst->src[i].file != UNIFORM)
             continue;
@@ -1999,7 +1999,7 @@ fs_visitor::opt_algebraic()
 {
    bool progress = false;
 
-   foreach_in_list(fs_inst, inst, &instructions) {
+   foreach_block_and_inst(block, fs_inst, inst, cfg) {
       switch (inst->opcode) {
       case BRW_OPCODE_MUL:
         if (inst->src[1].file != IMM)
@@ -2112,7 +2112,7 @@ fs_visitor::opt_register_renaming()
    int remap[virtual_grf_count];
    memset(remap, -1, sizeof(int) * virtual_grf_count);
 
-   foreach_in_list(fs_inst, inst, &this->instructions) {
+   foreach_block_and_inst(block, fs_inst, inst, cfg) {
       if (inst->opcode == BRW_OPCODE_IF || inst->opcode == BRW_OPCODE_DO) {
          depth++;
       } else if (inst->opcode == BRW_OPCODE_ENDIF ||
@@ -2827,7 +2827,7 @@ fs_visitor::dump_instructions(const char *name)
    }
 
    int ip = 0, max_pressure = 0;
-   foreach_in_list(backend_instruction, inst, &instructions) {
+   foreach_block_and_inst(block, backend_instruction, inst, cfg) {
       max_pressure = MAX2(max_pressure, regs_live_at_ip[ip]);
       fprintf(file, "{%3d} %4d: ", regs_live_at_ip[ip], ip);
       dump_instruction(inst, file);

diff --git a/src/mesa/drivers/dri/i965/brw_fs_reg_allocate.cpp b/src/mesa/drivers/dri/i965/brw_fs_reg_allocate.cpp
index eb443d22d8344d30aa89b06ca9429ddf10755a3d..24c3d3a21d64c52219e7e2ca7e678c12320d511a 100644 (file)
--- a/src/mesa/drivers/dri/i965/brw_fs_reg_allocate.cpp
+++ b/src/mesa/drivers/dri/i965/brw_fs_reg_allocate.cpp
@@ -55,7 +55,7 @@ fs_visitor::assign_regs_trivial()
    }
    this->grf_used = hw_reg_mapping[this->virtual_grf_count];
 
-   foreach_in_list(fs_inst, inst, &instructions) {
+   foreach_block_and_inst(block, fs_inst, inst, cfg) {
       assign_reg(hw_reg_mapping, &inst->dst, reg_width);
       for (i = 0; i < inst->sources; i++) {
          assign_reg(hw_reg_mapping, &inst->src[i], reg_width);
@@ -242,7 +242,7 @@ fs_visitor::setup_payload_interference(struct ra_graph *g,
    int payload_last_use_ip[payload_node_count];
    memset(payload_last_use_ip, 0, sizeof(payload_last_use_ip));
    int ip = 0;
-   foreach_in_list(fs_inst, inst, &instructions) {
+   foreach_block_and_inst(block, fs_inst, inst, cfg) {
       switch (inst->opcode) {
       case BRW_OPCODE_DO:
          loop_depth++;
@@ -362,7 +362,7 @@ fs_visitor::get_used_mrfs(bool *mrf_used)
 
    memset(mrf_used, 0, BRW_MAX_MRF * sizeof(bool));
 
-   foreach_in_list(fs_inst, inst, &instructions) {
+   foreach_block_and_inst(block, fs_inst, inst, cfg) {
       if (inst->dst.file == MRF) {
          int reg = inst->dst.reg & ~BRW_MRF_COMPR4;
          mrf_used[reg] = true;
@@ -520,7 +520,7 @@ fs_visitor::assign_regs(bool allow_spilling)
                            reg_width);
    }
 
-   foreach_in_list(fs_inst, inst, &instructions) {
+   foreach_block_and_inst(block, fs_inst, inst, cfg) {
       assign_reg(hw_reg_mapping, &inst->dst, reg_width);
       for (int i = 0; i < inst->sources; i++) {
          assign_reg(hw_reg_mapping, &inst->src[i], reg_width);
@@ -578,7 +578,7 @@ fs_visitor::choose_spill_reg(struct ra_graph *g)
     * spill/unspill we'll have to do, and guess that the insides of
     * loops run 10 times.
     */
-   foreach_in_list(fs_inst, inst, &instructions) {
+   foreach_block_and_inst(block, fs_inst, inst, cfg) {
       for (unsigned int i = 0; i < inst->sources; i++) {
         if (inst->src[i].file == GRF) {
            spill_costs[inst->src[i].reg] += loop_scale;

diff --git a/src/mesa/drivers/dri/i965/brw_fs_register_coalesce.cpp b/src/mesa/drivers/dri/i965/brw_fs_register_coalesce.cpp
index b4abb4caa7e9c6961076338404b0e4af2f3b028d..318bfa610d4c41c721f51a6921b8ab19ab81e515 100644 (file)
--- a/src/mesa/drivers/dri/i965/brw_fs_register_coalesce.cpp
+++ b/src/mesa/drivers/dri/i965/brw_fs_register_coalesce.cpp
@@ -113,7 +113,7 @@ is_coalesce_candidate(const fs_inst *inst, const int *virtual_grf_sizes)
 
 static bool
 can_coalesce_vars(brw::fs_live_variables *live_intervals,
-                  const exec_list *instructions, const fs_inst *inst,
+                  const cfg_t *cfg, const fs_inst *inst,
                   int var_to, int var_from)
 {
    if (!live_intervals->vars_interfere(var_from, var_to))
@@ -132,7 +132,7 @@ can_coalesce_vars(brw::fs_live_variables *live_intervals,
    int start_ip = MIN2(start_to, start_from);
    int scan_ip = -1;
 
-   foreach_in_list(fs_inst, scan_inst, instructions) {
+   foreach_block_and_inst(block, fs_inst, scan_inst, cfg) {
       scan_ip++;
 
       if (scan_ip < start_ip)
@@ -145,7 +145,7 @@ can_coalesce_vars(brw::fs_live_variables *live_intervals,
          continue;
 
       if (scan_ip > live_intervals->end[var_to])
-         break;
+         return true;
 
       if (scan_inst->dst.equals(inst->dst) ||
           scan_inst->dst.equals(inst->src[0]))
@@ -170,7 +170,7 @@ fs_visitor::register_coalesce()
    int var_to[MAX_SAMPLER_MESSAGE_SIZE];
    int var_from[MAX_SAMPLER_MESSAGE_SIZE];
 
-   foreach_in_list(fs_inst, inst, &instructions) {
+   foreach_block_and_inst(block, fs_inst, inst, cfg) {
       if (!is_coalesce_candidate(inst, virtual_grf_sizes))
          continue;
 
@@ -216,7 +216,7 @@ fs_visitor::register_coalesce()
          var_to[i] = live_intervals->var_from_vgrf[reg_to] + reg_to_offset[i];
          var_from[i] = live_intervals->var_from_vgrf[reg_from] + i;
 
-         if (!can_coalesce_vars(live_intervals, &instructions, inst,
+         if (!can_coalesce_vars(live_intervals, cfg, inst,
                                 var_to[i], var_from[i])) {
             can_coalesce = false;
             reg_from = -1;
@@ -241,7 +241,7 @@ fs_visitor::register_coalesce()
          }
       }
 
-      foreach_in_list(fs_inst, scan_inst, &instructions) {
+      foreach_block_and_inst(block, fs_inst, scan_inst, cfg) {
          for (int i = 0; i < src_size; i++) {
             if (mov[i] || was_load_payload) {
                if (scan_inst->dst.file == GRF &&