[tint] Polyfill subgroupShuffle id param range (clamped)
Bug: 435246627
Change-Id: I1cd5aba1760881109ecba92500effffe694bab70
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/255322
Commit-Queue: Peter McNeeley <petermcneeley@google.com>
Reviewed-by: James Price <jrprice@google.com>
diff --git a/src/dawn/native/Toggles.cpp b/src/dawn/native/Toggles.cpp
index 8e72dea..52df799 100644
--- a/src/dawn/native/Toggles.cpp
+++ b/src/dawn/native/Toggles.cpp
@@ -575,6 +575,10 @@
{Toggle::ScalarizeMaxMinClamp,
{"scalarize_max_min_clamp", "Scalarize max, min, and clamp builtins.",
"https://crbug.com/422144514", ToggleStage::Device}},
+ {Toggle::SubgroupShuffleClamped,
+ {"subgroup_shuffle_clamped",
+ "Polyfill subgroupShuffle by clamping the id param to within maximum possible subgroup size.",
+ "https://crbug.com/dawn/2502", ToggleStage::Device}},
{Toggle::MetalDisableModuleConstantF16,
{"metal_disable_module_constant_f16",
"Disable module constant hoisting for values that contain f16 types.",
diff --git a/src/dawn/native/Toggles.h b/src/dawn/native/Toggles.h
index 82e81b3..980aead 100644
--- a/src/dawn/native/Toggles.h
+++ b/src/dawn/native/Toggles.h
@@ -140,6 +140,7 @@
ExposeWGSLExperimentalFeatures,
DisablePolyfillsOnIntegerDivisonAndModulo,
ScalarizeMaxMinClamp,
+ SubgroupShuffleClamped,
MetalDisableModuleConstantF16,
EnableImmediateErrorHandling,
VulkanUseStorageInputOutput16,
diff --git a/src/dawn/native/vulkan/PhysicalDeviceVk.cpp b/src/dawn/native/vulkan/PhysicalDeviceVk.cpp
index bfbdfde..d471d6a 100644
--- a/src/dawn/native/vulkan/PhysicalDeviceVk.cpp
+++ b/src/dawn/native/vulkan/PhysicalDeviceVk.cpp
@@ -920,6 +920,12 @@
deviceToggles->Default(Toggle::VulkanDirectVariableAccessTransformHandle, true);
}
+ if (IsIntelMesa()) {
+ // Polyfill a clamp of `id` param in subgroupShuffle to follow spec limitations.
+ // See crbug.com/435246627
+ deviceToggles->Default(Toggle::SubgroupShuffleClamped, true);
+ }
+
if (IsIntelMesa() && gpu_info::IsIntelGen12LP(GetVendorId(), GetDeviceId())) {
// dawn:1688: Intel Mesa driver has a bug about reusing the VkDeviceMemory that was
// previously bound to a 2D VkImage. To work around that bug we have to disable the resource
diff --git a/src/dawn/native/vulkan/ShaderModuleVk.cpp b/src/dawn/native/vulkan/ShaderModuleVk.cpp
index 6d5df11..cb13755 100644
--- a/src/dawn/native/vulkan/ShaderModuleVk.cpp
+++ b/src/dawn/native/vulkan/ShaderModuleVk.cpp
@@ -281,6 +281,8 @@
GetDevice()->IsToggleEnabled(Toggle::DisablePolyfillsOnIntegerDivisonAndModulo);
req.tintOptions.scalarize_max_min_clamp =
GetDevice()->IsToggleEnabled(Toggle::ScalarizeMaxMinClamp);
+ req.tintOptions.subgroup_shuffle_clamped =
+ GetDevice()->IsToggleEnabled(Toggle::SubgroupShuffleClamped);
req.tintOptions.use_vulkan_memory_model =
GetDevice()->IsToggleEnabled(Toggle::UseVulkanMemoryModel);
req.tintOptions.spirv_version = GetDevice()->IsToggleEnabled(Toggle::UseSpirv14)
diff --git a/src/tint/lang/spirv/writer/common/options.h b/src/tint/lang/spirv/writer/common/options.h
index ba8deb9e0..8b4bc8b 100644
--- a/src/tint/lang/spirv/writer/common/options.h
+++ b/src/tint/lang/spirv/writer/common/options.h
@@ -208,6 +208,10 @@
/// `unpack4x8unorm` builtins
bool polyfill_pack_unpack_4x8_norm = false;
+ /// Set to `true` to generate a polyfill clamp of `id` param of subgroupShuffle to within the
+ /// spec max subgroup size.
+ bool subgroup_shuffle_clamped = false;
+
/// Set to `true` to disable the polyfills on integer division and modulo.
bool disable_polyfill_integer_div_mod = false;
@@ -244,6 +248,7 @@
pass_matrix_by_pointer,
polyfill_dot_4x8_packed,
polyfill_pack_unpack_4x8_norm,
+ subgroup_shuffle_clamped,
disable_polyfill_integer_div_mod,
scalarize_max_min_clamp,
use_vulkan_memory_model,
diff --git a/src/tint/lang/spirv/writer/raise/builtin_polyfill.cc b/src/tint/lang/spirv/writer/raise/builtin_polyfill.cc
index 026db12..edfeb8c 100644
--- a/src/tint/lang/spirv/writer/raise/builtin_polyfill.cc
+++ b/src/tint/lang/spirv/writer/raise/builtin_polyfill.cc
@@ -48,6 +48,7 @@
#include "src/tint/lang/spirv/ir/literal_operand.h"
#include "src/tint/lang/spirv/type/sampled_image.h"
#include "src/tint/utils/ice/ice.h"
+#include "src/tint/utils/internal_limits.h"
using namespace tint::core::number_suffixes; // NOLINT
using namespace tint::core::fluent_types; // NOLINT
@@ -280,7 +281,7 @@
SubgroupBroadcast(builtin);
break;
case core::BuiltinFn::kSubgroupShuffle:
- SubgroupShuffle(builtin);
+ SubgroupShuffle(builtin, config.subgroup_shuffle_clamped);
break;
case core::BuiltinFn::kTextureDimensions:
TextureDimensions(builtin);
@@ -1093,7 +1094,7 @@
/// Handle a SubgroupShuffle() builtin.
/// @param builtin the builtin call instruction
- void SubgroupShuffle(core::ir::CoreBuiltinCall* builtin) {
+ void SubgroupShuffle(core::ir::CoreBuiltinCall* builtin, bool clamp_subgroup_shuffle) {
TINT_ASSERT(builtin->Args().Length() == 2);
auto* id = builtin->Args()[1];
@@ -1103,6 +1104,17 @@
cast->InsertBefore(builtin);
builtin->SetArg(1, cast->Result());
}
+
+ /// Polyfill a `subgroupShuffle()` builtin call with one that has clamped the 'id' param
+ if (clamp_subgroup_shuffle) {
+ auto* shuffle_id = builtin->Args()[1];
+ auto* mask_max_subgroup_size =
+ b.Constant(core::u32(tint::internal_limits::kMaxSubgroupSize - 1));
+ b.InsertBefore(builtin, [&] {
+ auto* clamp_via_masking_and = b.And<u32>(shuffle_id, mask_max_subgroup_size);
+ builtin->SetArg(1, clamp_via_masking_and->Result());
+ });
+ }
}
/// Handle a SubgroupBroadcast() builtin.
diff --git a/src/tint/lang/spirv/writer/raise/builtin_polyfill.h b/src/tint/lang/spirv/writer/raise/builtin_polyfill.h
index 98b063d..457e8d0 100644
--- a/src/tint/lang/spirv/writer/raise/builtin_polyfill.h
+++ b/src/tint/lang/spirv/writer/raise/builtin_polyfill.h
@@ -42,6 +42,7 @@
struct PolyfillConfig {
bool use_vulkan_memory_model = false;
SpvVersion version = SpvVersion::kSpv13;
+ bool subgroup_shuffle_clamped = false;
};
/// BuiltinPolyfill is a transform that replaces calls to builtins with polyfills and calls to
diff --git a/src/tint/lang/spirv/writer/raise/builtin_polyfill_test.cc b/src/tint/lang/spirv/writer/raise/builtin_polyfill_test.cc
index 278a22f..77dd296 100644
--- a/src/tint/lang/spirv/writer/raise/builtin_polyfill_test.cc
+++ b/src/tint/lang/spirv/writer/raise/builtin_polyfill_test.cc
@@ -4777,5 +4777,192 @@
EXPECT_EQ(expect, str());
}
+TEST_F(SpirvWriter_BuiltinPolyfillTest, SubgroupShuffleClamped_NoPolyfill) {
+ auto* func = b.Function("foo", ty.void_());
+ auto* arg1 = b.FunctionParam("arg1", ty.i32());
+ auto* arg2 = b.FunctionParam("arg2", ty.i32());
+ func->SetParams({arg1, arg2});
+
+ b.Append(func->Block(), [&] {
+ b.Let("a", b.Call(ty.i32(), core::BuiltinFn::kSubgroupShuffle, arg1, arg2));
+ b.Return(func);
+ });
+ auto* src = R"(
+%foo = func(%arg1:i32, %arg2:i32):void {
+ $B1: {
+ %4:i32 = subgroupShuffle %arg1, %arg2
+ %a:i32 = let %4
+ ret
+ }
+}
+)";
+
+ auto* expect = R"(
+%foo = func(%arg1:i32, %arg2:i32):void {
+ $B1: {
+ %4:u32 = bitcast %arg2
+ %5:i32 = subgroupShuffle %arg1, %4
+ %a:i32 = let %5
+ ret
+ }
+}
+)";
+ EXPECT_EQ(src, str());
+
+ PolyfillConfig config;
+ config.subgroup_shuffle_clamped = false;
+ Run(BuiltinPolyfill, config);
+ EXPECT_EQ(expect, str());
+}
+
+TEST_F(SpirvWriter_BuiltinPolyfillTest, SubgroupShuffleClamped_I32) {
+ auto* func = b.Function("foo", ty.void_());
+ auto* arg1 = b.FunctionParam("arg1", ty.i32());
+ auto* arg2 = b.FunctionParam("arg2", ty.i32());
+ func->SetParams({arg1, arg2});
+ b.Append(func->Block(), [&] {
+ b.Let("a", b.Call(ty.i32(), core::BuiltinFn::kSubgroupShuffle, arg1, arg2));
+ b.Return(func);
+ });
+ auto* src = R"(
+%foo = func(%arg1:i32, %arg2:i32):void {
+ $B1: {
+ %4:i32 = subgroupShuffle %arg1, %arg2
+ %a:i32 = let %4
+ ret
+ }
+}
+)";
+ auto* expect = R"(
+%foo = func(%arg1:i32, %arg2:i32):void {
+ $B1: {
+ %4:u32 = bitcast %arg2
+ %5:u32 = and %4, 127u
+ %6:i32 = subgroupShuffle %arg1, %5
+ %a:i32 = let %6
+ ret
+ }
+}
+)";
+
+ EXPECT_EQ(src, str());
+
+ PolyfillConfig config;
+ config.subgroup_shuffle_clamped = true;
+ Run(BuiltinPolyfill, config);
+ EXPECT_EQ(expect, str());
+}
+
+TEST_F(SpirvWriter_BuiltinPolyfillTest, SubgroupShuffleClamped_U32) {
+ auto* func = b.Function("foo", ty.void_());
+ auto* arg1 = b.FunctionParam("arg1", ty.u32());
+ auto* arg2 = b.FunctionParam("arg2", ty.u32());
+ func->SetParams({arg1, arg2});
+ b.Append(func->Block(), [&] {
+ b.Let("a", b.Call(ty.u32(), core::BuiltinFn::kSubgroupShuffle, arg1, arg2));
+ b.Return(func);
+ });
+ auto* src = R"(
+%foo = func(%arg1:u32, %arg2:u32):void {
+ $B1: {
+ %4:u32 = subgroupShuffle %arg1, %arg2
+ %a:u32 = let %4
+ ret
+ }
+}
+)";
+ auto* expect = R"(
+%foo = func(%arg1:u32, %arg2:u32):void {
+ $B1: {
+ %4:u32 = and %arg2, 127u
+ %5:u32 = subgroupShuffle %arg1, %4
+ %a:u32 = let %5
+ ret
+ }
+}
+)";
+
+ EXPECT_EQ(src, str());
+
+ PolyfillConfig config;
+ config.subgroup_shuffle_clamped = true;
+ Run(BuiltinPolyfill, config);
+ EXPECT_EQ(expect, str());
+}
+
+TEST_F(SpirvWriter_BuiltinPolyfillTest, SubgroupShuffleClamped_F32) {
+ auto* func = b.Function("foo", ty.void_());
+ auto* arg1 = b.FunctionParam("arg1", ty.f32());
+ auto* arg2 = b.FunctionParam("arg2", ty.u32());
+ func->SetParams({arg1, arg2});
+ b.Append(func->Block(), [&] {
+ b.Let("a", b.Call(ty.f32(), core::BuiltinFn::kSubgroupShuffle, arg1, arg2));
+ b.Return(func);
+ });
+ auto* src = R"(
+%foo = func(%arg1:f32, %arg2:u32):void {
+ $B1: {
+ %4:f32 = subgroupShuffle %arg1, %arg2
+ %a:f32 = let %4
+ ret
+ }
+}
+)";
+ auto* expect = R"(
+%foo = func(%arg1:f32, %arg2:u32):void {
+ $B1: {
+ %4:u32 = and %arg2, 127u
+ %5:f32 = subgroupShuffle %arg1, %4
+ %a:f32 = let %5
+ ret
+ }
+}
+)";
+
+ EXPECT_EQ(src, str());
+
+ PolyfillConfig config;
+ config.subgroup_shuffle_clamped = true;
+ Run(BuiltinPolyfill, config);
+ EXPECT_EQ(expect, str());
+}
+
+TEST_F(SpirvWriter_BuiltinPolyfillTest, SubgroupShuffleClamped_Vec2F32) {
+ auto* func = b.Function("foo", ty.void_());
+ auto* arg1 = b.FunctionParam("arg1", ty.vec2<f32>());
+ auto* arg2 = b.FunctionParam("arg2", ty.u32());
+ func->SetParams({arg1, arg2});
+ b.Append(func->Block(), [&] {
+ b.Let("a", b.Call(ty.vec2<f32>(), core::BuiltinFn::kSubgroupShuffle, arg1, arg2));
+ b.Return(func);
+ });
+ auto* src = R"(
+%foo = func(%arg1:vec2<f32>, %arg2:u32):void {
+ $B1: {
+ %4:vec2<f32> = subgroupShuffle %arg1, %arg2
+ %a:vec2<f32> = let %4
+ ret
+ }
+}
+)";
+ auto* expect = R"(
+%foo = func(%arg1:vec2<f32>, %arg2:u32):void {
+ $B1: {
+ %4:u32 = and %arg2, 127u
+ %5:vec2<f32> = subgroupShuffle %arg1, %4
+ %a:vec2<f32> = let %5
+ ret
+ }
+}
+)";
+
+ EXPECT_EQ(src, str());
+
+ PolyfillConfig config;
+ config.subgroup_shuffle_clamped = true;
+ Run(BuiltinPolyfill, config);
+ EXPECT_EQ(expect, str());
+}
+
} // namespace
} // namespace tint::spirv::writer::raise
diff --git a/src/tint/lang/spirv/writer/raise/raise.cc b/src/tint/lang/spirv/writer/raise/raise.cc
index 7600dcb..3fc32f1 100644
--- a/src/tint/lang/spirv/writer/raise/raise.cc
+++ b/src/tint/lang/spirv/writer/raise/raise.cc
@@ -177,7 +177,8 @@
}
raise::PolyfillConfig config = {.use_vulkan_memory_model = options.use_vulkan_memory_model,
- .version = options.spirv_version};
+ .version = options.spirv_version,
+ .subgroup_shuffle_clamped = options.subgroup_shuffle_clamped};
RUN_TRANSFORM(raise::BuiltinPolyfill, module, config);
RUN_TRANSFORM(raise::ExpandImplicitSplats, module);
