radeonsi: enable ACO by default

NIR+ACO is the best SSA-based shader compiler for AMD GPUs that exists.

There are many reasons why NIR+ACO is better than LLVM, and I have a long
list that I've collected over the years, but the major ones are better GPU
performance (faster GPU memory access thanks to better clauses and
scheduling, a lot less SGPR/VGPR spilling, better loop support, slightly
smaller shader binaries), 8x lower shader compile times, and smaller memory
footprint of the IR.

It also shows that NIR is a mature SSA-based shader compiler that helps
drivers generate optimized code very quickly.

And most importantly, radeonsi has slightly better Viewperf performance
with NIR+ACO than LLVM, and that's difficult to ignore.

Reviewed-by: Qiang Yu <yuq825@gmail.com>
Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/38070>

4 files changed, 4 insertions, 26 deletions

diff --git a/docs/envvars.rst b/docs/envvars.rst
index b94c23f4517..377d667f0a2 100644
--- a/docs/envvars.rst
+++ b/docs/envvars.rst
@@ -1784,8 +1784,8 @@ RadeonSI driver environment variables
       Use old-style monolithic shaders compiled on demand
    ``nooptvariant``
       Disable compiling optimized shader variants.
-   ``useaco``
-      Use ACO as shader compiler when possible
+   ``usellvm``
+      Use LLVM as shader compiler when possible
    ``nowc``
       Disable GTT write combining
    ``check_vm``
diff --git a/src/gallium/drivers/radeonsi/ci/gfx11-navi31-fail.csv b/src/gallium/drivers/radeonsi/ci/gfx11-navi31-fail.csv
index e03bd2bdf37..114963fa289 100644
--- a/src/gallium/drivers/radeonsi/ci/gfx11-navi31-fail.csv
+++ b/src/gallium/drivers/radeonsi/ci/gfx11-navi31-fail.csv
@@ -27,8 +27,6 @@ spec@!opengl 1.0@rasterpos@glsl_vs_tes_linked,Fail
 # This is caused by lowering mediump before linking:
 spec@glsl-es-1.00@linker@glsl-mismatched-uniform-precision-unused,Fail
 
-KHR-GL46.shaders.uniform_block.random.nested_structs_instance_arrays.0,Fail
-
 # See Khronos issue 5587: the test expects one-dimensional (array) texture to work while
 # it's explicitely marked as non-supported by EXT_sparse_texture2.
 KHR-GL46.sparse_texture2_tests.StandardPageSizesTestCase_texture_1d_array_r11f_g11f_b10f,Fail
@@ -427,12 +425,3 @@ KHR-GL46.transform_feedback_overflow_query_ARB.multiple-streams-one-buffer-per-s
 KHR-GL46.texture_query_lod.sampler1D_test,Fail
 KHR-GL46.texture_query_lod.sampler2D_test,Fail
 KHR-GL46.texture_query_lod.sampler3D_test,Fail
-
-# escts failures (pass with ACO)
-KHR-GLES3.shaders.uniform_block.random.nested_structs_instance_arrays.0,Fail
-KHR-GLES31.core.shader_image_load_store.basic-allFormats-store-fs,Fail
-
-# deqp failures (pass with ACO)
-dEQP-GLES3.functional.ubo.random.all_per_block_buffers.29,Fail
-dEQP-GLES3.functional.ubo.random.basic_instance_arrays.22,Fail
-dEQP-GLES3.functional.ubo.random.nested_structs.4,Fail
diff --git a/src/gallium/drivers/radeonsi/si_pipe.c b/src/gallium/drivers/radeonsi/si_pipe.c
index 8b9bd3e17a8..97f8ed4ef2e 100644
--- a/src/gallium/drivers/radeonsi/si_pipe.c
+++ b/src/gallium/drivers/radeonsi/si_pipe.c
@@ -121,7 +121,6 @@ static const struct debug_named_value radeonsi_shader_debug_options[] = {
    {"checkir", DBG(CHECK_IR), "Enable additional sanity checks on shader IR"},
    {"mono", DBG(MONOLITHIC_SHADERS), "Use old-style monolithic shaders compiled on demand"},
    {"nooptvariant", DBG(NO_OPT_VARIANT), "Disable compiling optimized shader variants."},
-   {"useaco", DBG(USE_ACO), "Use ACO as shader compiler when possible"},
    {"usellvm", DBG(USE_LLVM), "Use LLVM as shader compiler when possible"},
 
    DEBUG_NAMED_VALUE_END /* must be last */
@@ -1327,17 +1326,8 @@ static struct pipe_screen *radeonsi_screen_create_impl(struct radeon_winsys *ws,
    bool support_aco = aco_is_gpu_supported(&sscreen->info);
 
 #if AMD_LLVM_AVAILABLE
-   /* For GFX11.5, LLVM < 19 is missing a workaround that can cause GPU hangs. ACO is the only
-    * alternative that has the workaround and is always available. Same for GFX12.
-    */
-   if ((sscreen->info.gfx_level == GFX12 && LLVM_VERSION_MAJOR < 20) ||
-       (sscreen->info.gfx_level == GFX11_5 && LLVM_VERSION_MAJOR < 19))
-      sscreen->use_aco = true;
-   else if (sscreen->info.gfx_level >= GFX10)
-      sscreen->use_aco = (sscreen->shader_debug_flags & DBG(USE_ACO));
-   else
-      sscreen->use_aco = support_aco && sscreen->info.has_image_opcodes &&
-                         !(sscreen->shader_debug_flags & DBG(USE_LLVM));
+   sscreen->use_aco = support_aco && sscreen->info.has_image_opcodes &&
+                      !(sscreen->shader_debug_flags & DBG(USE_LLVM));
 #else
    sscreen->use_aco = true;
 #endif
diff --git a/src/gallium/drivers/radeonsi/si_pipe.h b/src/gallium/drivers/radeonsi/si_pipe.h
index c3aab10c457..84e44aada42 100644
--- a/src/gallium/drivers/radeonsi/si_pipe.h
+++ b/src/gallium/drivers/radeonsi/si_pipe.h
@@ -257,7 +257,6 @@ enum
    DBG_MONOLITHIC_SHADERS,
    DBG_NO_OPT_VARIANT,
 
-   DBG_USE_ACO,
    DBG_USE_LLVM,
 };