[AMDGPU] Add SubtargetFeature for dynamic VGPR mode llvm#130030

Author: qiaojbao

llvm/docs/AMDGPUUsage.rst

@@ -772,6 +772,12 @@ For example:
                                                   enabled will execute correctly but may be less
                                                   performant than code generated for XNACK replay
                                                   disabled.
+
+     dynamic-vgpr    TODO                         Represents the "Dynamic VGPR" hardware mode, introduced in GFX12.
+                                                  Waves launched in this mode may allocate or deallocate the VGPRs
+                                                  using dedicated instructions, but may not send the DEALLOC_VGPRS
+                                                  message.
+
      =============== ============================ ==================================================
 
 .. _amdgpu-target-id:

llvm/lib/Target/AMDGPU/AMDGPU.td

@@ -1245,6 +1245,12 @@ def FeatureXF32Insts : SubtargetFeature<"xf32-insts",
    "v_mfma_f32_16x16x8_xf32 and v_mfma_f32_32x32x4_xf32"
  >;
 
+def FeatureDynamicVGPR : SubtargetFeature <"dynamic-vgpr",
+  "DynamicVGPR",
+  "true",
+  "Enable dynamic VGPR mode"
+>;
+
 // Enable the use of SCRATCH_STORE/LOAD_BLOCK instructions for saving and
 // restoring the callee-saved registers.
 def FeatureUseBlockVGPROpsForCSR : SubtargetFeature<"block-vgpr-csr",

llvm/lib/Target/AMDGPU/AMDGPUAsmPrinter.cpp

@@ -1417,6 +1417,9 @@ static void EmitPALMetadataCommon(AMDGPUPALMetadata *MD,
     MD->setHwStage(CC, ".trap_present",
                    (bool)CurrentProgramInfo.TrapHandlerEnable);
     MD->setHwStage(CC, ".excp_en", CurrentProgramInfo.EXCPEnable);
+
+    if (ST.isDynamicVGPREnabled())
+      MD->setComputeRegisters(".dynamic_vgpr_en", true);
   }
 
   MD->setHwStage(CC, ".lds_size",

llvm/lib/Target/AMDGPU/GCNSubtarget.h

@@ -190,6 +190,7 @@ class GCNSubtarget final : public AMDGPUGenSubtargetInfo,
   /// indicates a lack of S_CLAUSE support.
   unsigned MaxHardClauseLength = 0;
   bool SupportsSRAMECC = false;
+  bool DynamicVGPR = false;
 
   // This should not be used directly. 'TargetID' tracks the dynamic settings
   // for SRAMECC.
@@ -1660,6 +1661,8 @@ class GCNSubtarget final : public AMDGPUGenSubtargetInfo,
     return true;
   }
 
+  bool isDynamicVGPREnabled() const { return DynamicVGPR; }
+
   bool requiresDisjointEarlyClobberAndUndef() const override {
     // AMDGPU doesn't care if early-clobber and undef operands are allocated
     // to the same register.

llvm/test/CodeGen/AMDGPU/pal-metadata-3.0.ll

@@ -1,4 +1,6 @@
-; RUN: llc -mtriple=amdgcn--amdpal -mcpu=gfx1100 <%s | FileCheck %s
+; RUN: llc -mtriple=amdgcn--amdpal -mcpu=gfx1100 <%s | FileCheck %s --check-prefixes=CHECK,GFX11
+; RUN: llc -mtriple=amdgcn--amdpal -mcpu=gfx1200 <%s | FileCheck %s --check-prefixes=CHECK
+; RUN: llc -mtriple=amdgcn--amdpal -mcpu=gfx1200 -mattr=+dynamic-vgpr <%s | FileCheck %s --check-prefixes=CHECK,DVGPR
 
 ; CHECK-LABEL: {{^}}_amdgpu_cs_main:
 ; CHECK: ; TotalNumSgprs: 4
@@ -8,6 +10,7 @@
 ; CHECK-NEXT: amdpal.pipelines:
 ; CHECK-NEXT:   - .api:            Vulkan
 ; CHECK-NEXT:     .compute_registers:
+; DVGPR-NEXT:       .dynamic_vgpr_en:   true
 ; CHECK-NEXT:       .tg_size_en:     true
 ; CHECK-NEXT:       .tgid_x_en:      false
 ; CHECK-NEXT:       .tgid_y_en:      false
@@ -57,7 +60,7 @@
 ; CHECK-NEXT:        .entry_point_symbol:    _amdgpu_cs_main
 ; CHECK-NEXT:        .excp_en:        0
 ; CHECK-NEXT:        .float_mode:     0xc0
-; CHECK-NEXT:        .ieee_mode:      false
+; GFX11-NEXT:        .ieee_mode:      false
 ; CHECK-NEXT:        .image_op:       false
 ; CHECK-NEXT:        .lds_size:       0
 ; CHECK-NEXT:        .mem_ordered:    true
@@ -112,7 +115,7 @@
 ; CHECK-NEXT:        .debug_mode:     false
 ; CHECK-NEXT:        .entry_point:    _amdgpu_gs
 ; CHECK-NEXT:        .entry_point_symbol:    gs_shader
-; CHECK-NEXT:        .ieee_mode:      false
+; GFX11-NEXT:        .ieee_mode:      false
 ; CHECK-NEXT:        .lds_size:       0x200
 ; CHECK-NEXT:        .mem_ordered:    true
 ; CHECK-NEXT:        .scratch_en:     false
@@ -124,7 +127,7 @@
 ; CHECK-NEXT:        .debug_mode:     false
 ; CHECK-NEXT:        .entry_point:    _amdgpu_hs
 ; CHECK-NEXT:        .entry_point_symbol:    hs_shader
-; CHECK-NEXT:        .ieee_mode:      false
+; GFX11-NEXT:        .ieee_mode:      false
 ; CHECK-NEXT:        .lds_size:       0x1000
 ; CHECK-NEXT:        .mem_ordered:    true
 ; CHECK-NEXT:        .scratch_en:     false
@@ -136,7 +139,7 @@
 ; CHECK-NEXT:        .debug_mode:     false
 ; CHECK-NEXT:        .entry_point:    _amdgpu_ps
 ; CHECK-NEXT:        .entry_point_symbol:    ps_shader
-; CHECK-NEXT:        .ieee_mode:      false
+; GFX11-NEXT:        .ieee_mode:      false
 ; CHECK-NEXT:        .lds_size:       0
 ; CHECK-NEXT:        .mem_ordered:    true
 ; CHECK-NEXT:        .scratch_en:     false