AMDGPU: Drop legacy r600.read.global.size intrinsics from amdgcn

[arsenm]

These ancient intrinsics were still consumed by the backend for libclc,
which no longer uses them.

[arsenm]

• AMDGPU: Drop legacy r600.read.global.size intrinsics from amdgcn #128700 👈 (View in Graphite)
• main

This stack of pull requests is managed by Graphite. Learn more about stacking.

[llvmbot]

@llvm/pr-subscribers-backend-amdgpu

Author: Matt Arsenault (arsenm)

Changes

These ancient intrinsics were still consumed by the backend for libclc,
which no longer uses them.

---

Full diff: https://github.com/llvm/llvm-project/pull/128700.diff

3 Files Affected:

• (modified) llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp (+2-7)
• (modified) llvm/lib/Target/AMDGPU/SIISelLowering.cpp (-21)
• (modified) llvm/test/CodeGen/AMDGPU/amdgpu.work-item-intrinsics.deprecated.ll (-49)

diff --git a/llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp b/llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp
index 649deee346e90..ef8aaf22aded0 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp
@@ -7400,13 +7400,8 @@ bool AMDGPULegalizerInfo::legalizeIntrinsic(LegalizerHelper &Helper,
     return legalizeKernargMemParameter(MI, B,  SI::KernelInputOffsets::LOCAL_SIZE_Y);
     // TODO: Could insert G_ASSERT_ZEXT from s16
   case Intrinsic::r600_read_local_size_z:
-    return legalizeKernargMemParameter(MI, B, SI::KernelInputOffsets::LOCAL_SIZE_Z);
-  case Intrinsic::r600_read_global_size_x:
-    return legalizeKernargMemParameter(MI, B, SI::KernelInputOffsets::GLOBAL_SIZE_X);
-  case Intrinsic::r600_read_global_size_y:
-    return legalizeKernargMemParameter(MI, B, SI::KernelInputOffsets::GLOBAL_SIZE_Y);
-  case Intrinsic::r600_read_global_size_z:
-    return legalizeKernargMemParameter(MI, B, SI::KernelInputOffsets::GLOBAL_SIZE_Z);
+    return legalizeKernargMemParameter(MI, B,
+                                       SI::KernelInputOffsets::LOCAL_SIZE_Z);
   case Intrinsic::amdgcn_fdiv_fast:
     return legalizeFDIVFastIntrin(MI, MRI, B);
   case Intrinsic::amdgcn_is_shared:
diff --git a/llvm/lib/Target/AMDGPU/SIISelLowering.cpp b/llvm/lib/Target/AMDGPU/SIISelLowering.cpp
index 6ed09253c51e1..be7cdde802b51 100644
--- a/llvm/lib/Target/AMDGPU/SIISelLowering.cpp
+++ b/llvm/lib/Target/AMDGPU/SIISelLowering.cpp
@@ -8770,27 +8770,6 @@ SDValue SITargetLowering::LowerINTRINSIC_WO_CHAIN(SDValue Op,
     return lowerKernargMemParameter(DAG, VT, VT, DL, DAG.getEntryNode(),
                                     SI::KernelInputOffsets::NGROUPS_Z, Align(4),
                                     false);
-  case Intrinsic::r600_read_global_size_x:
-    if (Subtarget->isAmdHsaOS())
-      return emitNonHSAIntrinsicError(DAG, DL, VT);
-
-    return lowerKernargMemParameter(DAG, VT, VT, DL, DAG.getEntryNode(),
-                                    SI::KernelInputOffsets::GLOBAL_SIZE_X,
-                                    Align(4), false);
-  case Intrinsic::r600_read_global_size_y:
-    if (Subtarget->isAmdHsaOS())
-      return emitNonHSAIntrinsicError(DAG, DL, VT);
-
-    return lowerKernargMemParameter(DAG, VT, VT, DL, DAG.getEntryNode(),
-                                    SI::KernelInputOffsets::GLOBAL_SIZE_Y,
-                                    Align(4), false);
-  case Intrinsic::r600_read_global_size_z:
-    if (Subtarget->isAmdHsaOS())
-      return emitNonHSAIntrinsicError(DAG, DL, VT);
-
-    return lowerKernargMemParameter(DAG, VT, VT, DL, DAG.getEntryNode(),
-                                    SI::KernelInputOffsets::GLOBAL_SIZE_Z,
-                                    Align(4), false);
   case Intrinsic::r600_read_local_size_x:
     if (Subtarget->isAmdHsaOS())
       return emitNonHSAIntrinsicError(DAG, DL, VT);
diff --git a/llvm/test/CodeGen/AMDGPU/amdgpu.work-item-intrinsics.deprecated.ll b/llvm/test/CodeGen/AMDGPU/amdgpu.work-item-intrinsics.deprecated.ll
index ad646a36a7443..e2510bb976be7 100644
--- a/llvm/test/CodeGen/AMDGPU/amdgpu.work-item-intrinsics.deprecated.ll
+++ b/llvm/test/CodeGen/AMDGPU/amdgpu.work-item-intrinsics.deprecated.ll
@@ -53,51 +53,6 @@ entry:
   ret void
 }
 
-; FUNC-LABEL: {{^}}global_size_x:
-; SI-NOHSA: s_load_dword [[VAL:s[0-9]+]], s[4:5], 0x3
-; VI-NOHSA: s_load_dword [[VAL:s[0-9]+]], s[4:5], 0xc
-; GCN-NOHSA: v_mov_b32_e32 [[VVAL:v[0-9]+]], [[VAL]]
-; GCN-NOHSA: buffer_store_dword [[VVAL]]
-
-; EG: MEM_RAT_CACHELESS STORE_RAW [[VAL:T[0-9]+\.X]]
-; EG: MOV {{\*? *}}[[VAL]], KC0[0].W
-define amdgpu_kernel void @global_size_x (ptr addrspace(1) %out) {
-entry:
-  %0 = call i32 @llvm.r600.read.global.size.x() #0
-  store i32 %0, ptr addrspace(1) %out
-  ret void
-}
-
-; FUNC-LABEL: {{^}}global_size_y:
-; SI-NOHSA: s_load_dword [[VAL:s[0-9]+]], s[4:5], 0x4
-; VI-NOHSA: s_load_dword [[VAL:s[0-9]+]], s[4:5], 0x10
-; GCN-NOHSA: v_mov_b32_e32 [[VVAL:v[0-9]+]], [[VAL]]
-; GCN-NOHSA: buffer_store_dword [[VVAL]]
-
-; EG: MEM_RAT_CACHELESS STORE_RAW [[VAL:T[0-9]+\.X]]
-; EG: MOV {{\*? *}}[[VAL]], KC0[1].X
-define amdgpu_kernel void @global_size_y (ptr addrspace(1) %out) {
-entry:
-  %0 = call i32 @llvm.r600.read.global.size.y() #0
-  store i32 %0, ptr addrspace(1) %out
-  ret void
-}
-
-; FUNC-LABEL: {{^}}global_size_z:
-; SI-NOHSA: s_load_dword [[VAL:s[0-9]+]], s[4:5], 0x5
-; VI-NOHSA: s_load_dword [[VAL:s[0-9]+]], s[4:5], 0x14
-; GCN-NOHSA: v_mov_b32_e32 [[VVAL:v[0-9]+]], [[VAL]]
-; GCN-NOHSA: buffer_store_dword [[VVAL]]
-
-; EG: MEM_RAT_CACHELESS STORE_RAW [[VAL:T[0-9]+\.X]]
-; EG: MOV {{\*? *}}[[VAL]], KC0[1].Y
-define amdgpu_kernel void @global_size_z (ptr addrspace(1) %out) {
-entry:
-  %0 = call i32 @llvm.r600.read.global.size.z() #0
-  store i32 %0, ptr addrspace(1) %out
-  ret void
-}
-
 ; FUNC-LABEL: {{^}}local_size_x:
 ; SI-NOHSA: s_load_dword [[VAL:s[0-9]+]], s[4:5], 0x6
 ; VI-NOHSA: s_load_dword [[VAL:s[0-9]+]], s[4:5], 0x18
@@ -147,10 +102,6 @@ declare i32 @llvm.r600.read.ngroups.x() #0
 declare i32 @llvm.r600.read.ngroups.y() #0
 declare i32 @llvm.r600.read.ngroups.z() #0
 
-declare i32 @llvm.r600.read.global.size.x() #0
-declare i32 @llvm.r600.read.global.size.y() #0
-declare i32 @llvm.r600.read.global.size.z() #0
-
 declare i32 @llvm.r600.read.local.size.x() #0
 declare i32 @llvm.r600.read.local.size.y() #0
 declare i32 @llvm.r600.read.local.size.z() #0

[jayfoad]

LGTM