Mesa (master): aco: do not use the vec3 variant for loads on GFX6

[GitLab Mirror]

Module: Mesa
Branch: master
Commit: fe9157a700341f38b448fa9279fb60d7e4a05b32
URL:    http://cgit.freedesktop.org/mesa/mesa/commit/?id=fe9157a700341f38b448fa9279fb60d7e4a05b32

Author: Samuel Pitoiset <samuel.pitoiset at gmail.com>
Date:   Thu Jan 16 14:44:02 2020 +0100

aco: do not use the vec3 variant for loads on GFX6

GFX6 only supports vec3 with load/store format.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset at gmail.com>
Reviewed-By: Timur Kristóf <timur.kristof at gmail.com>
Reviewed-by: Daniel Schürmann <daniel at schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3432>

---

 src/amd/compiler/aco_instruction_selection.cpp | 14 ++++++++++++++
 1 file changed, 14 insertions(+)

diff --git a/src/amd/compiler/aco_instruction_selection.cpp b/src/amd/compiler/aco_instruction_selection.cpp
index 250f7011b04..94800100c00 100644
--- a/src/amd/compiler/aco_instruction_selection.cpp
+++ b/src/amd/compiler/aco_instruction_selection.cpp
@@ -3408,6 +3408,9 @@ void load_buffer(isel_context *ctx, unsigned num_components, Temp dst,
          emit_split_vector(ctx, lower, 2);
          num_bytes -= 16;
          const_offset = 16;
+      } else if (num_bytes == 12 && ctx->options->chip_class == GFX6) {
+         /* GFX6 doesn't support loading vec3, expand to vec4. */
+         num_bytes = 16;
       }
 
       switch (num_bytes) {
@@ -3418,6 +3421,7 @@ void load_buffer(isel_context *ctx, unsigned num_components, Temp dst,
             op = aco_opcode::buffer_load_dwordx2;
             break;
          case 12:
+            assert(ctx->options->chip_class > GFX6);
             op = aco_opcode::buffer_load_dwordx3;
             break;
          case 16:
@@ -3451,6 +3455,16 @@ void load_buffer(isel_context *ctx, unsigned num_components, Temp dst,
          instr->operands[2] = Operand(emit_extract_vector(ctx, upper, 0, v2));
          if (dst.size() == 8)
             instr->operands[3] = Operand(emit_extract_vector(ctx, upper, 1, v2));
+      } else if (dst.size() == 3 && ctx->options->chip_class == GFX6) {
+         Temp vec = bld.tmp(v4);
+         instr->definitions[0] = Definition(vec);
+         bld.insert(std::move(instr));
+         emit_split_vector(ctx, vec, 4);
+
+         instr.reset(create_instruction<Pseudo_instruction>(aco_opcode::p_create_vector, Format::PSEUDO, 3, 1));
+         instr->operands[0] = Operand(emit_extract_vector(ctx, vec, 0, v1));
+         instr->operands[1] = Operand(emit_extract_vector(ctx, vec, 1, v1));
+         instr->operands[2] = Operand(emit_extract_vector(ctx, vec, 2, v1));
       }
 
       if (dst.type() == RegType::sgpr) {