[Mesa-dev] [PATCH 4/7] i965/fs: Return a fs_reg from shuffle_64bit_data_for_32bit_write

[Jason Ekstrand]

All callers of this function allocate a fs_reg expressly to pass into
it.  It's much easier if we just let the helper allocate the register.
While we're here, we switch it to doing the MOVs with an integer type so
that we don't accidentally canonicalize floats on half of a double.
---
 src/intel/compiler/brw_fs.h       |  7 +++----
 src/intel/compiler/brw_fs_nir.cpp | 34 +++++++++-------------------------
 2 files changed, 12 insertions(+), 29 deletions(-)

diff --git a/src/intel/compiler/brw_fs.h b/src/intel/compiler/brw_fs.h
index f1ba193..e281242 100644
--- a/src/intel/compiler/brw_fs.h
+++ b/src/intel/compiler/brw_fs.h
@@ -493,10 +493,9 @@ void shuffle_32bit_load_result_to_64bit_data(const brw::fs_builder &bld,
                                              const fs_reg &src,
                                              uint32_t components);
 
-void shuffle_64bit_data_for_32bit_write(const brw::fs_builder &bld,
-                                        const fs_reg &dst,
-                                        const fs_reg &src,
-                                        uint32_t components);
+fs_reg shuffle_64bit_data_for_32bit_write(const brw::fs_builder &bld,
+                                          const fs_reg &src,
+                                          uint32_t components);
 fs_reg setup_imm_df(const brw::fs_builder &bld,
                     double v);
 
diff --git a/src/intel/compiler/brw_fs_nir.cpp b/src/intel/compiler/brw_fs_nir.cpp
index 4080495..c01c201 100644
--- a/src/intel/compiler/brw_fs_nir.cpp
+++ b/src/intel/compiler/brw_fs_nir.cpp
@@ -2641,10 +2641,8 @@ fs_visitor::nir_emit_tcs_intrinsic(const fs_builder &bld,
                 * expected by our 32-bit URB write messages. We use a temporary
                 * for that.
                 */
-               fs_reg dest = fs_reg(VGRF, alloc.allocate(2), value.type);
                unsigned channel = iter * 2 + i;
-               shuffle_64bit_data_for_32bit_write(bld,
-                  retype(dest, BRW_REGISTER_TYPE_F),
+               fs_reg dest = shuffle_64bit_data_for_32bit_write(bld,
                   retype(offset(value, bld, 2 * channel), BRW_REGISTER_TYPE_DF),
                   1);
 
@@ -3496,14 +3494,9 @@ fs_visitor::nir_emit_cs_intrinsic(const fs_builder &bld,
       unsigned type_size = 4;
       if (nir_src_bit_size(instr->src[0]) == 64) {
          type_size = 8;
-         fs_reg tmp =
-           fs_reg(VGRF, alloc.allocate(alloc.sizes[val_reg.nr]), val_reg.type);
-         shuffle_64bit_data_for_32bit_write(
-            bld,
-            retype(tmp, BRW_REGISTER_TYPE_F),
+         val_reg = shuffle_64bit_data_for_32bit_write(bld,
             retype(val_reg, BRW_REGISTER_TYPE_DF),
             instr->num_components);
-         val_reg = tmp;
       }
 
       unsigned type_slots = type_size / 4;
@@ -4001,13 +3994,9 @@ fs_visitor::nir_emit_intrinsic(const fs_builder &bld, nir_intrinsic_instr *instr
       unsigned type_size = 4;
       if (nir_src_bit_size(instr->src[0]) == 64) {
          type_size = 8;
-         fs_reg tmp =
-           fs_reg(VGRF, alloc.allocate(alloc.sizes[val_reg.nr]), val_reg.type);
-         shuffle_64bit_data_for_32bit_write(bld,
-            retype(tmp, BRW_REGISTER_TYPE_F),
+         val_reg = shuffle_64bit_data_for_32bit_write(bld,
             retype(val_reg, BRW_REGISTER_TYPE_DF),
             instr->num_components);
-         val_reg = tmp;
       }
 
       unsigned type_slots = type_size / 4;
@@ -4065,11 +4054,8 @@ fs_visitor::nir_emit_intrinsic(const fs_builder &bld, nir_intrinsic_instr *instr
       unsigned num_components = instr->num_components;
       unsigned first_component = nir_intrinsic_component(instr);
       if (nir_src_bit_size(instr->src[0]) == 64) {
-         fs_reg tmp =
-            fs_reg(VGRF, alloc.allocate(2 * num_components),
-                   BRW_REGISTER_TYPE_F);
-         shuffle_64bit_data_for_32bit_write(
-            bld, tmp, retype(src, BRW_REGISTER_TYPE_DF), num_components);
+         fs_reg tmp = shuffle_64bit_data_for_32bit_write(bld,
+            retype(src, BRW_REGISTER_TYPE_DF), num_components);
          src = retype(tmp, src.type);
          num_components *= 2;
       }
@@ -4711,24 +4697,22 @@ shuffle_32bit_load_result_to_64bit_data(const fs_builder &bld,
  * 64-bit data they are about to write. Because of this the function checks
  * that the src and dst regions involved in the operation do not overlap.
  */
-void
+fs_reg
 shuffle_64bit_data_for_32bit_write(const fs_builder &bld,
-                                   const fs_reg &dst,
                                    const fs_reg &src,
                                    uint32_t components)
 {
    assert(type_sz(src.type) == 8);
-   assert(type_sz(dst.type) == 4);
 
-   assert(!regions_overlap(
-             dst, 2 * components * dst.component_size(bld.dispatch_width()),
-             src, components * src.component_size(bld.dispatch_width())));
+   fs_reg dst = bld.vgrf(BRW_REGISTER_TYPE_D, 2 * components);
 
    for (unsigned i = 0; i < components; i++) {
       const fs_reg component_i = offset(src, bld, i);
       bld.MOV(offset(dst, bld, 2 * i), subscript(component_i, dst.type, 0));
       bld.MOV(offset(dst, bld, 2 * i + 1), subscript(component_i, dst.type, 1));
    }
+
+   return dst;
 }
 
 fs_reg
-- 
2.5.0.400.gff86faf