Mesa (master): i965/fs: Fix code emission for imul_high in NIR.

[Matt Turner]

Module: Mesa
Branch: master
Commit: 0c06d019bcf626b289ae94ca791dc25c216c1e5c
URL:    http://cgit.freedesktop.org/mesa/mesa/commit/?id=0c06d019bcf626b289ae94ca791dc25c216c1e5c

Author: Matt Turner <mattst88 at gmail.com>
Date:   Fri Apr 24 11:28:05 2015 -0700

i965/fs: Fix code emission for imul_high in NIR.

Copy over from brw_fs_visitor.cpp.

Reviewed-by: Kenneth Graunke <kenneth at whitecape.org>

---

 src/mesa/drivers/dri/i965/brw_fs_nir.cpp |   24 +++++++++++++++++++++++-
 1 file changed, 23 insertions(+), 1 deletion(-)

diff --git a/src/mesa/drivers/dri/i965/brw_fs_nir.cpp b/src/mesa/drivers/dri/i965/brw_fs_nir.cpp
index 9564764..523e56d 100644
--- a/src/mesa/drivers/dri/i965/brw_fs_nir.cpp
+++ b/src/mesa/drivers/dri/i965/brw_fs_nir.cpp
@@ -827,8 +827,30 @@ fs_visitor::nir_emit_alu(nir_alu_instr *instr)
 
       struct brw_reg acc = retype(brw_acc_reg(dispatch_width), result.type);
 
-      emit(MUL(acc, op[0], op[1]));
+      fs_inst *mul = emit(MUL(acc, op[0], op[1]));
       emit(MACH(result, op[0], op[1]));
+
+      /* Until Gen8, integer multiplies read 32-bits from one source, and
+       * 16-bits from the other, and relying on the MACH instruction to
+       * generate the high bits of the result.
+       *
+       * On Gen8, the multiply instruction does a full 32x32-bit multiply,
+       * but in order to do a 64x64-bit multiply we have to simulate the
+       * previous behavior and then use a MACH instruction.
+       *
+       * FINISHME: Don't use source modifiers on src1.
+       */
+      if (devinfo->gen >= 8) {
+         assert(mul->src[1].type == BRW_REGISTER_TYPE_D ||
+                mul->src[1].type == BRW_REGISTER_TYPE_UD);
+         if (mul->src[1].type == BRW_REGISTER_TYPE_D) {
+            mul->src[1].type = BRW_REGISTER_TYPE_W;
+            mul->src[1].stride = 2;
+         } else {
+            mul->src[1].type = BRW_REGISTER_TYPE_UW;
+            mul->src[1].stride = 2;
+         }
+      }
       break;
    }