i965/vec4: split DF instructions and later double its execsize in IVB/BYT

We need to split DF instructions in two on IVB/BYT as it needs an
execsize 8 to process 4 DF values (one GRF in total).

v2:
- Rename helper and make it static inline function (Matt).
- Fix indention and add braces (Matt).

v3:
- Don't edit IR instruction when doubling exec_size (Curro)
- Add comment into the code (Curro).
- Manage ARF registers like the others (Curro)

v4:
- Add get_exec_type() function and use it to calculate the execution
  size.

Signed-off-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
[ Francisco Jerez: Fix bogus 'type != BAD_FILE' check.  Take
  destination type as execution type where there is no valid source.
  Assert-fail if the deduced execution type is byte.  Clarify comment
  in get_lowered_simd_width().  Move SIMD width workaround outside of
  'if (...inst->size_written > REG_SIZE)' conditional block, since the
  problem should be independent of whether the amount of data written
  by the instruction is greater or lower than a GRF.  Drop redundant
  is_ivb_df definition.  Drop bogus inst->exec_size < 8 check.
  Simplify channel group assertion. ]
Reviewed-by: Francisco Jerez <currojerez@riseup.net>

3 files changed, 53 insertions, 1 deletions

diff --git a/src/intel/compiler/brw_ir_vec4.h b/src/intel/compiler/brw_ir_vec4.h
index bd026eb2ae..56548c3883 100644
--- a/src/intel/compiler/brw_ir_vec4.h
+++ b/src/intel/compiler/brw_ir_vec4.h
@@ -404,6 +404,39 @@ regs_read(const vec4_instruction *inst, unsigned i)
                        reg_size);
 }
 
+static inline enum brw_reg_type
+get_exec_type(const vec4_instruction *inst)
+{
+   enum brw_reg_type exec_type = BRW_REGISTER_TYPE_B;
+
+   for (int i = 0; i < 3; i++) {
+      if (inst->src[i].file != BAD_FILE) {
+         const brw_reg_type t = get_exec_type(brw_reg_type(inst->src[i].type));
+         if (type_sz(t) > type_sz(exec_type))
+            exec_type = t;
+         else if (type_sz(t) == type_sz(exec_type) &&
+                  brw_reg_type_is_floating_point(t))
+            exec_type = t;
+      }
+   }
+
+   if (exec_type == BRW_REGISTER_TYPE_B)
+      exec_type = inst->dst.type;
+
+   /* TODO: We need to handle half-float conversions. */
+   assert(exec_type != BRW_REGISTER_TYPE_HF ||
+          inst->dst.type == BRW_REGISTER_TYPE_HF);
+   assert(exec_type != BRW_REGISTER_TYPE_B);
+
+   return exec_type;
+}
+
+static inline unsigned
+get_exec_type_size(const vec4_instruction *inst)
+{
+   return type_sz(get_exec_type(inst));
+}
+
 } /* namespace brw */
 
 #endif
diff --git a/src/intel/compiler/brw_vec4.cpp b/src/intel/compiler/brw_vec4.cpp
index d7c0909303..adbd85036e 100644
--- a/src/intel/compiler/brw_vec4.cpp
+++ b/src/intel/compiler/brw_vec4.cpp
@@ -2115,6 +2115,16 @@ get_lowered_simd_width(const struct gen_device_info *devinfo,
       }
    }
 
+   /* IvyBridge can manage a maximum of 4 DFs per SIMD4x2 instruction, since
+    * it doesn't support compression in Align16 mode, no matter if it has
+    * force_writemask_all enabled or disabled (the latter is affected by the
+    * compressed instruction bug in gen7, which is another reason to enforce
+    * this limit).
+    */
+   if (devinfo->gen == 7 && !devinfo->is_haswell &&
+       (get_exec_type_size(inst) == 8 || type_sz(inst->dst.type) == 8))
+      lowered_width = MIN2(lowered_width, 4);
+
    return lowered_width;
 }
 
diff --git a/src/intel/compiler/brw_vec4_generator.cpp b/src/intel/compiler/brw_vec4_generator.cpp
index d3192ab7db..15d6d290fc 100644
--- a/src/intel/compiler/brw_vec4_generator.cpp
+++ b/src/intel/compiler/brw_vec4_generator.cpp
@@ -1522,7 +1522,6 @@ generate_code(struct brw_codegen *p,
       brw_set_default_saturate(p, inst->saturate);
       brw_set_default_mask_control(p, inst->force_writemask_all);
       brw_set_default_acc_write_control(p, inst->writes_accumulator);
-      brw_set_default_exec_size(p, cvt(inst->exec_size) - 1);
 
       assert(inst->group % inst->exec_size == 0);
       assert(inst->group % 8 == 0 ||
@@ -1530,6 +1529,16 @@ generate_code(struct brw_codegen *p,
              inst->src[0].type == BRW_REGISTER_TYPE_DF ||
              inst->src[1].type == BRW_REGISTER_TYPE_DF ||
              inst->src[2].type == BRW_REGISTER_TYPE_DF);
+
+      unsigned exec_size = inst->exec_size;
+      if (devinfo->gen == 7 &&
+          !devinfo->is_haswell &&
+          (get_exec_type_size(inst) == 8 ||
+           inst->dst.type == BRW_REGISTER_TYPE_DF))
+         exec_size *= 2;
+
+      brw_set_default_exec_size(p, cvt(exec_size) - 1);
+
       if (!inst->force_writemask_all)
          brw_set_default_group(p, inst->group);