From: Matt Turner <mattst88@gmail.com>
Date: Sat, 16 Nov 2013 21:16:50 +0000 (-0800)
Subject: i965/fs: Don't emit SIMD16 BFI instructions.
X-Git-Url: https://git.libre-soc.org/?a=commitdiff_plain;h=c4464c9eea65a287aa74f6bb59cd00b1ef665a36;p=mesa.git

i965/fs: Don't emit SIMD16 BFI instructions.

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

diff --git a/src/mesa/drivers/dri/i965/brw_fs_generator.cpp b/src/mesa/drivers/dri/i965/brw_fs_generator.cpp
index a97a016b8b3..6626a8cdf16 100644
--- a/src/mesa/drivers/dri/i965/brw_fs_generator.cpp
+++ b/src/mesa/drivers/dri/i965/brw_fs_generator.cpp
@@ -1530,11 +1530,32 @@ fs_generator::generate_code(exec_list *instructions)
 
       case BRW_OPCODE_BFI1:
          assert(brw->gen >= 7);
-         brw_BFI1(p, dst, src[0], src[1]);
+         /* The Haswell WaForceSIMD8ForBFIInstruction workaround says that we
+          * should
+          *
+          *    "Force BFI instructions to be executed always in SIMD8."
+          */
+         if (dispatch_width == 16 && brw->is_haswell) {
+            brw_set_compression_control(p, BRW_COMPRESSION_NONE);
+            brw_BFI1(p, dst, src[0], src[1]);
+            brw_set_compression_control(p, BRW_COMPRESSION_2NDHALF);
+            brw_BFI1(p, sechalf(dst), sechalf(src[0]), sechalf(src[1]));
+            brw_set_compression_control(p, BRW_COMPRESSION_COMPRESSED);
+         } else {
+            brw_BFI1(p, dst, src[0], src[1]);
+         }
          break;
       case BRW_OPCODE_BFI2:
          assert(brw->gen >= 7);
          brw_set_access_mode(p, BRW_ALIGN_16);
+         /* The Haswell WaForceSIMD8ForBFIInstruction workaround says that we
+          * should
+          *
+          *    "Force BFI instructions to be executed always in SIMD8."
+          *
+          * Otherwise we would be able to emit compressed instructions like we
+          * do for the other three-source instructions.
+          */
          if (dispatch_width == 16) {
             brw_set_compression_control(p, BRW_COMPRESSION_NONE);
             brw_BFI2(p, dst, src[0], src[1], src[2]);