Mesa (master): i965: Do int64 lowering in NIR

[Jason Ekstrand]

Module: Mesa
Branch: master
Commit: 424ac809bfac60ed8482abfb7ef8c8aa88877b55
URL:    http://cgit.freedesktop.org/mesa/mesa/commit/?id=424ac809bfac60ed8482abfb7ef8c8aa88877b55

Author: Jason Ekstrand <jason.ekstrand at intel.com>
Date:   Thu Feb 23 14:54:13 2017 -0800

i965: Do int64 lowering in NIR

Reviewed-by: Samuel Iglesias Gonsálvez <siglesias at igalia.com>

---

 src/compiler/nir/nir_lower_int64.c     | 107 ++++++++++++++++-----------------
 src/mesa/drivers/dri/i965/brw_link.cpp |   5 --
 src/mesa/drivers/dri/i965/brw_nir.c    |   4 ++
 3 files changed, 57 insertions(+), 59 deletions(-)

diff --git a/src/compiler/nir/nir_lower_int64.c b/src/compiler/nir/nir_lower_int64.c
index 4914091..4de905b 100644
--- a/src/compiler/nir/nir_lower_int64.c
+++ b/src/compiler/nir/nir_lower_int64.c
@@ -25,7 +25,7 @@
 #include "nir_builder.h"
 
 static nir_ssa_def *
-lower_umul64(nir_builder *b, nir_ssa_def *x, nir_ssa_def *y)
+lower_imul64(nir_builder *b, nir_ssa_def *x, nir_ssa_def *y)
 {
    nir_ssa_def *x_lo = nir_unpack_64_2x32_split_x(b, x);
    nir_ssa_def *x_hi = nir_unpack_64_2x32_split_y(b, x);
@@ -187,6 +187,50 @@ lower_imod64(nir_builder *b, nir_ssa_def *n, nir_ssa_def *d)
    return nir_bcsel(b, negate, nir_ineg(b, r), r);
 }
 
+static nir_lower_int64_options
+opcode_to_options_mask(nir_op opcode)
+{
+   switch (opcode) {
+   case nir_op_imul:
+      return nir_lower_imul64;
+   case nir_op_isign:
+      return nir_lower_isign64;
+   case nir_op_udiv:
+   case nir_op_idiv:
+   case nir_op_umod:
+   case nir_op_imod:
+   case nir_op_irem:
+      return nir_lower_divmod64;
+   default:
+      return 0;
+   }
+}
+
+static nir_ssa_def *
+lower_int64_alu_instr(nir_builder *b, nir_alu_instr *alu)
+{
+   nir_ssa_def *src[4];
+   for (unsigned i = 0; i < nir_op_infos[alu->op].num_inputs; i++)
+      src[i] = nir_ssa_for_alu_src(b, alu, i);
+
+   switch (alu->op) {
+   case nir_op_imul:
+      return lower_imul64(b, src[0], src[1]);
+   case nir_op_isign:
+      return lower_isign64(b, src[0]);
+   case nir_op_udiv:
+      return lower_udiv64(b, src[0], src[1]);
+   case nir_op_idiv:
+      return lower_idiv64(b, src[0], src[1]);
+   case nir_op_umod:
+      return lower_umod64(b, src[0], src[1]);
+   case nir_op_imod:
+      return lower_imod64(b, src[0], src[1]);
+   default:
+      unreachable("Invalid ALU opcode to lower");
+   }
+}
+
 static bool
 lower_int64_impl(nir_function_impl *impl, nir_lower_int64_options options)
 {
@@ -204,61 +248,16 @@ lower_int64_impl(nir_function_impl *impl, nir_lower_int64_options options)
          if (alu->dest.dest.ssa.bit_size != 64)
             continue;
 
-         b.cursor = nir_before_instr(instr);
-
-         nir_ssa_def *new_def = NULL;
-         switch (alu->op) {
-         case nir_op_imul:
-            if (options & nir_lower_imul64) {
-               new_def = lower_umul64(&b, nir_ssa_for_alu_src(&b, alu, 0),
-                                          nir_ssa_for_alu_src(&b, alu, 1));
-            }
-            break;
-
-         case nir_op_isign:
-            if (options & nir_lower_isign64)
-               new_def = lower_isign64(&b, nir_ssa_for_alu_src(&b, alu, 0));
-            break;
-
-         case nir_op_udiv:
-            if (options & nir_lower_divmod64) {
-               new_def = lower_udiv64(&b, nir_ssa_for_alu_src(&b, alu, 0),
-                                          nir_ssa_for_alu_src(&b, alu, 1));
-            }
-            break;
-
-         case nir_op_idiv:
-            if (options & nir_lower_divmod64) {
-               new_def = lower_idiv64(&b, nir_ssa_for_alu_src(&b, alu, 0),
-                                          nir_ssa_for_alu_src(&b, alu, 1));
-            }
-            break;
-
-         case nir_op_umod:
-            if (options & nir_lower_divmod64) {
-               new_def = lower_umod64(&b, nir_ssa_for_alu_src(&b, alu, 0),
-                                          nir_ssa_for_alu_src(&b, alu, 1));
-            }
-            break;
-
-         case nir_op_imod:
-            if (options & nir_lower_divmod64) {
-               new_def = lower_imod64(&b, nir_ssa_for_alu_src(&b, alu, 0),
-                                          nir_ssa_for_alu_src(&b, alu, 1));
-            }
-            break;
-
-         default:
-            /* Not lowered */
+         if (!(options & opcode_to_options_mask(alu->op)))
             continue;
-         }
 
-         if (new_def) {
-            nir_ssa_def_rewrite_uses(&alu->dest.dest.ssa,
-                                     nir_src_for_ssa(new_def));
-            nir_instr_remove(&alu->instr);
-            progress = true;
-         }
+         b.cursor = nir_before_instr(instr);
+
+         nir_ssa_def *lowered = lower_int64_alu_instr(&b, alu);
+         nir_ssa_def_rewrite_uses(&alu->dest.dest.ssa,
+                                  nir_src_for_ssa(lowered));
+         nir_instr_remove(&alu->instr);
+         progress = true;
       }
    }
 
diff --git a/src/mesa/drivers/dri/i965/brw_link.cpp b/src/mesa/drivers/dri/i965/brw_link.cpp
index 8611d48..e124f92 100644
--- a/src/mesa/drivers/dri/i965/brw_link.cpp
+++ b/src/mesa/drivers/dri/i965/brw_link.cpp
@@ -118,11 +118,6 @@ process_glsl_ir(struct brw_context *brw,
    }
 
    lower_instructions(shader->ir, instructions_to_lower);
-   lower_64bit_integer_instructions(shader->ir,
-                                    MUL64 |
-                                    DIV64 |
-                                    MOD64 |
-                                    SIGN64);
 
    /* Pre-gen6 HW can only nest if-statements 16 deep.  Beyond this,
     * if-statements need to be flattened.
diff --git a/src/mesa/drivers/dri/i965/brw_nir.c b/src/mesa/drivers/dri/i965/brw_nir.c
index 3cf2c98..f863085 100644
--- a/src/mesa/drivers/dri/i965/brw_nir.c
+++ b/src/mesa/drivers/dri/i965/brw_nir.c
@@ -579,6 +579,10 @@ brw_preprocess_nir(const struct brw_compiler *compiler, nir_shader *nir)
 
    nir_lower_indirect_derefs(nir, indirect_mask);
 
+   nir_lower_int64(nir, nir_lower_imul64 |
+                        nir_lower_isign64 |
+                        nir_lower_divmod64);
+
    /* Get rid of split copies */
    nir = nir_optimize(nir, compiler, is_scalar);