[DirectX] Set shader feature flags MinimumPrecision and NativeLowPrecision, and refactor the logic for setting low-precision-related flags

llvm/include/llvm/BinaryFormat/DXContainerConstants.def

@@ -34,7 +34,7 @@ SHADER_FEATURE_FLAG(14, 19, WaveOps, "Wave level operations")
 SHADER_FEATURE_FLAG(15, 20, Int64Ops, "64-Bit integer")
 SHADER_FEATURE_FLAG(16, 21, ViewID, "View Instancing")
 SHADER_FEATURE_FLAG(17, 22, Barycentrics, "Barycentrics")
-SHADER_FEATURE_FLAG(18, -1, NativeLowPrecision, "Use native low precision")
+SHADER_FEATURE_FLAG(18, 23, NativeLowPrecision, "Use native low precision")
 SHADER_FEATURE_FLAG(19, 24, ShadingRate, "Shading Rate")
 SHADER_FEATURE_FLAG(20, 25, Raytracing_Tier_1_1, "Raytracing tier 1.1 features")
 SHADER_FEATURE_FLAG(21, 26, SamplerFeedback, "Sampler feedback")
@@ -117,7 +117,6 @@ DXIL_MODULE_FLAG( 3,  ForceEarlyDepthStencil, "Force early depth-stencil test")
 DXIL_MODULE_FLAG( 4,  EnableRawAndStructuredBuffers, "Raw and structured buffers")
 DXIL_MODULE_FLAG( 5,  LowPrecisionPresent, "Low-precision data types")
 DXIL_MODULE_FLAG( 8,  AllResourcesBound, "All resources bound for the duration of shader execution")
-DXIL_MODULE_FLAG(23,  UseNativeLowPrecision, "Use native low precision")
 DXIL_MODULE_FLAG(33,  ResMayNotAlias, "Any UAV may not alias any other UAV")
 
 #undef DXIL_MODULE_FLAG

llvm/lib/Target/DirectX/DXILShaderFlags.cpp

@@ -142,6 +142,13 @@ void ModuleShaderFlags::updateFunctionFlags(ComputedShaderFlags &CSF,
     }
   }
 
+  if (CSF.LowPrecisionPresent) {
+    if (NativeLowPrecisionAvailable)
+      CSF.NativeLowPrecision = true;
+    else
+      CSF.MinimumPrecision = true;
+  }
+
   if (!CSF.Int64Ops)
     CSF.Int64Ops = I.getType()->isIntegerTy(64);
 
@@ -206,14 +213,23 @@ void ModuleShaderFlags::initialize(Module &M, DXILResourceTypeMap &DRTM,
                                    const ModuleMetadataInfo &MMDI) {
 
   CanSetResMayNotAlias = MMDI.DXILVersion >= VersionTuple(1, 7);
-
-  // Check if -res-may-alias was provided on the command line.
-  // The command line option will set the dx.resmayalias module flag to 1.
+  // The command line option -res-may-alias will set the dx.resmayalias module
+  // flag to 1 and disable the ability to set the ResMayNotAlias flag
   if (auto *RMA = mdconst::extract_or_null<ConstantInt>(
           M.getModuleFlag("dx.resmayalias")))
     if (RMA->getValue() != 0)
       CanSetResMayNotAlias = false;
 
+  // NativeLowPrecision can only be set when the command line option
+  // -enable-16bit-types is provided. This is indicated by the dx.nativelowprec
+  // module flag being set
+  NativeLowPrecisionAvailable = false;
+  if (auto *NativeLowPrec = mdconst::extract_or_null<ConstantInt>(
+          M.getModuleFlag("dx.nativelowprec")))
+    if (MMDI.ShaderModelVersion >= VersionTuple(6, 2) &&
+        NativeLowPrec->getValue() != 0)
+      NativeLowPrecisionAvailable = true;
+
   CallGraph CG(M);
 
   // Compute Shader Flags Mask for all functions using post-order visit of SCC
@@ -243,13 +259,6 @@ void ModuleShaderFlags::initialize(Module &M, DXILResourceTypeMap &DRTM,
       if (CanSetResMayNotAlias && MMDI.ValidatorVersion < VersionTuple(1, 8))
         SCCSF.ResMayNotAlias = !DRM.uavs().empty();
 
-      // Set UseNativeLowPrecision using dx.nativelowprec module metadata
-      if (auto *NativeLowPrec = mdconst::extract_or_null<ConstantInt>(
-              M.getModuleFlag("dx.nativelowprec")))
-        if (MMDI.ShaderModelVersion >= VersionTuple(6, 2) &&
-            NativeLowPrec->getValue() != 0)
-          SCCSF.UseNativeLowPrecision = true;
-
       ComputedShaderFlags CSF;
       for (const auto &BB : *F)
         for (const auto &I : BB)

llvm/lib/Target/DirectX/DXILShaderFlags.h

@@ -92,6 +92,7 @@ struct ModuleShaderFlags {
 
 private:
   bool CanSetResMayNotAlias;
+  bool NativeLowPrecisionAvailable;
   /// Map of Function-Shader Flag Mask pairs representing properties of each of
   /// the functions in the module. Shader Flags of each function represent both
   /// module-level and function-level flags

llvm/test/CodeGen/DirectX/ShaderFlags/low-precision.ll

@@ -1,30 +1,44 @@
 ; RUN: opt -S --passes="print-dx-shader-flags" 2>&1 %s | FileCheck %s
+; RUN: llc %s --filetype=obj -o - | obj2yaml | FileCheck %s --check-prefix=DXC
+
+; Check that when the dx.nativelowprec module flag is not specified, the
+; module-level shader flag UseNativeLowPrecision is not set, and the
+; MinimumPrecision feature flag is set due to the presence of half and i16
+; without native low precision.
 
 target triple = "dxil-pc-shadermodel6.7-library"
 
 ;CHECK: ; Combined Shader Flags for Module
 ;CHECK-NEXT: ; Shader Flags Value: 0x00000020
 ;CHECK-NEXT: ;
 ;CHECK-NEXT: ; Note: shader requires additional functionality:
+;CHECK-NEXT: ;       Minimum-precision data types
 ;CHECK-NEXT: ; Note: extra DXIL module flags:
 ;CHECK-NEXT: ;       Low-precision data types
 ;CHECK-NEXT: ;
 ;CHECK-NEXT: ; Shader Flags for Module Functions
 
 ;CHECK-LABEL: ; Function add_i16 : 0x00000020
-define i16 @add_i16(i16 %a, i16 %b) {
+define i16 @add_i16(i16 %a, i16 %b) "hlsl.export" {
   %sum = add i16 %a, %b
   ret i16 %sum
 }
 
 ;CHECK-LABEL: ; Function add_i32 : 0x00000000
-define i32 @add_i32(i32 %a, i32 %b) {
+define i32 @add_i32(i32 %a, i32 %b) "hlsl.export" {
   %sum = add i32 %a, %b
   ret i32 %sum
 }
 
 ;CHECK-LABEL: ; Function add_half : 0x00000020
-define half @add_half(half %a, half %b) {
+define half @add_half(half %a, half %b) "hlsl.export" {
   %sum = fadd half %a, %b
   ret half %sum
 }
+
+; DXC: - Name:            SFI0
+; DXC-NEXT:     Size:            8
+; DXC-NEXT:     Flags:
+; DXC:      MinimumPrecision: true
+; DXC:      NativeLowPrecision: false
+; DXC: ...

llvm/test/CodeGen/DirectX/ShaderFlags/use-native-low-precision-0.ll

@@ -1,37 +1,47 @@
 ; RUN: opt -S --passes="print-dx-shader-flags" 2>&1 %s | FileCheck %s
+; RUN: llc %s --filetype=obj -o - | obj2yaml | FileCheck %s --check-prefix=DXC
 
 ; Check that when the dx.nativelowprec module flag is set to 0, the module-level
-; shader flag UseNativeLowPrecision is not set
+; shader flag UseNativeLowPrecision is not set, and the MinimumPrecision feature
+; flag is set due to the presence of half and i16 without native low precision.
 
 target triple = "dxil-pc-shadermodel6.7-library"
 
 ;CHECK: ; Combined Shader Flags for Module
 ;CHECK-NEXT: ; Shader Flags Value: 0x00000020
 ;CHECK-NEXT: ;
 ;CHECK-NEXT: ; Note: shader requires additional functionality:
+;CHECK-NEXT: ;       Minimum-precision data types
 ;CHECK-NEXT: ; Note: extra DXIL module flags:
 ;CHECK-NEXT: ;       Low-precision data types
 ;CHECK-NOT:  ;       Native 16bit types enabled
 ;CHECK-NEXT: ;
 ;CHECK-NEXT: ; Shader Flags for Module Functions
 
 ;CHECK-LABEL: ; Function add_i16 : 0x00000020
-define i16 @add_i16(i16 %a, i16 %b) {
+define i16 @add_i16(i16 %a, i16 %b) "hlsl.export" {
   %sum = add i16 %a, %b
   ret i16 %sum
 }
 
 ;CHECK-LABEL: ; Function add_i32 : 0x00000000
-define i32 @add_i32(i32 %a, i32 %b) {
+define i32 @add_i32(i32 %a, i32 %b) "hlsl.export" {
   %sum = add i32 %a, %b
   ret i32 %sum
 }
 
 ;CHECK-LABEL: ; Function add_half : 0x00000020
-define half @add_half(half %a, half %b) {
+define half @add_half(half %a, half %b) "hlsl.export" {
   %sum = fadd half %a, %b
   ret half %sum
 }
 
 !llvm.module.flags = !{!0}
 !0 = !{i32 1, !"dx.nativelowprec", i32 0}
+
+; DXC: - Name:            SFI0
+; DXC-NEXT:     Size:            8
+; DXC-NEXT:     Flags:
+; DXC:      MinimumPrecision: true
+; DXC:      NativeLowPrecision: false
+; DXC: ...

llvm/test/CodeGen/DirectX/ShaderFlags/use-native-low-precision-1.ll

@@ -1,37 +1,49 @@
 ; RUN: opt -S --passes="print-dx-shader-flags" 2>&1 %s | FileCheck %s
+; RUN: llc %s --filetype=obj -o - | obj2yaml | FileCheck %s --check-prefix=DXC
+
+; Check that when the dx.nativelowprec module flag is set to 1, the module-level
+; shader flag UseNativeLowPrecision is set, and the NativeLowPrecision feature
+; flag is set
 
 target triple = "dxil-pc-shadermodel6.7-library"
 
 ;CHECK: ; Combined Shader Flags for Module
 ;CHECK-NEXT: ; Shader Flags Value: 0x00800020
 ;CHECK-NEXT: ;
 ;CHECK-NEXT: ; Note: shader requires additional functionality:
+;CHECK-NEXT: ;       Use native low precision
 ;CHECK-NEXT: ; Note: extra DXIL module flags:
 ;CHECK-NEXT: ;       Low-precision data types
-;CHECK-NEXT: ;       Use native low precision
 ;CHECK-NEXT: ;
 ;CHECK-NEXT: ; Shader Flags for Module Functions
 
 ;CHECK-LABEL: ; Function add_i16 : 0x00800020
-define i16 @add_i16(i16 %a, i16 %b) {
+define i16 @add_i16(i16 %a, i16 %b) "hlsl.export" {
   %sum = add i16 %a, %b
   ret i16 %sum
 }
 
-; NOTE: The flag for native low precision (0x80000) is set for every function
-; in the module regardless of whether or not the function uses low precision
-; data types (flag 0x20). This matches the behavior in DXC
-;CHECK-LABEL: ; Function add_i32 : 0x00800000
-define i32 @add_i32(i32 %a, i32 %b) {
+; NOTE: In DXC, the flag for native low precision (0x80000) is set for every
+; function in the module regardless of whether or not the function uses low
+; precision data types (flag 0x20). However, this will not be the case for Clang
+;CHECK-LABEL: ; Function add_i32 : 0x00000000
+define i32 @add_i32(i32 %a, i32 %b) "hlsl.export" {
   %sum = add i32 %a, %b
   ret i32 %sum
 }
 
 ;CHECK-LABEL: ; Function add_half : 0x00800020
-define half @add_half(half %a, half %b) {
+define half @add_half(half %a, half %b) "hlsl.export" {
   %sum = fadd half %a, %b
   ret half %sum
 }
 
 !llvm.module.flags = !{!0}
 !0 = !{i32 1, !"dx.nativelowprec", i32 1}
+
+; DXC: - Name:            SFI0
+; DXC-NEXT:     Size:            8
+; DXC-NEXT:     Flags:
+; DXC:      MinimumPrecision: false
+; DXC:      NativeLowPrecision: true
+; DXC: ...