aco: set vm for pos0 exports on GFX10

RADV's LLVM backend and radeonsi does the same thing.

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Cc: 19.3 <mesa-stable@lists.freedesktop.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3081>

diff --git a/src/amd/compiler/aco_assembler.cpp b/src/amd/compiler/aco_assembler.cpp
index 207c40acf49d56831f463c446f16ba1ee7a65b60..69b8a9b33c6c529977e0ff0eafc4ba1508b4ada2 100644 (file)
--- a/src/amd/compiler/aco_assembler.cpp
+++ b/src/amd/compiler/aco_assembler.cpp
@@ -627,7 +627,7 @@ void fix_exports(asm_context& ctx, std::vector<uint32_t>& out, Program* program)
       exp->enabled_mask = 0;
       exp->compressed = false;
       exp->done = true;
-      exp->valid_mask = program->stage & hw_fs;
+      exp->valid_mask = (program->stage & hw_fs) || program->chip_class >= GFX10;
       if (program->stage & hw_fs)
          exp->dest = 9; /* NULL */
       else

diff --git a/src/amd/compiler/aco_instruction_selection.cpp b/src/amd/compiler/aco_instruction_selection.cpp
index de2132c91db46c2cad835853c0dd0fb4c9f8cab0..d6b7dc696d52f336061fc1587bf38a77060bfbdb 100644 (file)
--- a/src/amd/compiler/aco_instruction_selection.cpp
+++ b/src/amd/compiler/aco_instruction_selection.cpp
@@ -7739,7 +7739,10 @@ static void export_vs_varying(isel_context *ctx, int slot, bool is_pos, int *nex
       else
          exp->operands[i] = Operand(v1);
    }
-   exp->valid_mask = false;
+   /* Navi10-14 skip POS0 exports if EXEC=0 and DONE=0, causing a hang.
+    * Setting valid_mask=1 prevents it and has no other effect.
+    */
+   exp->valid_mask = ctx->options->chip_class >= GFX10 && is_pos && *next_pos == 0;
    exp->done = false;
    exp->compressed = false;
    if (is_pos)
@@ -7779,7 +7782,7 @@ static void export_vs_psiz_layer_viewport(isel_context *ctx, int *next_pos)
          exp->enabled_mask |= 0x4;
       }
    }
-   exp->valid_mask = false;
+   exp->valid_mask = ctx->options->chip_class >= GFX10 && *next_pos == 0;
    exp->done = false;
    exp->compressed = false;
    exp->dest = V_008DFC_SQ_EXP_POS + (*next_pos)++;