<div dir="ltr">I did some benchmarking of Assassin's Creed Odyssey today on ANV. Disabling the SSBO optimization possibilities in ANV that moving to derefs unlocks (I just moved the lowering super-early; moving back to index/offset would be insane) drops the perf of ACO by 20%. You want SSBO derefs. :D<br></div><br><div class="gmail_quote"><div dir="ltr" class="gmail_attr">On Thu, Mar 28, 2019 at 9:08 AM Jason Ekstrand <<a href="mailto:jason@jlekstrand.net">jason@jlekstrand.net</a>> wrote:<br></div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"><div dir="ltr"><div class="gmail_quote"><div dir="ltr" class="gmail_attr">On Wed, Mar 27, 2019 at 4:13 AM Samuel Pitoiset <<a href="mailto:samuel.pitoiset@gmail.com" target="_blank">samuel.pitoiset@gmail.com</a>> wrote:<br></div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex">This helps few compute shaders, mostly for F12017.<br>
<br>
27670 shaders in 14347 tests<br>
Totals:<br>
SGPRS: 1231173 -> 1231173 (0.00 %)<br>
VGPRS: 866056 -> 865928 (-0.01 %)<br>
Spilled SGPRs: 24201 -> 24201 (0.00 %)<br>
Code Size: 46137040 -> 46144868 (0.02 %) bytes<br>
Max Waves: 232287 -> 232302 (0.01 %)<br>
<br>
Totals from affected shaders:<br>
SGPRS: 24624 -> 24624 (0.00 %)<br>
VGPRS: 25960 -> 25832 (-0.49 %)<br>
Spilled SGPRs: 0 -> 0 (0.00 %)<br>
Code Size: 2922632 -> 2930460 (0.27 %) bytes<br>
Max Waves: 1216 -> 1231 (1.23 %)<br>
<br>
Suggested-by: <Jason Ekstrand <a href="mailto:jason@jlekstrand.net" target="_blank">jason@jlekstrand.net</a>><br>
Signed-off-by: Samuel Pitoiset <<a href="mailto:samuel.pitoiset@gmail.com" target="_blank">samuel.pitoiset@gmail.com</a>><br>
---<br>
src/amd/common/ac_nir_to_llvm.c | 8 ++++++++<br>
src/amd/vulkan/radv_shader.c | 5 ++++-<br>
2 files changed, 12 insertions(+), 1 deletion(-)<br>
<br>
diff --git a/src/amd/common/ac_nir_to_llvm.c b/src/amd/common/ac_nir_to_llvm.c<br>
index b25cc6a0a84..c46d98e6dd9 100644<br>
--- a/src/amd/common/ac_nir_to_llvm.c<br>
+++ b/src/amd/common/ac_nir_to_llvm.c<br>
@@ -3275,6 +3275,14 @@ static void visit_intrinsic(struct ac_nir_context *ctx,<br>
case nir_intrinsic_vulkan_resource_reindex:<br>
result = visit_vulkan_resource_reindex(ctx, instr);<br>
break;<br>
+ case nir_intrinsic_load_vulkan_descriptor: {<br>
+ LLVMValueRef values[2] = {<br>
+ get_src(ctx, instr->src[0]),<br>
+ ctx->ac.i32_0,<br>
+ };<br>
+ result = ac_build_gather_values(&ctx->ac, values, 2);<br>
+ break;<br>
+ }<br>
case nir_intrinsic_store_ssbo:<br>
visit_store_ssbo(ctx, instr);<br>
break;<br>
diff --git a/src/amd/vulkan/radv_shader.c b/src/amd/vulkan/radv_shader.c<br>
index 19a807df199..2751302e8b9 100644<br>
--- a/src/amd/vulkan/radv_shader.c<br>
+++ b/src/amd/vulkan/radv_shader.c<br>
@@ -220,7 +220,6 @@ radv_shader_compile_to_nir(struct radv_device *device,<br>
}<br>
}<br>
const struct spirv_to_nir_options spirv_options = {<br>
- .lower_ubo_ssbo_access_to_offsets = true,<br>
.caps = {<br>
.descriptor_array_dynamic_indexing = true,<br>
.device_group = true,<br>
@@ -306,6 +305,10 @@ radv_shader_compile_to_nir(struct radv_device *device,<br>
NIR_PASS_V(nir, nir_lower_system_values);<br>
NIR_PASS_V(nir, nir_lower_clip_cull_distance_arrays);<br>
NIR_PASS_V(nir, nir_lower_frexp);<br>
+<br>
+ NIR_PASS_V(nir, nir_lower_explicit_io,<br>
+ nir_var_mem_ubo | nir_var_mem_ssbo,<br>
+ nir_address_format_32bit_index_offset);<br></blockquote><div><br></div><div>If you actually want to get SSBO access optimization, you need to call this *after* your first call to your main optimization loop. Otherwise, the change is basically just a no-op which just shuffles around the way address calculations are done a bit.</div><div><br></div><div>--Jason<br></div><div> </div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex">
}<br>
<br>
/* Vulkan uses the separate-shader linking model */<br>
-- <br>
2.21.0<br>
<br>
_______________________________________________<br>
mesa-dev mailing list<br>
<a href="mailto:mesa-dev@lists.freedesktop.org" target="_blank">mesa-dev@lists.freedesktop.org</a><br>
<a href="https://lists.freedesktop.org/mailman/listinfo/mesa-dev" rel="noreferrer" target="_blank">https://lists.freedesktop.org/mailman/listinfo/mesa-dev</a></blockquote></div></div>
</blockquote></div>