[llvm] [AMDGPU] Add intrinsic exposing s_alloc_vgpr (PR #163951)

Diana Picus via llvm-commits llvm-commits at lists.llvm.org
Mon Feb 9 03:08:31 PST 2026


https://github.com/rovka updated https://github.com/llvm/llvm-project/pull/163951

>From f084590dd6b12d28e17cb2a7fd54e3da7828a924 Mon Sep 17 00:00:00 2001
From: Diana Picus <diana-magda.picus at amd.com>
Date: Thu, 16 Oct 2025 13:48:09 +0200
Subject: [PATCH 1/6] [AMDGPU] Add intrinsic exposing s_alloc_vgpr

Make it possible to use `s_alloc_vgpr` at the IR level. This is a huge
footgun and use for anything other than compiler internal purposes is
heavily discouraged. The calling code must make sure that it does not
allocate fewer VGPRs than necessary - the intrinsic is NOT a request to
the backend to limit the number of VGPRs it uses (in essence it's not so
different from what we do with the dynamic VGPR flags of the
`amdgcn.cs.chain` intrinsic, it just makes it possible to use this
functionality in other scenarios).
---
 llvm/include/llvm/IR/IntrinsicsAMDGPU.td      | 11 ++++
 .../AMDGPU/AMDGPUInstructionSelector.cpp      | 16 +++++
 .../Target/AMDGPU/AMDGPURegisterBankInfo.cpp  |  4 ++
 .../Target/AMDGPU/AMDGPUSearchableTables.td   |  1 +
 llvm/lib/Target/AMDGPU/SOPInstructions.td     |  6 +-
 .../AMDGPU/always_uniform.ll                  |  9 +++
 .../AMDGPU/intrinsic-amdgcn-s-alloc-vgpr.ll   | 63 +++++++++++++++++++
 7 files changed, 108 insertions(+), 2 deletions(-)
 create mode 100644 llvm/test/CodeGen/AMDGPU/intrinsic-amdgcn-s-alloc-vgpr.ll

diff --git a/llvm/include/llvm/IR/IntrinsicsAMDGPU.td b/llvm/include/llvm/IR/IntrinsicsAMDGPU.td
index a8eba9ed126b7..fd7d00e824ac2 100644
--- a/llvm/include/llvm/IR/IntrinsicsAMDGPU.td
+++ b/llvm/include/llvm/IR/IntrinsicsAMDGPU.td
@@ -397,6 +397,17 @@ def int_amdgcn_s_wait_loadcnt        : AMDGPUWaitIntrinsic;
 def int_amdgcn_s_wait_samplecnt      : AMDGPUWaitIntrinsic;
 def int_amdgcn_s_wait_storecnt       : AMDGPUWaitIntrinsic;
 
+// Force the VGPR allocation of the current wave to (at least) the given value.
+// The actual number of allocated VGPRs may be rounded up to match hardware
+// block boundaries.
+// It is the responsibility of the calling code to ensure it does not allocate
+// below the VGPR requirements of the current shader.
+def int_amdgcn_s_alloc_vgpr :
+  Intrinsic<
+    [llvm_i1_ty], // Returns true if the allocation succeeded, false otherwise.
+    [llvm_i32_ty], // The number of VGPRs to allocate.
+    [NoUndef<RetIndex>, IntrNoMem, IntrHasSideEffects, IntrConvergent, IntrWillReturn, IntrNoCallback, IntrNoFree]>;
+
 def int_amdgcn_div_scale : PureIntrinsic<
   // 1st parameter: Numerator
   // 2nd parameter: Denominator
diff --git a/llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp b/llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp
index 15056a9a9cd65..7b8b4bd7b1062 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp
@@ -2391,6 +2391,22 @@ bool AMDGPUInstructionSelector::selectG_INTRINSIC_W_SIDE_EFFECTS(
   case Intrinsic::amdgcn_ds_bvh_stack_push8_pop1_rtn:
   case Intrinsic::amdgcn_ds_bvh_stack_push8_pop2_rtn:
     return selectDSBvhStackIntrinsic(I);
+  case Intrinsic::amdgcn_s_alloc_vgpr: {
+    // S_ALLOC_VGPR doesn't have a destination register, it just implicitly sets
+    // SCC. We then need to COPY it into the result vreg.
+    MachineBasicBlock *MBB = I.getParent();
+    const DebugLoc &DL = I.getDebugLoc();
+
+    Register ResReg = I.getOperand(0).getReg();
+
+    MachineInstr *AllocMI = BuildMI(*MBB, &I, DL, TII.get(AMDGPU::S_ALLOC_VGPR))
+                                .add(I.getOperand(2));
+    MachineInstr *CopyMI = BuildMI(*MBB, &I, DL, TII.get(AMDGPU::COPY), ResReg)
+                               .addReg(AMDGPU::SCC);
+    I.eraseFromParent();
+    return constrainSelectedInstRegOperands(*AllocMI, TII, TRI, RBI) &&
+           RBI.constrainGenericRegister(ResReg, AMDGPU::SReg_32RegClass, *MRI);
+  }
   case Intrinsic::amdgcn_s_barrier_init:
   case Intrinsic::amdgcn_s_barrier_signal_var:
     return selectNamedBarrierInit(I, IntrinsicID);
diff --git a/llvm/lib/Target/AMDGPU/AMDGPURegisterBankInfo.cpp b/llvm/lib/Target/AMDGPU/AMDGPURegisterBankInfo.cpp
index 7470fecd3c03f..4eaf728be6e89 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPURegisterBankInfo.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPURegisterBankInfo.cpp
@@ -5380,6 +5380,10 @@ AMDGPURegisterBankInfo::getInstrMapping(const MachineInstr &MI) const {
       OpdsMapping[6] = AMDGPU::getValueMapping(AMDGPU::VGPRRegBankID, 32);
       OpdsMapping[8] = getSGPROpMapping(MI.getOperand(8).getReg(), MRI, *TRI);
       break;
+    case Intrinsic::amdgcn_s_alloc_vgpr:
+      OpdsMapping[0] = AMDGPU::getValueMapping(AMDGPU::SGPRRegBankID, 1);
+      OpdsMapping[2] = AMDGPU::getValueMapping(AMDGPU::SGPRRegBankID, 32);
+      break;
     case Intrinsic::amdgcn_s_sendmsg:
     case Intrinsic::amdgcn_s_sendmsghalt: {
       // This must be an SGPR, but accept a VGPR.
diff --git a/llvm/lib/Target/AMDGPU/AMDGPUSearchableTables.td b/llvm/lib/Target/AMDGPU/AMDGPUSearchableTables.td
index 58a9b5511f2d0..963bb91b6a7df 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUSearchableTables.td
+++ b/llvm/lib/Target/AMDGPU/AMDGPUSearchableTables.td
@@ -407,6 +407,7 @@ def : AlwaysUniform<int_amdgcn_cluster_workgroup_max_flat_id>;
 def : AlwaysUniform<int_amdgcn_workgroup_id_x>;
 def : AlwaysUniform<int_amdgcn_workgroup_id_y>;
 def : AlwaysUniform<int_amdgcn_workgroup_id_z>;
+def : AlwaysUniform<int_amdgcn_s_alloc_vgpr>;
 def : AlwaysUniform<int_amdgcn_s_getpc>;
 def : AlwaysUniform<int_amdgcn_s_getreg>;
 def : AlwaysUniform<int_amdgcn_s_memrealtime>;
diff --git a/llvm/lib/Target/AMDGPU/SOPInstructions.td b/llvm/lib/Target/AMDGPU/SOPInstructions.td
index 99b352bdf6765..a4f8b9bc67346 100644
--- a/llvm/lib/Target/AMDGPU/SOPInstructions.td
+++ b/llvm/lib/Target/AMDGPU/SOPInstructions.td
@@ -433,8 +433,10 @@ let SubtargetPredicate = isGFX11Plus in {
 } // End SubtargetPredicate = isGFX11Plus
 
 let SubtargetPredicate = isGFX12Plus in {
-  let hasSideEffects = 1, Defs = [SCC] in {
-    def S_ALLOC_VGPR : SOP1_0_32 <"s_alloc_vgpr">;
+  let hasSideEffects = 1, isConvergent = 1, Defs = [SCC] in {
+    def S_ALLOC_VGPR : SOP1_0_32 <"s_alloc_vgpr",
+      [(set SCC, (int_amdgcn_s_alloc_vgpr SSrc_b32:$src0))]
+    >;
   }
 } // End SubtargetPredicate = isGFX12Plus
 
diff --git a/llvm/test/Analysis/UniformityAnalysis/AMDGPU/always_uniform.ll b/llvm/test/Analysis/UniformityAnalysis/AMDGPU/always_uniform.ll
index 9ff670bee0f89..3f56f12f3cb34 100644
--- a/llvm/test/Analysis/UniformityAnalysis/AMDGPU/always_uniform.ll
+++ b/llvm/test/Analysis/UniformityAnalysis/AMDGPU/always_uniform.ll
@@ -183,6 +183,15 @@ define void @cluster_workgroup_max_flat_id(ptr addrspace(1) inreg %out) {
   ret void
 }
 
+; CHECK-LABEL: for function 's_alloc_vgpr':
+; CHECK: ALL VALUES UNIFORM
+define void @s_alloc_vgpr(i32 inreg %n, ptr addrspace(1) inreg %out) {
+  %scc = call i1 @llvm.amdgcn.s.alloc.vgpr(i32 %n)
+  %sel = select i1 %scc, i32 1, i32 0
+  store i32 %sel, ptr addrspace(1) %out
+  ret void
+}
+
 ; CHECK-LABEL: for function 's_memtime':
 ; CHECK: ALL VALUES UNIFORM
 define void @s_memtime(ptr addrspace(1) inreg %out) {
diff --git a/llvm/test/CodeGen/AMDGPU/intrinsic-amdgcn-s-alloc-vgpr.ll b/llvm/test/CodeGen/AMDGPU/intrinsic-amdgcn-s-alloc-vgpr.ll
new file mode 100644
index 0000000000000..58b41b87962a0
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/intrinsic-amdgcn-s-alloc-vgpr.ll
@@ -0,0 +1,63 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 6
+; RUN: llc -global-isel=1 -mtriple=amdgcn--amdpal -mcpu=gfx1250 < %s | FileCheck %s --check-prefix=GISEL
+; RUN: llc -global-isel=0 -mtriple=amdgcn--amdpal -mcpu=gfx1250 < %s | FileCheck %s --check-prefix=DAGISEL
+
+declare i1 @llvm.amdgcn.s.alloc.vgpr(i32)
+
+define amdgpu_cs void @test_alloc_vreg_const(ptr addrspace(1) %out) #0 {
+; GISEL-LABEL: test_alloc_vreg_const:
+; GISEL:       ; %bb.0: ; %entry
+; GISEL-NEXT:    s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1
+; GISEL-NEXT:    s_alloc_vgpr 45
+; GISEL-NEXT:    s_cselect_b32 s0, 1, 0
+; GISEL-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1)
+; GISEL-NEXT:    s_and_b32 s0, s0, 1
+; GISEL-NEXT:    v_mov_b32_e32 v2, s0
+; GISEL-NEXT:    global_store_b32 v[0:1], v2, off
+; GISEL-NEXT:    s_endpgm
+;
+; DAGISEL-LABEL: test_alloc_vreg_const:
+; DAGISEL:       ; %bb.0: ; %entry
+; DAGISEL-NEXT:    s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1
+; DAGISEL-NEXT:    s_alloc_vgpr 45
+; DAGISEL-NEXT:    s_cselect_b32 s0, -1, 0
+; DAGISEL-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
+; DAGISEL-NEXT:    v_cndmask_b32_e64 v2, 0, 1, s0
+; DAGISEL-NEXT:    global_store_b32 v[0:1], v2, off
+; DAGISEL-NEXT:    s_endpgm
+entry:
+  %scc = call i1 @llvm.amdgcn.s.alloc.vgpr(i32 45)
+  %sel = select i1 %scc, i32 1, i32 0
+  store i32 %sel, ptr addrspace(1) %out
+  ret void
+}
+
+define amdgpu_cs void @test_alloc_vreg_var(i32 inreg %n, ptr addrspace(1) %out) #0 {
+; GISEL-LABEL: test_alloc_vreg_var:
+; GISEL:       ; %bb.0: ; %entry
+; GISEL-NEXT:    s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1
+; GISEL-NEXT:    s_alloc_vgpr s0
+; GISEL-NEXT:    s_cselect_b32 s0, 1, 0
+; GISEL-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1)
+; GISEL-NEXT:    s_and_b32 s0, s0, 1
+; GISEL-NEXT:    v_mov_b32_e32 v2, s0
+; GISEL-NEXT:    global_store_b32 v[0:1], v2, off
+; GISEL-NEXT:    s_endpgm
+;
+; DAGISEL-LABEL: test_alloc_vreg_var:
+; DAGISEL:       ; %bb.0: ; %entry
+; DAGISEL-NEXT:    s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1
+; DAGISEL-NEXT:    s_alloc_vgpr s0
+; DAGISEL-NEXT:    s_cselect_b32 s0, -1, 0
+; DAGISEL-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
+; DAGISEL-NEXT:    v_cndmask_b32_e64 v2, 0, 1, s0
+; DAGISEL-NEXT:    global_store_b32 v[0:1], v2, off
+; DAGISEL-NEXT:    s_endpgm
+entry:
+  %scc = call i1 @llvm.amdgcn.s.alloc.vgpr(i32 %n)
+  %sel = select i1 %scc, i32 1, i32 0
+  store i32 %sel, ptr addrspace(1) %out
+  ret void
+}
+
+attributes #0 = { "amdgpu-dynamic-vgpr-block-sze" = "16" }

>From 7c316c7ff745e022e566d041293f7d139796565b Mon Sep 17 00:00:00 2001
From: Diana Picus <diana-magda.picus at amd.com>
Date: Fri, 17 Oct 2025 14:52:46 +0200
Subject: [PATCH 2/6] Silence warning

---
 llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp b/llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp
index 7b8b4bd7b1062..ee219ff2d3514 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp
@@ -2401,8 +2401,8 @@ bool AMDGPUInstructionSelector::selectG_INTRINSIC_W_SIDE_EFFECTS(
 
     MachineInstr *AllocMI = BuildMI(*MBB, &I, DL, TII.get(AMDGPU::S_ALLOC_VGPR))
                                 .add(I.getOperand(2));
-    MachineInstr *CopyMI = BuildMI(*MBB, &I, DL, TII.get(AMDGPU::COPY), ResReg)
-                               .addReg(AMDGPU::SCC);
+    (void)BuildMI(*MBB, &I, DL, TII.get(AMDGPU::COPY), ResReg)
+        .addReg(AMDGPU::SCC);
     I.eraseFromParent();
     return constrainSelectedInstRegOperands(*AllocMI, TII, TRI, RBI) &&
            RBI.constrainGenericRegister(ResReg, AMDGPU::SReg_32RegClass, *MRI);

>From d7ebd9c44f0366d934d72d18af54356cc18ff7c4 Mon Sep 17 00:00:00 2001
From: Diana Picus <diana-magda.picus at amd.com>
Date: Tue, 21 Oct 2025 10:54:53 +0200
Subject: [PATCH 3/6] Address review comments

---
 llvm/include/llvm/IR/IntrinsicsAMDGPU.td | 9 ++++-----
 1 file changed, 4 insertions(+), 5 deletions(-)

diff --git a/llvm/include/llvm/IR/IntrinsicsAMDGPU.td b/llvm/include/llvm/IR/IntrinsicsAMDGPU.td
index fd7d00e824ac2..656e4c0d108d4 100644
--- a/llvm/include/llvm/IR/IntrinsicsAMDGPU.td
+++ b/llvm/include/llvm/IR/IntrinsicsAMDGPU.td
@@ -397,16 +397,15 @@ def int_amdgcn_s_wait_loadcnt        : AMDGPUWaitIntrinsic;
 def int_amdgcn_s_wait_samplecnt      : AMDGPUWaitIntrinsic;
 def int_amdgcn_s_wait_storecnt       : AMDGPUWaitIntrinsic;
 
-// Force the VGPR allocation of the current wave to (at least) the given value.
-// The actual number of allocated VGPRs may be rounded up to match hardware
-// block boundaries.
+// Request the hardware to allocate the given number of VGPRs. The actual number
+// of allocated VGPRs may be rounded up to match hardware block boundaries.
 // It is the responsibility of the calling code to ensure it does not allocate
 // below the VGPR requirements of the current shader.
 def int_amdgcn_s_alloc_vgpr :
-  Intrinsic<
+  DefaultAttrsIntrinsic<
     [llvm_i1_ty], // Returns true if the allocation succeeded, false otherwise.
     [llvm_i32_ty], // The number of VGPRs to allocate.
-    [NoUndef<RetIndex>, IntrNoMem, IntrHasSideEffects, IntrConvergent, IntrWillReturn, IntrNoCallback, IntrNoFree]>;
+    [NoUndef<RetIndex>, IntrNoMem, IntrHasSideEffects, IntrConvergent]>;
 
 def int_amdgcn_div_scale : PureIntrinsic<
   // 1st parameter: Numerator

>From 35e13a4ea755e38fa7eb8d45967e30fb8ff3408b Mon Sep 17 00:00:00 2001
From: Diana Picus <diana-magda.picus at amd.com>
Date: Wed, 4 Feb 2026 15:06:20 +0100
Subject: [PATCH 4/6] Address review comments

---
 llvm/include/llvm/IR/IntrinsicsAMDGPU.td      |  3 +-
 .../Target/AMDGPU/AMDGPURegisterBankInfo.cpp  |  3 +
 llvm/lib/Target/AMDGPU/SIISelLowering.cpp     | 13 ++++
 .../AMDGPU/intrinsic-amdgcn-s-alloc-vgpr.ll   | 74 +++++++++++++++++--
 4 files changed, 85 insertions(+), 8 deletions(-)

diff --git a/llvm/include/llvm/IR/IntrinsicsAMDGPU.td b/llvm/include/llvm/IR/IntrinsicsAMDGPU.td
index 656e4c0d108d4..633e0a8fededb 100644
--- a/llvm/include/llvm/IR/IntrinsicsAMDGPU.td
+++ b/llvm/include/llvm/IR/IntrinsicsAMDGPU.td
@@ -400,7 +400,8 @@ def int_amdgcn_s_wait_storecnt       : AMDGPUWaitIntrinsic;
 // Request the hardware to allocate the given number of VGPRs. The actual number
 // of allocated VGPRs may be rounded up to match hardware block boundaries.
 // It is the responsibility of the calling code to ensure it does not allocate
-// below the VGPR requirements of the current shader.
+// below the VGPR requirements of the current shader. This intrinsic is only
+// available on targets that support dynamic VGPR mode.
 def int_amdgcn_s_alloc_vgpr :
   DefaultAttrsIntrinsic<
     [llvm_i1_ty], // Returns true if the allocation succeeded, false otherwise.
diff --git a/llvm/lib/Target/AMDGPU/AMDGPURegisterBankInfo.cpp b/llvm/lib/Target/AMDGPU/AMDGPURegisterBankInfo.cpp
index 4eaf728be6e89..21634e15e16b1 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPURegisterBankInfo.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPURegisterBankInfo.cpp
@@ -3282,6 +3282,9 @@ void AMDGPURegisterBankInfo::applyMappingImpl(
       constrainOpWithReadfirstlane(B, MI, 2); // M0
       return;
     }
+    case Intrinsic::amdgcn_s_alloc_vgpr:
+      constrainOpWithReadfirstlane(B, MI, 2);
+      return;
     case Intrinsic::amdgcn_s_sendmsg:
     case Intrinsic::amdgcn_s_sendmsghalt: {
       // FIXME: Should this use a waterfall loop?
diff --git a/llvm/lib/Target/AMDGPU/SIISelLowering.cpp b/llvm/lib/Target/AMDGPU/SIISelLowering.cpp
index 17b64c0330318..e63921df35430 100644
--- a/llvm/lib/Target/AMDGPU/SIISelLowering.cpp
+++ b/llvm/lib/Target/AMDGPU/SIISelLowering.cpp
@@ -10970,6 +10970,19 @@ SDValue SITargetLowering::LowerINTRINSIC_W_CHAIN(SDValue Op,
     return DAG.getAtomic(Opcode, SDLoc(Op), M->getMemoryVT(), M->getVTList(),
                          Ops, M->getMemOperand());
   }
+  case Intrinsic::amdgcn_s_alloc_vgpr: {
+    SDValue NumVGPRs = Op.getOperand(2);
+    if (!NumVGPRs->isDivergent())
+      return Op;
+
+    SDValue ReadFirstLaneID =
+        DAG.getTargetConstant(Intrinsic::amdgcn_readfirstlane, DL, MVT::i32);
+    NumVGPRs = DAG.getNode(ISD::INTRINSIC_WO_CHAIN, DL, MVT::i32,
+                           ReadFirstLaneID, NumVGPRs);
+
+    return DAG.getNode(ISD::INTRINSIC_W_CHAIN, DL, Op->getVTList(),
+                       Op.getOperand(0), Op.getOperand(1), NumVGPRs);
+  }
   case Intrinsic::amdgcn_s_get_barrier_state:
   case Intrinsic::amdgcn_s_get_named_barrier_state: {
     SDValue Chain = Op->getOperand(0);
diff --git a/llvm/test/CodeGen/AMDGPU/intrinsic-amdgcn-s-alloc-vgpr.ll b/llvm/test/CodeGen/AMDGPU/intrinsic-amdgcn-s-alloc-vgpr.ll
index 58b41b87962a0..49426872ed580 100644
--- a/llvm/test/CodeGen/AMDGPU/intrinsic-amdgcn-s-alloc-vgpr.ll
+++ b/llvm/test/CodeGen/AMDGPU/intrinsic-amdgcn-s-alloc-vgpr.ll
@@ -7,26 +7,35 @@ declare i1 @llvm.amdgcn.s.alloc.vgpr(i32)
 define amdgpu_cs void @test_alloc_vreg_const(ptr addrspace(1) %out) #0 {
 ; GISEL-LABEL: test_alloc_vreg_const:
 ; GISEL:       ; %bb.0: ; %entry
+; GISEL-NEXT:    s_getreg_b32 s33, hwreg(HW_REG_WAVE_HW_ID2, 8, 2)
+; GISEL-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_4) | instid1(SALU_CYCLE_1)
+; GISEL-NEXT:    s_cmp_lg_u32 0, s33
+; GISEL-NEXT:    s_cmovk_i32 s33, 0x1c0
 ; GISEL-NEXT:    s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1
 ; GISEL-NEXT:    s_alloc_vgpr 45
 ; GISEL-NEXT:    s_cselect_b32 s0, 1, 0
-; GISEL-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1)
 ; GISEL-NEXT:    s_and_b32 s0, s0, 1
+; GISEL-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
 ; GISEL-NEXT:    v_mov_b32_e32 v2, s0
 ; GISEL-NEXT:    global_store_b32 v[0:1], v2, off
+; GISEL-NEXT:    s_alloc_vgpr 0
 ; GISEL-NEXT:    s_endpgm
 ;
 ; DAGISEL-LABEL: test_alloc_vreg_const:
 ; DAGISEL:       ; %bb.0: ; %entry
+; DAGISEL-NEXT:    s_getreg_b32 s33, hwreg(HW_REG_WAVE_HW_ID2, 8, 2)
+; DAGISEL-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_4) | instid1(SALU_CYCLE_1)
+; DAGISEL-NEXT:    s_cmp_lg_u32 0, s33
+; DAGISEL-NEXT:    s_cmovk_i32 s33, 0x1c0
 ; DAGISEL-NEXT:    s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1
 ; DAGISEL-NEXT:    s_alloc_vgpr 45
 ; DAGISEL-NEXT:    s_cselect_b32 s0, -1, 0
-; DAGISEL-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
 ; DAGISEL-NEXT:    v_cndmask_b32_e64 v2, 0, 1, s0
 ; DAGISEL-NEXT:    global_store_b32 v[0:1], v2, off
+; DAGISEL-NEXT:    s_alloc_vgpr 0
 ; DAGISEL-NEXT:    s_endpgm
 entry:
-  %scc = call i1 @llvm.amdgcn.s.alloc.vgpr(i32 45)
+  %scc = call i1 @llvm.amdgcn.s.alloc.vgpr(i32 inreg 45)
   %sel = select i1 %scc, i32 1, i32 0
   store i32 %sel, ptr addrspace(1) %out
   ret void
@@ -35,29 +44,80 @@ entry:
 define amdgpu_cs void @test_alloc_vreg_var(i32 inreg %n, ptr addrspace(1) %out) #0 {
 ; GISEL-LABEL: test_alloc_vreg_var:
 ; GISEL:       ; %bb.0: ; %entry
+; GISEL-NEXT:    s_getreg_b32 s33, hwreg(HW_REG_WAVE_HW_ID2, 8, 2)
+; GISEL-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_4) | instid1(SALU_CYCLE_1)
+; GISEL-NEXT:    s_cmp_lg_u32 0, s33
+; GISEL-NEXT:    s_cmovk_i32 s33, 0x1c0
 ; GISEL-NEXT:    s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1
 ; GISEL-NEXT:    s_alloc_vgpr s0
 ; GISEL-NEXT:    s_cselect_b32 s0, 1, 0
-; GISEL-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1)
 ; GISEL-NEXT:    s_and_b32 s0, s0, 1
+; GISEL-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
 ; GISEL-NEXT:    v_mov_b32_e32 v2, s0
 ; GISEL-NEXT:    global_store_b32 v[0:1], v2, off
+; GISEL-NEXT:    s_alloc_vgpr 0
 ; GISEL-NEXT:    s_endpgm
 ;
 ; DAGISEL-LABEL: test_alloc_vreg_var:
 ; DAGISEL:       ; %bb.0: ; %entry
+; DAGISEL-NEXT:    s_getreg_b32 s33, hwreg(HW_REG_WAVE_HW_ID2, 8, 2)
+; DAGISEL-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_4) | instid1(SALU_CYCLE_1)
+; DAGISEL-NEXT:    s_cmp_lg_u32 0, s33
+; DAGISEL-NEXT:    s_cmovk_i32 s33, 0x1c0
 ; DAGISEL-NEXT:    s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1
 ; DAGISEL-NEXT:    s_alloc_vgpr s0
 ; DAGISEL-NEXT:    s_cselect_b32 s0, -1, 0
-; DAGISEL-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
 ; DAGISEL-NEXT:    v_cndmask_b32_e64 v2, 0, 1, s0
 ; DAGISEL-NEXT:    global_store_b32 v[0:1], v2, off
+; DAGISEL-NEXT:    s_alloc_vgpr 0
+; DAGISEL-NEXT:    s_endpgm
+entry:
+  %scc = call i1 @llvm.amdgcn.s.alloc.vgpr(i32 inreg %n)
+  %sel = select i1 %scc, i32 1, i32 0
+  store i32 %sel, ptr addrspace(1) %out
+  ret void
+}
+
+define amdgpu_cs void @test_alloc_vreg_vgpr(i32 %n, ptr addrspace(1) %out) #0 {
+; GISEL-LABEL: test_alloc_vreg_vgpr:
+; GISEL:       ; %bb.0: ; %entry
+; GISEL-NEXT:    s_getreg_b32 s33, hwreg(HW_REG_WAVE_HW_ID2, 8, 2)
+; GISEL-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
+; GISEL-NEXT:    s_cmp_lg_u32 0, s33
+; GISEL-NEXT:    s_cmovk_i32 s33, 0x1c0
+; GISEL-NEXT:    s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1
+; GISEL-NEXT:    v_readfirstlane_b32 s0, v0
+; GISEL-NEXT:    v_dual_mov_b32 v4, v1 :: v_dual_mov_b32 v5, v2
+; GISEL-NEXT:    s_alloc_vgpr s0
+; GISEL-NEXT:    s_cselect_b32 s0, 1, 0
+; GISEL-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1)
+; GISEL-NEXT:    s_and_b32 s0, s0, 1
+; GISEL-NEXT:    v_mov_b32_e32 v0, s0
+; GISEL-NEXT:    global_store_b32 v[4:5], v0, off
+; GISEL-NEXT:    s_alloc_vgpr 0
+; GISEL-NEXT:    s_endpgm
+;
+; DAGISEL-LABEL: test_alloc_vreg_vgpr:
+; DAGISEL:       ; %bb.0: ; %entry
+; DAGISEL-NEXT:    s_getreg_b32 s33, hwreg(HW_REG_WAVE_HW_ID2, 8, 2)
+; DAGISEL-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
+; DAGISEL-NEXT:    s_cmp_lg_u32 0, s33
+; DAGISEL-NEXT:    s_cmovk_i32 s33, 0x1c0
+; DAGISEL-NEXT:    s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1
+; DAGISEL-NEXT:    v_readfirstlane_b32 s0, v0
+; DAGISEL-NEXT:    v_dual_mov_b32 v3, v2 :: v_dual_mov_b32 v2, v1
+; DAGISEL-NEXT:    s_alloc_vgpr s0
+; DAGISEL-NEXT:    s_cselect_b32 s0, -1, 0
+; DAGISEL-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
+; DAGISEL-NEXT:    v_cndmask_b32_e64 v0, 0, 1, s0
+; DAGISEL-NEXT:    global_store_b32 v[2:3], v0, off
+; DAGISEL-NEXT:    s_alloc_vgpr 0
 ; DAGISEL-NEXT:    s_endpgm
 entry:
-  %scc = call i1 @llvm.amdgcn.s.alloc.vgpr(i32 %n)
+  %scc = call i1 @llvm.amdgcn.s.alloc.vgpr(i32 inreg %n)
   %sel = select i1 %scc, i32 1, i32 0
   store i32 %sel, ptr addrspace(1) %out
   ret void
 }
 
-attributes #0 = { "amdgpu-dynamic-vgpr-block-sze" = "16" }
+attributes #0 = { "amdgpu-dynamic-vgpr-block-size" = "16" }

>From 2b8cd809f621c98e6b07143cbffabc9a128bc865 Mon Sep 17 00:00:00 2001
From: Diana Picus <diana-magda.picus at amd.com>
Date: Mon, 9 Feb 2026 12:07:48 +0100
Subject: [PATCH 5/6] Update after upstream changes

---
 llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp b/llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp
index 37449dc47a6d4..c1af146e2cddd 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp
@@ -2414,8 +2414,8 @@ bool AMDGPUInstructionSelector::selectG_INTRINSIC_W_SIDE_EFFECTS(
     (void)BuildMI(*MBB, &I, DL, TII.get(AMDGPU::COPY), ResReg)
         .addReg(AMDGPU::SCC);
     I.eraseFromParent();
-    return constrainSelectedInstRegOperands(*AllocMI, TII, TRI, RBI) &&
-           RBI.constrainGenericRegister(ResReg, AMDGPU::SReg_32RegClass, *MRI);
+    constrainSelectedInstRegOperands(*AllocMI, TII, TRI, RBI);
+    return RBI.constrainGenericRegister(ResReg, AMDGPU::SReg_32RegClass, *MRI);
   }
   case Intrinsic::amdgcn_s_barrier_init:
   case Intrinsic::amdgcn_s_barrier_signal_var:

>From b169123464d9e215fa621d1bb21cb6bf023f6384 Mon Sep 17 00:00:00 2001
From: Diana Picus <diana-magda.picus at amd.com>
Date: Mon, 9 Feb 2026 12:08:02 +0100
Subject: [PATCH 6/6] Address review comments

---
 llvm/include/llvm/IR/IntrinsicsAMDGPU.td                  | 8 +++++---
 llvm/test/CodeGen/AMDGPU/intrinsic-amdgcn-s-alloc-vgpr.ll | 6 +++---
 2 files changed, 8 insertions(+), 6 deletions(-)

diff --git a/llvm/include/llvm/IR/IntrinsicsAMDGPU.td b/llvm/include/llvm/IR/IntrinsicsAMDGPU.td
index 42bfec2a35678..7a627119cc916 100644
--- a/llvm/include/llvm/IR/IntrinsicsAMDGPU.td
+++ b/llvm/include/llvm/IR/IntrinsicsAMDGPU.td
@@ -402,11 +402,13 @@ def int_amdgcn_s_wait_storecnt       : AMDGPUWaitIntrinsic;
 // It is the responsibility of the calling code to ensure it does not allocate
 // below the VGPR requirements of the current shader. This intrinsic is only
 // available on targets that support dynamic VGPR mode.
-def int_amdgcn_s_alloc_vgpr :
-  DefaultAttrsIntrinsic<
+def int_amdgcn_s_alloc_vgpr : DefaultAttrsIntrinsic<
     [llvm_i1_ty], // Returns true if the allocation succeeded, false otherwise.
     [llvm_i32_ty], // The number of VGPRs to allocate.
-    [NoUndef<RetIndex>, IntrNoMem, IntrHasSideEffects, IntrConvergent]>;
+    [NoUndef<RetIndex>, NoUndef<ArgIndex<0>>,
+     IntrNoMem, IntrHasSideEffects, IntrConvergent
+    ]
+>;
 
 def int_amdgcn_div_scale : PureIntrinsic<
   // 1st parameter: Numerator
diff --git a/llvm/test/CodeGen/AMDGPU/intrinsic-amdgcn-s-alloc-vgpr.ll b/llvm/test/CodeGen/AMDGPU/intrinsic-amdgcn-s-alloc-vgpr.ll
index 49426872ed580..2a7f93f68b8f7 100644
--- a/llvm/test/CodeGen/AMDGPU/intrinsic-amdgcn-s-alloc-vgpr.ll
+++ b/llvm/test/CodeGen/AMDGPU/intrinsic-amdgcn-s-alloc-vgpr.ll
@@ -35,7 +35,7 @@ define amdgpu_cs void @test_alloc_vreg_const(ptr addrspace(1) %out) #0 {
 ; DAGISEL-NEXT:    s_alloc_vgpr 0
 ; DAGISEL-NEXT:    s_endpgm
 entry:
-  %scc = call i1 @llvm.amdgcn.s.alloc.vgpr(i32 inreg 45)
+  %scc = call i1 @llvm.amdgcn.s.alloc.vgpr(i32 45)
   %sel = select i1 %scc, i32 1, i32 0
   store i32 %sel, ptr addrspace(1) %out
   ret void
@@ -72,7 +72,7 @@ define amdgpu_cs void @test_alloc_vreg_var(i32 inreg %n, ptr addrspace(1) %out)
 ; DAGISEL-NEXT:    s_alloc_vgpr 0
 ; DAGISEL-NEXT:    s_endpgm
 entry:
-  %scc = call i1 @llvm.amdgcn.s.alloc.vgpr(i32 inreg %n)
+  %scc = call i1 @llvm.amdgcn.s.alloc.vgpr(i32 %n)
   %sel = select i1 %scc, i32 1, i32 0
   store i32 %sel, ptr addrspace(1) %out
   ret void
@@ -114,7 +114,7 @@ define amdgpu_cs void @test_alloc_vreg_vgpr(i32 %n, ptr addrspace(1) %out) #0 {
 ; DAGISEL-NEXT:    s_alloc_vgpr 0
 ; DAGISEL-NEXT:    s_endpgm
 entry:
-  %scc = call i1 @llvm.amdgcn.s.alloc.vgpr(i32 inreg %n)
+  %scc = call i1 @llvm.amdgcn.s.alloc.vgpr(i32 %n)
   %sel = select i1 %scc, i32 1, i32 0
   store i32 %sel, ptr addrspace(1) %out
   ret void



More information about the llvm-commits mailing list