commit 59e6c0d8aee718cf58198d5a5b2adce3e01391a6
author Samuel Iglesias Gonsálvez <siglesias@igalia.com> Mon Feb 13 13:24:18 2017 +0100
committer Emil Velikov <emil.l.velikov@gmail.com> Wed Mar 15 18:54:31 2017 +0000
tree 2e0bf6c4c0200a390ed95fe00b87bcca543ee0b6
parent 7d3a10c516eb5b1b4058bee4640c5fbb22617f5b

i965/fs: fix indirect load DF uniforms on BSW/BXT

The lowered BSW/BXT indirect move instructions had incorrect
source types, which luckily wasn't causing incorrect assembly to be
generated due to the bug fixed in the next patch, but would have
confused the remaining back-end IR infrastructure due to the mismatch
between the IR source types and the emitted machine code.

v2:
- Improve commit log (Curro)
- Fix read_size (Curro)
- Fix DF uniform array detection in assign_constant_locations() when
  it is acceded with 32-bit MOV_INDIRECTs in BSW/BXT.

v3:
- Move changes in assign_constant_locations() to other patch.

Signed-off-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Cc: "17.0" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
(cherry picked from commit 56266df7ed9dbdf63acfd58944442893b4cd0c0b)

src/mesa/drivers/dri/i965/brw_fs_nir.cpp

diff --git a/src/mesa/drivers/dri/i965/brw_fs_nir.cpp b/src/mesa/drivers/dri/i965/brw_fs_nir.cpp
index ff29711..8c679dd 100644
--- a/src/mesa/drivers/dri/i965/brw_fs_nir.cpp
+++ b/src/mesa/drivers/dri/i965/brw_fs_nir.cpp
@@ -3792,31 +3792,30 @@
          unsigned read_size = instr->const_index[1] -
             (instr->num_components - 1) * type_sz(dest.type);
 
-         fs_reg indirect_chv_high_32bit;
-         bool is_chv_bxt_64bit =
-            (devinfo->is_cherryview || devinfo->is_broxton) &&
-            type_sz(dest.type) == 8;
-         if (is_chv_bxt_64bit) {
-            indirect_chv_high_32bit = vgrf(glsl_type::uint_type);
-            /* Calculate indirect address to read high 32 bits */
-            bld.ADD(indirect_chv_high_32bit, indirect, brw_imm_ud(4));
-         }
+         bool supports_64bit_indirects =
+            !devinfo->is_cherryview && !devinfo->is_broxton;
 
-         for (unsigned j = 0; j < instr->num_components; j++) {
-            if (!is_chv_bxt_64bit) {
+         if (type_sz(dest.type) != 8 || supports_64bit_indirects) {
+            for (unsigned j = 0; j < instr->num_components; j++) {
                bld.emit(SHADER_OPCODE_MOV_INDIRECT,
                         offset(dest, bld, j), offset(src, bld, j),
                         indirect, brw_imm_ud(read_size));
-            } else {
-               bld.emit(SHADER_OPCODE_MOV_INDIRECT,
-                        subscript(offset(dest, bld, j), BRW_REGISTER_TYPE_UD, 0),
-                        offset(src, bld, j),
-                        indirect, brw_imm_ud(read_size));
-
-               bld.emit(SHADER_OPCODE_MOV_INDIRECT,
-                        subscript(offset(dest, bld, j), BRW_REGISTER_TYPE_UD, 1),
-                        offset(src, bld, j),
-                        indirect_chv_high_32bit, brw_imm_ud(read_size));
+            }
+         } else {
+            const unsigned num_mov_indirects =
+               type_sz(dest.type) / type_sz(BRW_REGISTER_TYPE_UD);
+            /* We read a little bit less per MOV INDIRECT, as they are now
+             * 32-bits ones instead of 64-bit. Fix read_size then.
+             */
+            const unsigned read_size_32bit = read_size -
+                (num_mov_indirects - 1) * type_sz(BRW_REGISTER_TYPE_UD);
+            for (unsigned j = 0; j < instr->num_components; j++) {
+               for (unsigned i = 0; i < num_mov_indirects; i++) {
+                  bld.emit(SHADER_OPCODE_MOV_INDIRECT,
+                           subscript(offset(dest, bld, j), BRW_REGISTER_TYPE_UD, i),
+                           subscript(offset(src, bld, j), BRW_REGISTER_TYPE_UD, i),
+                           indirect, brw_imm_ud(read_size_32bit));
+               }
             }
          }
       }