i965/vec4: Don't spill non-GRF-aligned register regions.

A better fix would be to do something along the lines of the FS
back-end spilling code and emit a scratch read before any instruction
that overwrites the register to spill partially due to a non-zero
sub-register offset.  In the meantime mark registers used with a
non-zero sub-register offset as no-spill to prevent the spilling code
from miscompiling the program.

Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>

diff --git a/src/mesa/drivers/dri/i965/brw_vec4_reg_allocate.cpp b/src/mesa/drivers/dri/i965/brw_vec4_reg_allocate.cpp
index 947bb492bda7f766cf86a58a1cc4eef1b10204ea..228e04c44dec8e8b4bcccdb04a3f4ddae10f83e6 100644 (file)
--- a/src/mesa/drivers/dri/i965/brw_vec4_reg_allocate.cpp
+++ b/src/mesa/drivers/dri/i965/brw_vec4_reg_allocate.cpp
@@ -392,7 +392,8 @@ vec4_visitor::evaluate_spill_costs(float *spill_costs, bool *no_spill)
              */
             if (!can_use_scratch_for_source(inst, i, inst->src[i].nr)) {
                spill_costs[inst->src[i].nr] += loop_scale;
-               if (inst->src[i].reladdr)
+               if (inst->src[i].reladdr ||
+                   inst->src[i].offset % REG_SIZE != 0)
                   no_spill[inst->src[i].nr] = true;
             }
          }
@@ -400,7 +401,7 @@ vec4_visitor::evaluate_spill_costs(float *spill_costs, bool *no_spill)
 
       if (inst->dst.file == VGRF) {
          spill_costs[inst->dst.nr] += loop_scale;
-         if (inst->dst.reladdr)
+         if (inst->dst.reladdr || inst->dst.offset % REG_SIZE != 0)
             no_spill[inst->dst.nr] = true;
       }
 

diff --git a/src/mesa/drivers/dri/i965/brw_vec4_visitor.cpp b/src/mesa/drivers/dri/i965/brw_vec4_visitor.cpp
index a4e4f40ab2b4d5b7ddf77dcde87059330fe81c4b..c88d7b6f2291ec2633a5dff0948e9be4d461cff1 100644 (file)
--- a/src/mesa/drivers/dri/i965/brw_vec4_visitor.cpp
+++ b/src/mesa/drivers/dri/i965/brw_vec4_visitor.cpp
@@ -1481,6 +1481,7 @@ vec4_visitor::emit_scratch_read(bblock_t *block, vec4_instruction *inst,
                                dst_reg temp, src_reg orig_src,
                                int base_offset)
 {
+   assert(orig_src.offset % REG_SIZE == 0);
    int reg_offset = base_offset + orig_src.offset / REG_SIZE;
    src_reg index = get_scratch_offset(block, inst, orig_src.reladdr,
                                       reg_offset);
@@ -1498,6 +1499,7 @@ void
 vec4_visitor::emit_scratch_write(bblock_t *block, vec4_instruction *inst,
                                  int base_offset)
 {
+   assert(inst->dst.offset % REG_SIZE == 0);
    int reg_offset = base_offset + inst->dst.offset / REG_SIZE;
    src_reg index = get_scratch_offset(block, inst, inst->dst.reladdr,
                                       reg_offset);