[Mesa-dev] [PATCH v2 05/10] nir: Implement CSE on intrinsics that can be eliminated and reordered.

Kenneth Graunke kenneth at whitecape.org
Thu Jan 22 03:41:26 PST 2015


Matt and I noticed that one of the shaders hurt by INTEL_USE_NIR=1 had
load_input and load_uniform intrinsics repeated several times, with the
same parameters, but each one generating a distinct SSA value.  This
made ALU operations on those values appear distinct as well.

Generating distinct SSA values is silly - these are read only variables.
CSE'ing them makes everything use a single SSA value, which then allows
other operations to be CSE'd away as well.

Generalizing a bit, it seems like we should be able to safely CSE any
intrinsics that can be eliminated and reordered.  I didn't implement
support for variables for the time being.

v2: Assert that info->num_variables == 0 (requested by Jason).

total NIR instructions in shared programs: 2435936 -> 2023511 (-16.93%)
NIR instructions in affected programs:     2413496 -> 2001071 (-17.09%)
helped:                                    16872

total i965 instructions in shared programs: 6028987 -> 6008427 (-0.34%)
i965 instructions in affected programs:     640654 -> 620094 (-3.21%)
helped:                                     2071
HURT:                                       585
GAINED:                                     14
LOST:                                       25

Signed-off-by: Kenneth Graunke <kenneth at whitecape.org>
---
 src/glsl/nir/nir_opt_cse.c | 40 ++++++++++++++++++++++++++++++++++++++--
 1 file changed, 38 insertions(+), 2 deletions(-)

Here's v2 of CSE for intrinsics.  Sounds like it's good to go.

diff --git a/src/glsl/nir/nir_opt_cse.c b/src/glsl/nir/nir_opt_cse.c
index fef1678..b3e9c0d 100644
--- a/src/glsl/nir/nir_opt_cse.c
+++ b/src/glsl/nir/nir_opt_cse.c
@@ -112,7 +112,34 @@ nir_instrs_equal(nir_instr *instr1, nir_instr *instr2)
 
       return true;
    }
-   case nir_instr_type_intrinsic:
+   case nir_instr_type_intrinsic: {
+      nir_intrinsic_instr *intrinsic1 = nir_instr_as_intrinsic(instr1);
+      nir_intrinsic_instr *intrinsic2 = nir_instr_as_intrinsic(instr2);
+      const nir_intrinsic_info *info =
+         &nir_intrinsic_infos[intrinsic1->intrinsic];
+
+      if (intrinsic1->intrinsic != intrinsic2->intrinsic ||
+          intrinsic1->num_components != intrinsic2->num_components)
+         return false;
+
+      if (info->has_dest && intrinsic1->dest.ssa.num_components !=
+                            intrinsic2->dest.ssa.num_components)
+         return false;
+
+      for (unsigned i = 0; i < info->num_srcs; i++) {
+         if (!nir_srcs_equal(intrinsic1->src[i], intrinsic2->src[i]))
+            return false;
+      }
+
+      assert(info->num_variables == 0);
+
+      for (unsigned i = 0; i < info->num_indices; i++) {
+         if (intrinsic1->const_index[i] != intrinsic2->const_index[i])
+            return false;
+      }
+
+      return true;
+   }
    case nir_instr_type_call:
    case nir_instr_type_jump:
    case nir_instr_type_ssa_undef:
@@ -151,7 +178,13 @@ nir_instr_can_cse(nir_instr *instr)
       return true;
    case nir_instr_type_tex:
       return false; /* TODO */
-   case nir_instr_type_intrinsic:
+   case nir_instr_type_intrinsic: {
+      const nir_intrinsic_info *info =
+         &nir_intrinsic_infos[nir_instr_as_intrinsic(instr)->intrinsic];
+      return (info->flags & NIR_INTRINSIC_CAN_ELIMINATE) &&
+             (info->flags & NIR_INTRINSIC_CAN_REORDER) &&
+             info->num_variables == 0; /* not implemented yet */
+   }
    case nir_instr_type_call:
    case nir_instr_type_jump:
    case nir_instr_type_ssa_undef:
@@ -176,6 +209,9 @@ nir_instr_get_dest_ssa_def(nir_instr *instr)
    case nir_instr_type_phi:
       assert(nir_instr_as_phi(instr)->dest.is_ssa);
       return &nir_instr_as_phi(instr)->dest.ssa;
+   case nir_instr_type_intrinsic:
+      assert(nir_instr_as_intrinsic(instr)->dest.is_ssa);
+      return &nir_instr_as_intrinsic(instr)->dest.ssa;
    default:
       unreachable("We never ask for any of these");
    }
-- 
2.2.2



More information about the mesa-dev mailing list