match immwip/const-vf

1 files changed, 77 insertions, 0 deletions

diff --git a/src/mesa/drivers/dri/i965/brw_fs_combine_constants.cpp b/src/mesa/drivers/dri/i965/brw_fs_combine_constants.cpp
index dd519dc7cb6..fe9d8a67e00 100644
--- a/src/mesa/drivers/dri/i965/brw_fs_combine_constants.cpp
+++ b/src/mesa/drivers/dri/i965/brw_fs_combine_constants.cpp
@@ -190,6 +190,63 @@ compare(const void *_a, const void *_b)
    return a->first_use_ip - b->first_use_ip;
 }
 
+static void
+match_immediates(fs_visitor *v, struct table *table,
+                 const float value[8], float multiplicand,
+                 const uint8_t first_4_vf[4], const uint8_t second_4_vf[4])
+{
+   int block_num = -1;
+   int found = 0;
+   int idx[8];
+
+   for (int i = 0; i < table->len && found < 8; i++) {
+      for (int j = 0; j < 8; j++) {
+         if (table->imm[i].val == value[j]) {
+            if (block_num == -1) {
+               block_num = table->imm[i].block->num;
+            } else if (table->imm[i].block->num != block_num) {
+               return;
+            }
+            idx[j] = i;
+            found++;
+            break;
+         }
+      }
+   }
+
+   if (found != 8)
+      return;
+
+   fs_reg result(GRF, v->virtual_grf_alloc(1));
+   fs_reg reg(GRF, v->virtual_grf_alloc(1));
+   reg.width = 4;
+
+   for (int i = 0; i < 8; i++) {
+      table->imm[idx[i]].reg = result.reg;
+      table->imm[idx[i]].subreg_offset = i * sizeof(float);
+      table->imm[idx[i]].inserted = true;
+   }
+
+   fs_inst *res = v->MUL(result, reg, fs_reg(multiplicand));
+   fs_inst *_1 = v->MOV(reg, fs_reg(first_4_vf));
+   reg.subreg_offset = 16;
+   fs_inst *_2 = v->MOV(reg, fs_reg(second_4_vf));
+
+   bblock_t *block = v->cfg->blocks[block_num];
+   backend_instruction *inst = block->first_non_control_flow_inst();
+   inst->insert_before(block, _1);
+   inst->insert_before(block, _2);
+   inst->insert_before(block, res);
+
+   for (int i = 0; i < table->len; i++) {
+      if (table->imm[i].val == multiplicand) {
+         if (table->imm[i].inserted)
+            break;
+         table->imm[i].block = cfg_t::intersect(block, table->imm[i].block);
+      }
+   }
+}
+
 bool
 fs_visitor::opt_combine_constants()
 {
@@ -266,6 +323,23 @@ fs_visitor::opt_combine_constants()
    if (cfg->num_blocks != 1)
       qsort(table.imm, table.len, sizeof(struct imm), compare);
 
+   static const float ninths[8] = {
+      1/9., 2/9., 3/9., 4/9., 5/9., 6/9., 7/9., 8/9.
+   };
+   static const uint8_t one_to_four_vf[4] = { 0x30, 0x40, 0x48, 0x50 };
+   static const uint8_t five_to_eight_vf[4] = { 0x54, 0x58, 0x5c, 0x60 };
+
+   match_immediates(this, &table, ninths, 1.0f / 9.0f,
+                    one_to_four_vf, five_to_eight_vf);
+
+   static const float tenths[8] = {
+      0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9
+   };
+   static const uint8_t two_to_five_vf[4] = { 0x40, 0x48, 0x50, 0x54 };
+   static const uint8_t six_to_nine_vf[4] = { 0x58, 0x5c, 0x60, 0x62 };
+
+   match_immediates(this, &table, tenths, 0.1f,
+                    two_to_five_vf, six_to_nine_vf);
 
    /* Insert MOVs to load the constant values into GRFs. */
    fs_reg reg(GRF, alloc.allocate(dispatch_width / 8));
@@ -284,6 +358,9 @@ fs_visitor::opt_combine_constants()
             count = 0;
          }
 
+         if (table.imm[i].inserted)
+            continue;
+
          int vf = brw_float_to_vf(table.imm[i].val);
          if (vf == -1)
             continue;