i965/fs: Lower 32x32 bit multiplication on BXT.

[mesa.git] / src / mesa / drivers / dri / i965 / brw_fs_reg_allocate.cpp

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 364fc4a5ad26ef8b7a1cd0c56cc42c17be0443b6..6a7ed64415dccf8559f24d7e09c7c9c3aa380315 100644 (file)
--- a/src/mesa/drivers/dri/i965/brw_fs_reg_allocate.cpp
+++ b/src/mesa/drivers/dri/i965/brw_fs_reg_allocate.cpp
@@ -341,7 +341,9 @@ fs_visitor::setup_payload_interference(struct ra_graph *g,
    int loop_end_ip = 0;
 
    int payload_last_use_ip[payload_node_count];
-   memset(payload_last_use_ip, 0, sizeof(payload_last_use_ip));
+   for (int i = 0; i < payload_node_count; i++)
+      payload_last_use_ip[i] = -1;
+
    int ip = 0;
    foreach_block_and_inst(block, fs_inst, inst, cfg) {
       switch (inst->opcode) {
@@ -380,32 +382,15 @@ fs_visitor::setup_payload_interference(struct ra_graph *g,
             if (node_nr >= payload_node_count)
                continue;
 
-            payload_last_use_ip[node_nr] = use_ip;
+            for (int j = 0; j < inst->regs_read(i); j++) {
+               payload_last_use_ip[node_nr + j] = use_ip;
+               assert(node_nr + j < payload_node_count);
+            }
          }
       }
 
       /* Special case instructions which have extra implied registers used. */
       switch (inst->opcode) {
-      case FS_OPCODE_LINTERP:
-         /* On gen6+ in SIMD16, there are 4 adjacent registers used by
-          * PLN's sourcing of the deltas, while we list only the first one
-          * in the arguments.  Pre-gen6, the deltas are computed in normal
-          * VGRFs.
-          */
-         if (devinfo->gen >= 6) {
-            int delta_x_arg = 0;
-            if (inst->src[delta_x_arg].file == HW_REG &&
-                inst->src[delta_x_arg].fixed_hw_reg.file ==
-                BRW_GENERAL_REGISTER_FILE) {
-               for (int i = 1; i < 4; ++i) {
-                  int node = inst->src[delta_x_arg].fixed_hw_reg.nr + i;
-                  assert(node < payload_node_count);
-                  payload_last_use_ip[node] = use_ip;
-               }
-            }
-         }
-         break;
-
       case CS_OPCODE_CS_TERMINATE:
          payload_last_use_ip[0] = use_ip;
          break;
@@ -428,6 +413,9 @@ fs_visitor::setup_payload_interference(struct ra_graph *g,
    }
 
    for (int i = 0; i < payload_node_count; i++) {
+      if (payload_last_use_ip[i] == -1)
+         continue;
+
       /* Mark the payload node as interfering with any virtual grf that is
        * live between the start of the program and our last use of the payload
        * node.
@@ -706,10 +694,8 @@ fs_visitor::emit_unspill(bblock_t *block, fs_inst *inst, fs_reg dst,
                          uint32_t spill_offset, int count)
 {
    int reg_size = 1;
-   if (dispatch_width == 16 && count % 2 == 0) {
+   if (dispatch_width == 16 && count % 2 == 0)
       reg_size = 2;
-      dst.width = 16;
-   }
 
    const fs_builder ibld = bld.annotate(inst->annotation, inst->ir)
                               .group(reg_size * 8, 0)
@@ -752,7 +738,7 @@ fs_visitor::emit_spill(bblock_t *block, fs_inst *inst, fs_reg src,
 
    for (int i = 0; i < count / reg_size; i++) {
       fs_inst *spill_inst =
-         ibld.emit(SHADER_OPCODE_GEN4_SCRATCH_WRITE, bld.null_reg_f(), src);
+         ibld.emit(SHADER_OPCODE_GEN4_SCRATCH_WRITE, ibld.null_reg_f(), src);
       src.reg_offset += reg_size;
       spill_inst->offset = spill_offset + i * reg_size * REG_SIZE;
       spill_inst->mlen = 1 + reg_size; /* header, value */