radv: copy indirect lowering settings from radeonsi

It looks the original indirect mask was probably copied from
ANV.

Sascha Willems demo results:

tessellation ~4000 -> ~4200 fps

V2: continue lowering local indirects due to llvm deficiencies.

Tested-by: Alex Smith <asmith@feralinteractive.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>

diff --git a/src/amd/vulkan/radv_shader.c b/src/amd/vulkan/radv_shader.c
index a86ba2a01c0bb7650f02f89d592c9e64d54eb5db..3688680107f66ca79478196d1bb2241360a016ab 100644 (file)
--- a/src/amd/vulkan/radv_shader.c
+++ b/src/amd/vulkan/radv_shader.c
@@ -245,8 +245,33 @@ radv_shader_compile_to_nir(struct radv_device *device,
 
        nir_shader_gather_info(nir, entry_point->impl);
 
+       /* While it would be nice not to have this flag, we are constrained
+        * by the reality that LLVM 5.0 doesn't have working VGPR indexing
+        * on GFX9.
+        */
+       bool llvm_has_working_vgpr_indexing =
+               device->physical_device->rad_info.chip_class <= VI;
+
+       /* TODO: Indirect indexing of GS inputs is unimplemented.
+        *
+        * TCS and TES load inputs directly from LDS or offchip memory, so
+        * indirect indexing is trivial.
+        */
        nir_variable_mode indirect_mask = 0;
-       indirect_mask |= nir_var_shader_in;
+       if (nir->stage == MESA_SHADER_GEOMETRY ||
+           (nir->stage != MESA_SHADER_TESS_CTRL &&
+            nir->stage != MESA_SHADER_TESS_EVAL &&
+            !llvm_has_working_vgpr_indexing)) {
+               indirect_mask |= nir_var_shader_in;
+       }
+
+       /* TODO: We shouldn't need to do this, however LLVM isn't currently
+        * smart enough to handle indirects without causing excess spilling
+        * causing the gpu to hang.
+        *
+        * See the following thread for more details of the problem:
+        * https://lists.freedesktop.org/archives/mesa-dev/2017-July/162106.html
+        */
        indirect_mask |= nir_var_local;
 
        nir_lower_indirect_derefs(nir, indirect_mask);