[SPIRV] Add FloatControl2 capability #144371
Open
+55
−6
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Add handling for FPFastMathMode in SPIR-V shaders. This is a first pass that simply does a direct translation when the proper extension is available. This will unblock work for HLSL. However, it is not a full solution. The default math mode for spir-v is determined by the API. When targeting Vulkan many of the fast math options are assumed. We should do something particular when targeting Vulkan. We will also need to handle the hlsl "precise" keyword correctly when FPFastMathMode is not available. Unblockes llvm#140739, but we are keeing it open to track the remaining issues mentioned above.
|
@llvm/pr-subscribers-backend-spir-v Author: Steven Perron (s-perron) ChangesAdd handling for FPFastMathMode in SPIR-V shaders. This is a first pass that The default math mode for spir-v is determined by the API. When We will also need to handle the hlsl "precise" keyword correctly when Unblockes #140739, but we are Full diff: https://github.com/llvm/llvm-project/pull/144371.diff 5 Files Affected:
diff --git a/llvm/docs/SPIRVUsage.rst b/llvm/docs/SPIRVUsage.rst
index 1858bda6160d4..d96393e111717 100644
--- a/llvm/docs/SPIRVUsage.rst
+++ b/llvm/docs/SPIRVUsage.rst
@@ -217,6 +217,8 @@ list of supported SPIR-V extensions, sorted alphabetically by their extension na
- Adds an instruction to compute the matrix product of an M x K matrix with a K x N matrix and then add an M x N matrix.
* - ``SPV_INTEL_int4``
- Adds support for 4-bit integer type, and allow this type to be used in cooperative matrices.
+ * - ``SPV_KHR_float_controls2``
+ - Adds ability to specify the floating-point environment in shaders. It can be used on whole modules and individual instructions.
To enable multiple extensions, list them separated by comma. For example, to enable support for atomic operations on floating-point numbers and arbitrary precision integers, use:
diff --git a/llvm/lib/Target/SPIRV/SPIRVCommandLine.cpp b/llvm/lib/Target/SPIRV/SPIRVCommandLine.cpp
index fbaca4e0e4d81..1d19bc4617c7d 100644
--- a/llvm/lib/Target/SPIRV/SPIRVCommandLine.cpp
+++ b/llvm/lib/Target/SPIRV/SPIRVCommandLine.cpp
@@ -100,7 +100,9 @@ static const std::map<std::string, SPIRV::Extension::Extension, std::less<>>
SPIRV::Extension::Extension::SPV_INTEL_ternary_bitwise_function},
{"SPV_INTEL_2d_block_io",
SPIRV::Extension::Extension::SPV_INTEL_2d_block_io},
- {"SPV_INTEL_int4", SPIRV::Extension::Extension::SPV_INTEL_int4}};
+ {"SPV_INTEL_int4", SPIRV::Extension::Extension::SPV_INTEL_int4},
+ {"SPV_KHR_float_controls2",
+ SPIRV::Extension::Extension::SPV_KHR_float_controls2}};
bool SPIRVExtensionsParser::parse(cl::Option &O, StringRef ArgName,
StringRef ArgValue,
diff --git a/llvm/lib/Target/SPIRV/SPIRVModuleAnalysis.cpp b/llvm/lib/Target/SPIRV/SPIRVModuleAnalysis.cpp
index b71a9dd68dd44..ad976e5288927 100644
--- a/llvm/lib/Target/SPIRV/SPIRVModuleAnalysis.cpp
+++ b/llvm/lib/Target/SPIRV/SPIRVModuleAnalysis.cpp
@@ -70,6 +70,8 @@ getSymbolicOperandRequirements(SPIRV::OperandCategory::OperandCategory Category,
AvoidCapabilitiesSet AvoidCaps;
if (!ST.isShader())
AvoidCaps.S.insert(SPIRV::Capability::Shader);
+ else
+ AvoidCaps.S.insert(SPIRV::Capability::Kernel);
VersionTuple ReqMinVer = getSymbolicOperandMinVersion(Category, i);
VersionTuple ReqMaxVer = getSymbolicOperandMaxVersion(Category, i);
@@ -88,8 +90,11 @@ getSymbolicOperandRequirements(SPIRV::OperandCategory::OperandCategory Category,
} else if (MinVerOK && MaxVerOK) {
if (ReqCaps.size() == 1) {
auto Cap = ReqCaps[0];
- if (Reqs.isCapabilityAvailable(Cap))
+ if (Reqs.isCapabilityAvailable(Cap)) {
+ ReqExts.append(getSymbolicOperandExtensions(
+ SPIRV::OperandCategory::CapabilityOperand, Cap));
return {true, {Cap}, ReqExts, ReqMinVer, ReqMaxVer};
+ }
} else {
// By SPIR-V specification: "If an instruction, enumerant, or other
// feature specifies multiple enabling capabilities, only one such
@@ -103,8 +108,11 @@ getSymbolicOperandRequirements(SPIRV::OperandCategory::OperandCategory Category,
UseCaps.push_back(Cap);
for (size_t i = 0, Sz = UseCaps.size(); i < Sz; ++i) {
auto Cap = UseCaps[i];
- if (i == Sz - 1 || !AvoidCaps.S.contains(Cap))
+ if (i == Sz - 1 || !AvoidCaps.S.contains(Cap)) {
+ ReqExts.append(getSymbolicOperandExtensions(
+ SPIRV::OperandCategory::CapabilityOperand, Cap));
return {true, {Cap}, ReqExts, ReqMinVer, ReqMaxVer};
+ }
}
}
}
@@ -1975,6 +1983,14 @@ static unsigned getFastMathFlags(const MachineInstr &I) {
return Flags;
}
+static bool isFastMathMathModeAvailable(const SPIRVSubtarget &ST) {
+ if (ST.isKernel())
+ return true;
+ if (ST.getSPIRVVersion() < VersionTuple(1, 2))
+ return false;
+ return ST.canUseExtension(SPIRV::Extension::SPV_KHR_float_controls2);
+}
+
static void handleMIFlagDecoration(MachineInstr &I, const SPIRVSubtarget &ST,
const SPIRVInstrInfo &TII,
SPIRV::RequirementHandler &Reqs) {
@@ -1998,8 +2014,12 @@ static void handleMIFlagDecoration(MachineInstr &I, const SPIRVSubtarget &ST,
unsigned FMFlags = getFastMathFlags(I);
if (FMFlags == SPIRV::FPFastMathMode::None)
return;
- Register DstReg = I.getOperand(0).getReg();
- buildOpDecorate(DstReg, I, TII, SPIRV::Decoration::FPFastMathMode, {FMFlags});
+
+ if (isFastMathMathModeAvailable(ST)) {
+ Register DstReg = I.getOperand(0).getReg();
+ buildOpDecorate(DstReg, I, TII, SPIRV::Decoration::FPFastMathMode,
+ {FMFlags});
+ }
}
// Walk all functions and add decorations related to MI flags.
diff --git a/llvm/lib/Target/SPIRV/SPIRVSymbolicOperands.td b/llvm/lib/Target/SPIRV/SPIRVSymbolicOperands.td
index f1aae42ea2be0..548e9b717c161 100644
--- a/llvm/lib/Target/SPIRV/SPIRVSymbolicOperands.td
+++ b/llvm/lib/Target/SPIRV/SPIRVSymbolicOperands.td
@@ -319,6 +319,7 @@ defm SPV_INTEL_ternary_bitwise_function : ExtensionOperand<120>;
defm SPV_INTEL_subgroup_matrix_multiply_accumulate : ExtensionOperand<121>;
defm SPV_INTEL_2d_block_io : ExtensionOperand<122>;
defm SPV_INTEL_int4 : ExtensionOperand<123>;
+defm SPV_KHR_float_controls2 : ExtensionOperand<124>;
//===----------------------------------------------------------------------===//
// Multiclass used to define Capabilities enum values and at the same time
@@ -489,6 +490,8 @@ defm DotProductInput4x8Bit : CapabilityOperand<6017, 0x10600, 0, [SPV_KHR_intege
defm DotProductInput4x8BitPacked : CapabilityOperand<6018, 0x10600, 0, [SPV_KHR_integer_dot_product], []>;
defm DotProduct : CapabilityOperand<6019, 0x10600, 0, [SPV_KHR_integer_dot_product], []>;
defm GroupNonUniformRotateKHR : CapabilityOperand<6026, 0, 0, [SPV_KHR_subgroup_rotate], [GroupNonUniform]>;
+defm FloatControls2
+ : CapabilityOperand<6029, 0x10200, 0, [SPV_KHR_float_controls2], []>;
defm AtomicFloat32AddEXT : CapabilityOperand<6033, 0, 0, [SPV_EXT_shader_atomic_float_add], []>;
defm AtomicFloat64AddEXT : CapabilityOperand<6034, 0, 0, [SPV_EXT_shader_atomic_float_add], []>;
defm AtomicFloat16AddEXT : CapabilityOperand<6095, 0, 0, [SPV_EXT_shader_atomic_float16_add], []>;
@@ -1239,7 +1242,7 @@ defm XfbBuffer : DecorationOperand<36, 0, 0, [], [TransformFeedback]>;
defm XfbStride : DecorationOperand<37, 0, 0, [], [TransformFeedback]>;
defm FuncParamAttr : DecorationOperand<38, 0, 0, [], [Kernel]>;
defm FPRoundingMode : DecorationOperand<39, 0, 0, [], []>;
-defm FPFastMathMode : DecorationOperand<40, 0, 0, [], [Kernel]>;
+defm FPFastMathMode : DecorationOperand<40, 0, 0, [], [Kernel, FloatControls2]>;
defm LinkageAttributes : DecorationOperand<41, 0, 0, [], [Linkage]>;
defm NoContraction : DecorationOperand<42, 0, 0, [], [Shader]>;
defm InputAttachmentIndex : DecorationOperand<43, 0, 0, [], [InputAttachment]>;
diff --git a/llvm/test/CodeGen/SPIRV/capability-FloatControl2.ll b/llvm/test/CodeGen/SPIRV/capability-FloatControl2.ll
new file mode 100644
index 0000000000000..657bbb757199e
--- /dev/null
+++ b/llvm/test/CodeGen/SPIRV/capability-FloatControl2.ll
@@ -0,0 +1,28 @@
+; RUN: llc -O0 -mtriple=spirv1.6-vulkan1.3-compute %s -o - | FileCheck %s --check-prefix=CHECK-NOEXT
+; RUN: %if spirv-tools %{ llc -O0 -mtriple=spirv1.6-vulkan1.3-compute %s -o - -filetype=obj | spirv-val %}
+
+; RUN: llc -O0 -mtriple=spirv1.6-vulkan1.3-compute -spirv-ext=+SPV_KHR_float_controls2 %s -o - | FileCheck %s --check-prefix=CHECK-EXT
+; RUN: %if spirv-tools %{ llc -O0 -mtriple=spirv1.6-vulkan1.3-compute %s -o - -filetype=obj | spirv-val %}
+
+; CHECK-NOEXT-NOT: OpDecorate FPFastMathMode
+
+; CHECK-EXT: OpCapability FloatControls2
+; CHECK-EXT: OpExtension "SPV_KHR_float_controls2"
+; CHECK-EXT: OpDecorate {{%[0-9]+}} FPFastMathMode NotNaN|NotInf|NSZ|AllowRecip|Fast
+
+@.str = private unnamed_addr constant [3 x i8] c"In\00", align 1
+@.str.2 = private unnamed_addr constant [4 x i8] c"Out\00", align 1
+
+define void @main() local_unnamed_addr #0 {
+ %1 = tail call target("spirv.Image", float, 5, 2, 0, 0, 2, 0) @llvm.spv.resource.handlefrombinding.tspirv.Image_f32_5_2_0_0_2_0t(i32 0, i32 0, i32 1, i32 0, i1 false, ptr nonnull @.str)
+ %2 = tail call target("spirv.Image", float, 5, 2, 0, 0, 2, 0) @llvm.spv.resource.handlefrombinding.tspirv.Image_f32_5_2_0_0_2_0t(i32 0, i32 1, i32 1, i32 0, i1 false, ptr nonnull @.str.2)
+ %3 = tail call i32 @llvm.spv.thread.id.in.group(i32 0)
+ %4 = tail call noundef align 4 dereferenceable(4) ptr addrspace(11) @llvm.spv.resource.getpointer.p11.tspirv.Image_f32_5_2_0_0_2_0t(target("spirv.Image", float, 5, 2, 0, 0, 2, 0) %1, i32 %3)
+ %5 = load float, ptr addrspace(11) %4, align 4
+ %6 = fadd reassoc nnan ninf nsz arcp afn float %5, 0x4011333340000000
+ %7 = tail call noundef align 4 dereferenceable(4) ptr addrspace(11) @llvm.spv.resource.getpointer.p11.tspirv.Image_f32_5_2_0_0_2_0t(target("spirv.Image", float, 5, 2, 0, 0, 2, 0) %2, i32 %3)
+ store float %6, ptr addrspace(11) %7, align 4
+ ret void
+}
+
+attributes #0 = { mustprogress nofree noinline norecurse nosync nounwind willreturn memory(readwrite, inaccessiblemem: none) "approx-func-fp-math"="true" "frame-pointer"="all" "hlsl.numthreads"="8,1,1" "hlsl.shader"="compute" "no-infs-fp-math"="true" "no-nans-fp-math"="true" "no-signed-zeros-fp-math"="true" "no-trapping-math"="true" "stack-protector-buffer-size"="8" }
|
Keenuts
reviewed
Jun 16, 2025
Keenuts
reviewed
Jun 17, 2025
Keenuts
approved these changes
Jun 17, 2025
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Add handling for FPFastMathMode in SPIR-V shaders. This is a first pass that
simply does a direct translation when the proper extension is available.
This will unblock work for HLSL. However, it is not a full solution.
The default math mode for spir-v is determined by the API. When
targeting Vulkan many of the fast math options are assumed. We should do
something particular when targeting Vulkan.
We will also need to handle the hlsl "precise" keyword correctly when
FPFastMathMode is not available.
Unblockes #140739, but we are
keeing it open to track the remaining issues mentioned above.