Skip to content

AMDGPU: Use getSignedTargetConstant for ImmOffset in SelectScratchSVAddr #121978

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom
Merged
merged 1 commit into from
Jan 7, 2025
Merged

AMDGPU: Use getSignedTargetConstant for ImmOffset in SelectScratchSVAddr #121978

merged 1 commit into from
Jan 7, 2025

Conversation

changpeng
Copy link
Contributor

ImmOffset is signed and we will hit an assert with negative ImmOffset when getTargetConstant is used.

Fixes: SWDEV-506453

@changpeng
AMDGPU: Use getSignedTargetConstant for ImmOffset in SelectScratchSVAddr
2a7f674 
 ImmOffset is signed and we will hit an assert with negative ImmOffset
when getTargetConstant is uded.

Fixes: SWDEV-506453
@llvmbot
Copy link
Member

llvmbot commented Jan 7, 2025

@llvm/pr-subscribers-llvm-globalisel

Author: Changpeng Fang (changpeng)

Changes

ImmOffset is signed and we will hit an assert with negative ImmOffset when getTargetConstant is used.

Fixes: SWDEV-506453

Full diff: https://github.com/llvm/llvm-project/pull/121978.diff

3 Files Affected:

  • (modified) llvm/lib/Target/AMDGPU/AMDGPUISelDAGToDAG.cpp (+1-1)
  • (modified) llvm/test/CodeGen/AMDGPU/GlobalISel/flat-scratch.ll (+110)
  • (modified) llvm/test/CodeGen/AMDGPU/flat-scratch.ll (+108)
diff --git a/llvm/lib/Target/AMDGPU/AMDGPUISelDAGToDAG.cpp b/llvm/lib/Target/AMDGPU/AMDGPUISelDAGToDAG.cpp
index d9eaf82c521404..27e9018d68a03e 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUISelDAGToDAG.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUISelDAGToDAG.cpp
@@ -1997,7 +1997,7 @@ bool AMDGPUDAGToDAGISel::SelectScratchSVAddr(SDNode *N, SDValue Addr,
   if (checkFlatScratchSVSSwizzleBug(VAddr, SAddr, ImmOffset))
     return false;
   SAddr = SelectSAddrFI(CurDAG, SAddr);
-  Offset = CurDAG->getTargetConstant(ImmOffset, SDLoc(), MVT::i32);
+  Offset = CurDAG->getSignedTargetConstant(ImmOffset, SDLoc(), MVT::i32);
   return true;
 }
 
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/flat-scratch.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/flat-scratch.ll
index b0f2aac9a42d45..7cafa2f608a4bc 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/flat-scratch.ll
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/flat-scratch.ll
@@ -3990,6 +3990,116 @@ bb:
   ret void
 }
 
+define amdgpu_gs void @sgpr_base_plus_sgpr_plus_vgpr_plus_negative_imm_offset(ptr addrspace(5) inreg %sgpr_base, i32 inreg %sidx, i32 %vidx) {
+; GFX9-LABEL: sgpr_base_plus_sgpr_plus_vgpr_plus_negative_imm_offset:
+; GFX9:       ; %bb.0: ; %bb
+; GFX9-NEXT:    s_add_u32 flat_scratch_lo, s0, s5
+; GFX9-NEXT:    v_add_u32_e32 v0, s3, v0
+; GFX9-NEXT:    s_addc_u32 flat_scratch_hi, s1, 0
+; GFX9-NEXT:    v_add3_u32 v0, s2, v0, -16
+; GFX9-NEXT:    v_mov_b32_e32 v1, 15
+; GFX9-NEXT:    scratch_store_dword v0, v1, off
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_endpgm
+;
+; GFX10-LABEL: sgpr_base_plus_sgpr_plus_vgpr_plus_negative_imm_offset:
+; GFX10:       ; %bb.0: ; %bb
+; GFX10-NEXT:    s_add_u32 s0, s0, s5
+; GFX10-NEXT:    s_addc_u32 s1, s1, 0
+; GFX10-NEXT:    s_setreg_b32 hwreg(HW_REG_FLAT_SCR_LO), s0
+; GFX10-NEXT:    s_setreg_b32 hwreg(HW_REG_FLAT_SCR_HI), s1
+; GFX10-NEXT:    v_add_nc_u32_e32 v0, s3, v0
+; GFX10-NEXT:    v_mov_b32_e32 v1, 15
+; GFX10-NEXT:    v_add_nc_u32_e32 v0, s2, v0
+; GFX10-NEXT:    scratch_store_dword v0, v1, off offset:-16
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_endpgm
+;
+; GFX940-LABEL: sgpr_base_plus_sgpr_plus_vgpr_plus_negative_imm_offset:
+; GFX940:       ; %bb.0: ; %bb
+; GFX940-NEXT:    v_add_u32_e32 v0, s1, v0
+; GFX940-NEXT:    v_add3_u32 v0, s0, v0, -16
+; GFX940-NEXT:    v_mov_b32_e32 v1, 15
+; GFX940-NEXT:    scratch_store_dword v0, v1, off sc0 sc1
+; GFX940-NEXT:    s_waitcnt vmcnt(0)
+; GFX940-NEXT:    s_endpgm
+;
+; GFX11-LABEL: sgpr_base_plus_sgpr_plus_vgpr_plus_negative_imm_offset:
+; GFX11:       ; %bb.0: ; %bb
+; GFX11-NEXT:    v_dual_mov_b32 v1, 15 :: v_dual_add_nc_u32 v0, s1, v0
+; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX11-NEXT:    v_add_nc_u32_e32 v0, s0, v0
+; GFX11-NEXT:    scratch_store_b32 v0, v1, off offset:-16 dlc
+; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX11-NEXT:    s_endpgm
+;
+; GFX12-LABEL: sgpr_base_plus_sgpr_plus_vgpr_plus_negative_imm_offset:
+; GFX12:       ; %bb.0: ; %bb
+; GFX12-NEXT:    v_dual_mov_b32 v1, 15 :: v_dual_add_nc_u32 v0, s1, v0
+; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX12-NEXT:    v_add_nc_u32_e32 v0, s0, v0
+; GFX12-NEXT:    scratch_store_b32 v0, v1, off offset:-16 scope:SCOPE_SYS
+; GFX12-NEXT:    s_wait_storecnt 0x0
+; GFX12-NEXT:    s_endpgm
+;
+; UNALIGNED_GFX9-LABEL: sgpr_base_plus_sgpr_plus_vgpr_plus_negative_imm_offset:
+; UNALIGNED_GFX9:       ; %bb.0: ; %bb
+; UNALIGNED_GFX9-NEXT:    s_add_u32 flat_scratch_lo, s0, s5
+; UNALIGNED_GFX9-NEXT:    v_add_u32_e32 v0, s3, v0
+; UNALIGNED_GFX9-NEXT:    s_addc_u32 flat_scratch_hi, s1, 0
+; UNALIGNED_GFX9-NEXT:    v_add3_u32 v0, s2, v0, -16
+; UNALIGNED_GFX9-NEXT:    v_mov_b32_e32 v1, 15
+; UNALIGNED_GFX9-NEXT:    scratch_store_dword v0, v1, off
+; UNALIGNED_GFX9-NEXT:    s_waitcnt vmcnt(0)
+; UNALIGNED_GFX9-NEXT:    s_endpgm
+;
+; UNALIGNED_GFX10-LABEL: sgpr_base_plus_sgpr_plus_vgpr_plus_negative_imm_offset:
+; UNALIGNED_GFX10:       ; %bb.0: ; %bb
+; UNALIGNED_GFX10-NEXT:    s_add_u32 s0, s0, s5
+; UNALIGNED_GFX10-NEXT:    s_addc_u32 s1, s1, 0
+; UNALIGNED_GFX10-NEXT:    s_setreg_b32 hwreg(HW_REG_FLAT_SCR_LO), s0
+; UNALIGNED_GFX10-NEXT:    s_setreg_b32 hwreg(HW_REG_FLAT_SCR_HI), s1
+; UNALIGNED_GFX10-NEXT:    v_add_nc_u32_e32 v0, s3, v0
+; UNALIGNED_GFX10-NEXT:    v_mov_b32_e32 v1, 15
+; UNALIGNED_GFX10-NEXT:    v_add_nc_u32_e32 v0, s2, v0
+; UNALIGNED_GFX10-NEXT:    scratch_store_dword v0, v1, off offset:-16
+; UNALIGNED_GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; UNALIGNED_GFX10-NEXT:    s_endpgm
+;
+; UNALIGNED_GFX940-LABEL: sgpr_base_plus_sgpr_plus_vgpr_plus_negative_imm_offset:
+; UNALIGNED_GFX940:       ; %bb.0: ; %bb
+; UNALIGNED_GFX940-NEXT:    v_add_u32_e32 v0, s1, v0
+; UNALIGNED_GFX940-NEXT:    v_add3_u32 v0, s0, v0, -16
+; UNALIGNED_GFX940-NEXT:    v_mov_b32_e32 v1, 15
+; UNALIGNED_GFX940-NEXT:    scratch_store_dword v0, v1, off sc0 sc1
+; UNALIGNED_GFX940-NEXT:    s_waitcnt vmcnt(0)
+; UNALIGNED_GFX940-NEXT:    s_endpgm
+;
+; UNALIGNED_GFX11-LABEL: sgpr_base_plus_sgpr_plus_vgpr_plus_negative_imm_offset:
+; UNALIGNED_GFX11:       ; %bb.0: ; %bb
+; UNALIGNED_GFX11-NEXT:    v_dual_mov_b32 v1, 15 :: v_dual_add_nc_u32 v0, s1, v0
+; UNALIGNED_GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; UNALIGNED_GFX11-NEXT:    v_add_nc_u32_e32 v0, s0, v0
+; UNALIGNED_GFX11-NEXT:    scratch_store_b32 v0, v1, off offset:-16 dlc
+; UNALIGNED_GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
+; UNALIGNED_GFX11-NEXT:    s_endpgm
+;
+; UNALIGNED_GFX12-LABEL: sgpr_base_plus_sgpr_plus_vgpr_plus_negative_imm_offset:
+; UNALIGNED_GFX12:       ; %bb.0: ; %bb
+; UNALIGNED_GFX12-NEXT:    v_dual_mov_b32 v1, 15 :: v_dual_add_nc_u32 v0, s1, v0
+; UNALIGNED_GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; UNALIGNED_GFX12-NEXT:    v_add_nc_u32_e32 v0, s0, v0
+; UNALIGNED_GFX12-NEXT:    scratch_store_b32 v0, v1, off offset:-16 scope:SCOPE_SYS
+; UNALIGNED_GFX12-NEXT:    s_wait_storecnt 0x0
+; UNALIGNED_GFX12-NEXT:    s_endpgm
+bb:
+  %add1 = add nsw i32 %sidx, %vidx
+  %add2 = add nsw i32 %add1, -16
+  %gep = getelementptr inbounds [16 x i8], ptr addrspace(5) %sgpr_base, i32 0, i32 %add2
+  store volatile i32 15, ptr addrspace(5) %gep, align 4
+  ret void
+}
+
 define amdgpu_gs void @sgpr_base_negative_offset(ptr addrspace(1) %out, ptr addrspace(5) inreg %scevgep) {
 ; GFX9-LABEL: sgpr_base_negative_offset:
 ; GFX9:       ; %bb.0: ; %entry
diff --git a/llvm/test/CodeGen/AMDGPU/flat-scratch.ll b/llvm/test/CodeGen/AMDGPU/flat-scratch.ll
index 97d642b991f705..5415af02ef89ca 100644
--- a/llvm/test/CodeGen/AMDGPU/flat-scratch.ll
+++ b/llvm/test/CodeGen/AMDGPU/flat-scratch.ll
@@ -5249,6 +5249,114 @@ bb:
   ret void
 }
 
+define amdgpu_gs void @sgpr_base_plus_sgpr_plus_vgpr_plus_negative_imm_offset(ptr addrspace(5) inreg %sgpr_base, i32 inreg %sidx, i32 %vidx) {
+; GFX9-LABEL: sgpr_base_plus_sgpr_plus_vgpr_plus_negative_imm_offset:
+; GFX9:       ; %bb.0: ; %bb
+; GFX9-NEXT:    s_add_u32 flat_scratch_lo, s0, s5
+; GFX9-NEXT:    s_addc_u32 flat_scratch_hi, s1, 0
+; GFX9-NEXT:    s_add_i32 s2, s2, s3
+; GFX9-NEXT:    v_add_u32_e32 v0, s2, v0
+; GFX9-NEXT:    v_add_u32_e32 v0, -16, v0
+; GFX9-NEXT:    v_mov_b32_e32 v1, 15
+; GFX9-NEXT:    scratch_store_dword v0, v1, off
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_endpgm
+;
+; GFX10-LABEL: sgpr_base_plus_sgpr_plus_vgpr_plus_negative_imm_offset:
+; GFX10:       ; %bb.0: ; %bb
+; GFX10-NEXT:    s_add_u32 s0, s0, s5
+; GFX10-NEXT:    s_addc_u32 s1, s1, 0
+; GFX10-NEXT:    s_setreg_b32 hwreg(HW_REG_FLAT_SCR_LO), s0
+; GFX10-NEXT:    s_setreg_b32 hwreg(HW_REG_FLAT_SCR_HI), s1
+; GFX10-NEXT:    v_add3_u32 v0, s2, s3, v0
+; GFX10-NEXT:    v_mov_b32_e32 v1, 15
+; GFX10-NEXT:    scratch_store_dword v0, v1, off offset:-16
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_endpgm
+;
+; GFX11-LABEL: sgpr_base_plus_sgpr_plus_vgpr_plus_negative_imm_offset:
+; GFX11:       ; %bb.0: ; %bb
+; GFX11-NEXT:    v_add3_u32 v0, s0, s1, v0
+; GFX11-NEXT:    v_mov_b32_e32 v1, 15
+; GFX11-NEXT:    scratch_store_b32 v0, v1, off offset:-16 dlc
+; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX11-NEXT:    s_endpgm
+;
+; GFX12-LABEL: sgpr_base_plus_sgpr_plus_vgpr_plus_negative_imm_offset:
+; GFX12:       ; %bb.0: ; %bb
+; GFX12-NEXT:    v_mov_b32_e32 v1, 15
+; GFX12-NEXT:    s_add_co_i32 s0, s0, s1
+; GFX12-NEXT:    scratch_store_b32 v0, v1, s0 offset:-16 scope:SCOPE_SYS
+; GFX12-NEXT:    s_wait_storecnt 0x0
+; GFX12-NEXT:    s_endpgm
+;
+; GFX9-PAL-LABEL: sgpr_base_plus_sgpr_plus_vgpr_plus_negative_imm_offset:
+; GFX9-PAL:       ; %bb.0: ; %bb
+; GFX9-PAL-NEXT:    s_getpc_b64 s[2:3]
+; GFX9-PAL-NEXT:    s_mov_b32 s2, s8
+; GFX9-PAL-NEXT:    s_load_dwordx2 s[2:3], s[2:3], 0x0
+; GFX9-PAL-NEXT:    v_mov_b32_e32 v1, 15
+; GFX9-PAL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-PAL-NEXT:    s_and_b32 s3, s3, 0xffff
+; GFX9-PAL-NEXT:    s_add_u32 flat_scratch_lo, s2, s5
+; GFX9-PAL-NEXT:    s_addc_u32 flat_scratch_hi, s3, 0
+; GFX9-PAL-NEXT:    s_add_i32 s0, s0, s1
+; GFX9-PAL-NEXT:    v_add_u32_e32 v0, s0, v0
+; GFX9-PAL-NEXT:    v_add_u32_e32 v0, -16, v0
+; GFX9-PAL-NEXT:    scratch_store_dword v0, v1, off
+; GFX9-PAL-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-PAL-NEXT:    s_endpgm
+;
+; GFX940-LABEL: sgpr_base_plus_sgpr_plus_vgpr_plus_negative_imm_offset:
+; GFX940:       ; %bb.0: ; %bb
+; GFX940-NEXT:    s_add_i32 s0, s0, s1
+; GFX940-NEXT:    v_add_u32_e32 v0, s0, v0
+; GFX940-NEXT:    v_add_u32_e32 v0, -16, v0
+; GFX940-NEXT:    v_mov_b32_e32 v1, 15
+; GFX940-NEXT:    scratch_store_dword v0, v1, off sc0 sc1
+; GFX940-NEXT:    s_waitcnt vmcnt(0)
+; GFX940-NEXT:    s_endpgm
+;
+; GFX10-PAL-LABEL: sgpr_base_plus_sgpr_plus_vgpr_plus_negative_imm_offset:
+; GFX10-PAL:       ; %bb.0: ; %bb
+; GFX10-PAL-NEXT:    s_getpc_b64 s[2:3]
+; GFX10-PAL-NEXT:    s_mov_b32 s2, s8
+; GFX10-PAL-NEXT:    s_load_dwordx2 s[2:3], s[2:3], 0x0
+; GFX10-PAL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX10-PAL-NEXT:    s_and_b32 s3, s3, 0xffff
+; GFX10-PAL-NEXT:    s_add_u32 s2, s2, s5
+; GFX10-PAL-NEXT:    s_addc_u32 s3, s3, 0
+; GFX10-PAL-NEXT:    s_setreg_b32 hwreg(HW_REG_FLAT_SCR_LO), s2
+; GFX10-PAL-NEXT:    s_setreg_b32 hwreg(HW_REG_FLAT_SCR_HI), s3
+; GFX10-PAL-NEXT:    v_add3_u32 v0, s0, s1, v0
+; GFX10-PAL-NEXT:    v_mov_b32_e32 v1, 15
+; GFX10-PAL-NEXT:    scratch_store_dword v0, v1, off offset:-16
+; GFX10-PAL-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-PAL-NEXT:    s_endpgm
+;
+; GFX11-PAL-LABEL: sgpr_base_plus_sgpr_plus_vgpr_plus_negative_imm_offset:
+; GFX11-PAL:       ; %bb.0: ; %bb
+; GFX11-PAL-NEXT:    v_add3_u32 v0, s0, s1, v0
+; GFX11-PAL-NEXT:    v_mov_b32_e32 v1, 15
+; GFX11-PAL-NEXT:    scratch_store_b32 v0, v1, off offset:-16 dlc
+; GFX11-PAL-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX11-PAL-NEXT:    s_endpgm
+;
+; GFX12-PAL-LABEL: sgpr_base_plus_sgpr_plus_vgpr_plus_negative_imm_offset:
+; GFX12-PAL:       ; %bb.0: ; %bb
+; GFX12-PAL-NEXT:    v_mov_b32_e32 v1, 15
+; GFX12-PAL-NEXT:    s_add_co_i32 s0, s0, s1
+; GFX12-PAL-NEXT:    scratch_store_b32 v0, v1, s0 offset:-16 scope:SCOPE_SYS
+; GFX12-PAL-NEXT:    s_wait_storecnt 0x0
+; GFX12-PAL-NEXT:    s_endpgm
+bb:
+  %add1 = add nsw i32 %sidx, %vidx
+  %add2 = add nsw i32 %add1, -16
+  %gep = getelementptr inbounds [16 x i8], ptr addrspace(5) %sgpr_base, i32 0, i32 %add2
+  store volatile i32 15, ptr addrspace(5) %gep, align 4
+  ret void
+}
+
 define amdgpu_gs void @sgpr_base_negative_offset(ptr addrspace(1) %out, ptr addrspace(5) inreg %scevgep) {
 ; GFX9-LABEL: sgpr_base_negative_offset:
 ; GFX9:       ; %bb.0: ; %entry

@llvmbot
Copy link
Member

llvmbot commented Jan 7, 2025

@llvm/pr-subscribers-backend-amdgpu

Author: Changpeng Fang (changpeng)

Changes

ImmOffset is signed and we will hit an assert with negative ImmOffset when getTargetConstant is used.

Fixes: SWDEV-506453

Full diff: https://github.com/llvm/llvm-project/pull/121978.diff

3 Files Affected:

  • (modified) llvm/lib/Target/AMDGPU/AMDGPUISelDAGToDAG.cpp (+1-1)
  • (modified) llvm/test/CodeGen/AMDGPU/GlobalISel/flat-scratch.ll (+110)
  • (modified) llvm/test/CodeGen/AMDGPU/flat-scratch.ll (+108)
diff --git a/llvm/lib/Target/AMDGPU/AMDGPUISelDAGToDAG.cpp b/llvm/lib/Target/AMDGPU/AMDGPUISelDAGToDAG.cpp
index d9eaf82c521404..27e9018d68a03e 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUISelDAGToDAG.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUISelDAGToDAG.cpp
@@ -1997,7 +1997,7 @@ bool AMDGPUDAGToDAGISel::SelectScratchSVAddr(SDNode *N, SDValue Addr,
   if (checkFlatScratchSVSSwizzleBug(VAddr, SAddr, ImmOffset))
     return false;
   SAddr = SelectSAddrFI(CurDAG, SAddr);
-  Offset = CurDAG->getTargetConstant(ImmOffset, SDLoc(), MVT::i32);
+  Offset = CurDAG->getSignedTargetConstant(ImmOffset, SDLoc(), MVT::i32);
   return true;
 }
 
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/flat-scratch.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/flat-scratch.ll
index b0f2aac9a42d45..7cafa2f608a4bc 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/flat-scratch.ll
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/flat-scratch.ll
@@ -3990,6 +3990,116 @@ bb:
   ret void
 }
 
+define amdgpu_gs void @sgpr_base_plus_sgpr_plus_vgpr_plus_negative_imm_offset(ptr addrspace(5) inreg %sgpr_base, i32 inreg %sidx, i32 %vidx) {
+; GFX9-LABEL: sgpr_base_plus_sgpr_plus_vgpr_plus_negative_imm_offset:
+; GFX9:       ; %bb.0: ; %bb
+; GFX9-NEXT:    s_add_u32 flat_scratch_lo, s0, s5
+; GFX9-NEXT:    v_add_u32_e32 v0, s3, v0
+; GFX9-NEXT:    s_addc_u32 flat_scratch_hi, s1, 0
+; GFX9-NEXT:    v_add3_u32 v0, s2, v0, -16
+; GFX9-NEXT:    v_mov_b32_e32 v1, 15
+; GFX9-NEXT:    scratch_store_dword v0, v1, off
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_endpgm
+;
+; GFX10-LABEL: sgpr_base_plus_sgpr_plus_vgpr_plus_negative_imm_offset:
+; GFX10:       ; %bb.0: ; %bb
+; GFX10-NEXT:    s_add_u32 s0, s0, s5
+; GFX10-NEXT:    s_addc_u32 s1, s1, 0
+; GFX10-NEXT:    s_setreg_b32 hwreg(HW_REG_FLAT_SCR_LO), s0
+; GFX10-NEXT:    s_setreg_b32 hwreg(HW_REG_FLAT_SCR_HI), s1
+; GFX10-NEXT:    v_add_nc_u32_e32 v0, s3, v0
+; GFX10-NEXT:    v_mov_b32_e32 v1, 15
+; GFX10-NEXT:    v_add_nc_u32_e32 v0, s2, v0
+; GFX10-NEXT:    scratch_store_dword v0, v1, off offset:-16
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_endpgm
+;
+; GFX940-LABEL: sgpr_base_plus_sgpr_plus_vgpr_plus_negative_imm_offset:
+; GFX940:       ; %bb.0: ; %bb
+; GFX940-NEXT:    v_add_u32_e32 v0, s1, v0
+; GFX940-NEXT:    v_add3_u32 v0, s0, v0, -16
+; GFX940-NEXT:    v_mov_b32_e32 v1, 15
+; GFX940-NEXT:    scratch_store_dword v0, v1, off sc0 sc1
+; GFX940-NEXT:    s_waitcnt vmcnt(0)
+; GFX940-NEXT:    s_endpgm
+;
+; GFX11-LABEL: sgpr_base_plus_sgpr_plus_vgpr_plus_negative_imm_offset:
+; GFX11:       ; %bb.0: ; %bb
+; GFX11-NEXT:    v_dual_mov_b32 v1, 15 :: v_dual_add_nc_u32 v0, s1, v0
+; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX11-NEXT:    v_add_nc_u32_e32 v0, s0, v0
+; GFX11-NEXT:    scratch_store_b32 v0, v1, off offset:-16 dlc
+; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX11-NEXT:    s_endpgm
+;
+; GFX12-LABEL: sgpr_base_plus_sgpr_plus_vgpr_plus_negative_imm_offset:
+; GFX12:       ; %bb.0: ; %bb
+; GFX12-NEXT:    v_dual_mov_b32 v1, 15 :: v_dual_add_nc_u32 v0, s1, v0
+; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX12-NEXT:    v_add_nc_u32_e32 v0, s0, v0
+; GFX12-NEXT:    scratch_store_b32 v0, v1, off offset:-16 scope:SCOPE_SYS
+; GFX12-NEXT:    s_wait_storecnt 0x0
+; GFX12-NEXT:    s_endpgm
+;
+; UNALIGNED_GFX9-LABEL: sgpr_base_plus_sgpr_plus_vgpr_plus_negative_imm_offset:
+; UNALIGNED_GFX9:       ; %bb.0: ; %bb
+; UNALIGNED_GFX9-NEXT:    s_add_u32 flat_scratch_lo, s0, s5
+; UNALIGNED_GFX9-NEXT:    v_add_u32_e32 v0, s3, v0
+; UNALIGNED_GFX9-NEXT:    s_addc_u32 flat_scratch_hi, s1, 0
+; UNALIGNED_GFX9-NEXT:    v_add3_u32 v0, s2, v0, -16
+; UNALIGNED_GFX9-NEXT:    v_mov_b32_e32 v1, 15
+; UNALIGNED_GFX9-NEXT:    scratch_store_dword v0, v1, off
+; UNALIGNED_GFX9-NEXT:    s_waitcnt vmcnt(0)
+; UNALIGNED_GFX9-NEXT:    s_endpgm
+;
+; UNALIGNED_GFX10-LABEL: sgpr_base_plus_sgpr_plus_vgpr_plus_negative_imm_offset:
+; UNALIGNED_GFX10:       ; %bb.0: ; %bb
+; UNALIGNED_GFX10-NEXT:    s_add_u32 s0, s0, s5
+; UNALIGNED_GFX10-NEXT:    s_addc_u32 s1, s1, 0
+; UNALIGNED_GFX10-NEXT:    s_setreg_b32 hwreg(HW_REG_FLAT_SCR_LO), s0
+; UNALIGNED_GFX10-NEXT:    s_setreg_b32 hwreg(HW_REG_FLAT_SCR_HI), s1
+; UNALIGNED_GFX10-NEXT:    v_add_nc_u32_e32 v0, s3, v0
+; UNALIGNED_GFX10-NEXT:    v_mov_b32_e32 v1, 15
+; UNALIGNED_GFX10-NEXT:    v_add_nc_u32_e32 v0, s2, v0
+; UNALIGNED_GFX10-NEXT:    scratch_store_dword v0, v1, off offset:-16
+; UNALIGNED_GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; UNALIGNED_GFX10-NEXT:    s_endpgm
+;
+; UNALIGNED_GFX940-LABEL: sgpr_base_plus_sgpr_plus_vgpr_plus_negative_imm_offset:
+; UNALIGNED_GFX940:       ; %bb.0: ; %bb
+; UNALIGNED_GFX940-NEXT:    v_add_u32_e32 v0, s1, v0
+; UNALIGNED_GFX940-NEXT:    v_add3_u32 v0, s0, v0, -16
+; UNALIGNED_GFX940-NEXT:    v_mov_b32_e32 v1, 15
+; UNALIGNED_GFX940-NEXT:    scratch_store_dword v0, v1, off sc0 sc1
+; UNALIGNED_GFX940-NEXT:    s_waitcnt vmcnt(0)
+; UNALIGNED_GFX940-NEXT:    s_endpgm
+;
+; UNALIGNED_GFX11-LABEL: sgpr_base_plus_sgpr_plus_vgpr_plus_negative_imm_offset:
+; UNALIGNED_GFX11:       ; %bb.0: ; %bb
+; UNALIGNED_GFX11-NEXT:    v_dual_mov_b32 v1, 15 :: v_dual_add_nc_u32 v0, s1, v0
+; UNALIGNED_GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; UNALIGNED_GFX11-NEXT:    v_add_nc_u32_e32 v0, s0, v0
+; UNALIGNED_GFX11-NEXT:    scratch_store_b32 v0, v1, off offset:-16 dlc
+; UNALIGNED_GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
+; UNALIGNED_GFX11-NEXT:    s_endpgm
+;
+; UNALIGNED_GFX12-LABEL: sgpr_base_plus_sgpr_plus_vgpr_plus_negative_imm_offset:
+; UNALIGNED_GFX12:       ; %bb.0: ; %bb
+; UNALIGNED_GFX12-NEXT:    v_dual_mov_b32 v1, 15 :: v_dual_add_nc_u32 v0, s1, v0
+; UNALIGNED_GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; UNALIGNED_GFX12-NEXT:    v_add_nc_u32_e32 v0, s0, v0
+; UNALIGNED_GFX12-NEXT:    scratch_store_b32 v0, v1, off offset:-16 scope:SCOPE_SYS
+; UNALIGNED_GFX12-NEXT:    s_wait_storecnt 0x0
+; UNALIGNED_GFX12-NEXT:    s_endpgm
+bb:
+  %add1 = add nsw i32 %sidx, %vidx
+  %add2 = add nsw i32 %add1, -16
+  %gep = getelementptr inbounds [16 x i8], ptr addrspace(5) %sgpr_base, i32 0, i32 %add2
+  store volatile i32 15, ptr addrspace(5) %gep, align 4
+  ret void
+}
+
 define amdgpu_gs void @sgpr_base_negative_offset(ptr addrspace(1) %out, ptr addrspace(5) inreg %scevgep) {
 ; GFX9-LABEL: sgpr_base_negative_offset:
 ; GFX9:       ; %bb.0: ; %entry
diff --git a/llvm/test/CodeGen/AMDGPU/flat-scratch.ll b/llvm/test/CodeGen/AMDGPU/flat-scratch.ll
index 97d642b991f705..5415af02ef89ca 100644
--- a/llvm/test/CodeGen/AMDGPU/flat-scratch.ll
+++ b/llvm/test/CodeGen/AMDGPU/flat-scratch.ll
@@ -5249,6 +5249,114 @@ bb:
   ret void
 }
 
+define amdgpu_gs void @sgpr_base_plus_sgpr_plus_vgpr_plus_negative_imm_offset(ptr addrspace(5) inreg %sgpr_base, i32 inreg %sidx, i32 %vidx) {
+; GFX9-LABEL: sgpr_base_plus_sgpr_plus_vgpr_plus_negative_imm_offset:
+; GFX9:       ; %bb.0: ; %bb
+; GFX9-NEXT:    s_add_u32 flat_scratch_lo, s0, s5
+; GFX9-NEXT:    s_addc_u32 flat_scratch_hi, s1, 0
+; GFX9-NEXT:    s_add_i32 s2, s2, s3
+; GFX9-NEXT:    v_add_u32_e32 v0, s2, v0
+; GFX9-NEXT:    v_add_u32_e32 v0, -16, v0
+; GFX9-NEXT:    v_mov_b32_e32 v1, 15
+; GFX9-NEXT:    scratch_store_dword v0, v1, off
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_endpgm
+;
+; GFX10-LABEL: sgpr_base_plus_sgpr_plus_vgpr_plus_negative_imm_offset:
+; GFX10:       ; %bb.0: ; %bb
+; GFX10-NEXT:    s_add_u32 s0, s0, s5
+; GFX10-NEXT:    s_addc_u32 s1, s1, 0
+; GFX10-NEXT:    s_setreg_b32 hwreg(HW_REG_FLAT_SCR_LO), s0
+; GFX10-NEXT:    s_setreg_b32 hwreg(HW_REG_FLAT_SCR_HI), s1
+; GFX10-NEXT:    v_add3_u32 v0, s2, s3, v0
+; GFX10-NEXT:    v_mov_b32_e32 v1, 15
+; GFX10-NEXT:    scratch_store_dword v0, v1, off offset:-16
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_endpgm
+;
+; GFX11-LABEL: sgpr_base_plus_sgpr_plus_vgpr_plus_negative_imm_offset:
+; GFX11:       ; %bb.0: ; %bb
+; GFX11-NEXT:    v_add3_u32 v0, s0, s1, v0
+; GFX11-NEXT:    v_mov_b32_e32 v1, 15
+; GFX11-NEXT:    scratch_store_b32 v0, v1, off offset:-16 dlc
+; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX11-NEXT:    s_endpgm
+;
+; GFX12-LABEL: sgpr_base_plus_sgpr_plus_vgpr_plus_negative_imm_offset:
+; GFX12:       ; %bb.0: ; %bb
+; GFX12-NEXT:    v_mov_b32_e32 v1, 15
+; GFX12-NEXT:    s_add_co_i32 s0, s0, s1
+; GFX12-NEXT:    scratch_store_b32 v0, v1, s0 offset:-16 scope:SCOPE_SYS
+; GFX12-NEXT:    s_wait_storecnt 0x0
+; GFX12-NEXT:    s_endpgm
+;
+; GFX9-PAL-LABEL: sgpr_base_plus_sgpr_plus_vgpr_plus_negative_imm_offset:
+; GFX9-PAL:       ; %bb.0: ; %bb
+; GFX9-PAL-NEXT:    s_getpc_b64 s[2:3]
+; GFX9-PAL-NEXT:    s_mov_b32 s2, s8
+; GFX9-PAL-NEXT:    s_load_dwordx2 s[2:3], s[2:3], 0x0
+; GFX9-PAL-NEXT:    v_mov_b32_e32 v1, 15
+; GFX9-PAL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-PAL-NEXT:    s_and_b32 s3, s3, 0xffff
+; GFX9-PAL-NEXT:    s_add_u32 flat_scratch_lo, s2, s5
+; GFX9-PAL-NEXT:    s_addc_u32 flat_scratch_hi, s3, 0
+; GFX9-PAL-NEXT:    s_add_i32 s0, s0, s1
+; GFX9-PAL-NEXT:    v_add_u32_e32 v0, s0, v0
+; GFX9-PAL-NEXT:    v_add_u32_e32 v0, -16, v0
+; GFX9-PAL-NEXT:    scratch_store_dword v0, v1, off
+; GFX9-PAL-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-PAL-NEXT:    s_endpgm
+;
+; GFX940-LABEL: sgpr_base_plus_sgpr_plus_vgpr_plus_negative_imm_offset:
+; GFX940:       ; %bb.0: ; %bb
+; GFX940-NEXT:    s_add_i32 s0, s0, s1
+; GFX940-NEXT:    v_add_u32_e32 v0, s0, v0
+; GFX940-NEXT:    v_add_u32_e32 v0, -16, v0
+; GFX940-NEXT:    v_mov_b32_e32 v1, 15
+; GFX940-NEXT:    scratch_store_dword v0, v1, off sc0 sc1
+; GFX940-NEXT:    s_waitcnt vmcnt(0)
+; GFX940-NEXT:    s_endpgm
+;
+; GFX10-PAL-LABEL: sgpr_base_plus_sgpr_plus_vgpr_plus_negative_imm_offset:
+; GFX10-PAL:       ; %bb.0: ; %bb
+; GFX10-PAL-NEXT:    s_getpc_b64 s[2:3]
+; GFX10-PAL-NEXT:    s_mov_b32 s2, s8
+; GFX10-PAL-NEXT:    s_load_dwordx2 s[2:3], s[2:3], 0x0
+; GFX10-PAL-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX10-PAL-NEXT:    s_and_b32 s3, s3, 0xffff
+; GFX10-PAL-NEXT:    s_add_u32 s2, s2, s5
+; GFX10-PAL-NEXT:    s_addc_u32 s3, s3, 0
+; GFX10-PAL-NEXT:    s_setreg_b32 hwreg(HW_REG_FLAT_SCR_LO), s2
+; GFX10-PAL-NEXT:    s_setreg_b32 hwreg(HW_REG_FLAT_SCR_HI), s3
+; GFX10-PAL-NEXT:    v_add3_u32 v0, s0, s1, v0
+; GFX10-PAL-NEXT:    v_mov_b32_e32 v1, 15
+; GFX10-PAL-NEXT:    scratch_store_dword v0, v1, off offset:-16
+; GFX10-PAL-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-PAL-NEXT:    s_endpgm
+;
+; GFX11-PAL-LABEL: sgpr_base_plus_sgpr_plus_vgpr_plus_negative_imm_offset:
+; GFX11-PAL:       ; %bb.0: ; %bb
+; GFX11-PAL-NEXT:    v_add3_u32 v0, s0, s1, v0
+; GFX11-PAL-NEXT:    v_mov_b32_e32 v1, 15
+; GFX11-PAL-NEXT:    scratch_store_b32 v0, v1, off offset:-16 dlc
+; GFX11-PAL-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX11-PAL-NEXT:    s_endpgm
+;
+; GFX12-PAL-LABEL: sgpr_base_plus_sgpr_plus_vgpr_plus_negative_imm_offset:
+; GFX12-PAL:       ; %bb.0: ; %bb
+; GFX12-PAL-NEXT:    v_mov_b32_e32 v1, 15
+; GFX12-PAL-NEXT:    s_add_co_i32 s0, s0, s1
+; GFX12-PAL-NEXT:    scratch_store_b32 v0, v1, s0 offset:-16 scope:SCOPE_SYS
+; GFX12-PAL-NEXT:    s_wait_storecnt 0x0
+; GFX12-PAL-NEXT:    s_endpgm
+bb:
+  %add1 = add nsw i32 %sidx, %vidx
+  %add2 = add nsw i32 %add1, -16
+  %gep = getelementptr inbounds [16 x i8], ptr addrspace(5) %sgpr_base, i32 0, i32 %add2
+  store volatile i32 15, ptr addrspace(5) %gep, align 4
+  ret void
+}
+
 define amdgpu_gs void @sgpr_base_negative_offset(ptr addrspace(1) %out, ptr addrspace(5) inreg %scevgep) {
 ; GFX9-LABEL: sgpr_base_negative_offset:
 ; GFX9:       ; %bb.0: ; %entry

@changpeng changpeng requested a review from rampitec January 7, 2025 18:39
@changpeng changpeng merged commit 6869425 into llvm:main Jan 7, 2025
8 of 10 checks passed
@changpeng changpeng deleted the signed branch January 7, 2025 23:28
@shiltian shiltian mentioned this pull request Jan 8, 2025
Closed
searlmc1 pushed a commit to ROCm/llvm-project that referenced this pull request Feb 21, 2025
@changpeng
AMDGPU: Use getSignedTargetConstant for ImmOffset in SelectScratchSVA…
62533f7 
…ddr (llvm#121978)

ImmOffset is signed and we will hit an assert with negative ImmOffset
when getTargetConstant is used.

Fixes: SWDEV-506453
searlmc1 pushed a commit to ROCm/llvm-project that referenced this pull request Feb 21, 2025
@changpeng
AMDGPU: Use getSignedTargetConstant for ImmOffset in SelectScratchSVA…
1d7f33a 
…ddr(llvm#121978) (llvm#752)

ImmOffset is signed and we will hit an assert with negative ImmOffset when getTargetConstant is used.

Cherry-pick 6869425 to amd-mainkine to fix SWDEV-504260

NOTE: We cherry-pick two additional utility functions in SelectionDAG.h to make this cherry-pick possible:
getSignedConstant and getSignedTargetConstant
jrbyrnes pushed a commit to jrbyrnes/llvm-project that referenced this pull request Apr 29, 2025
@changpeng
AMDGPU: Use getSignedTargetConstant for ImmOffset in SelectScratchSVA…
2202700 
…ddr (llvm#121978)

ImmOffset is signed and we will hit an assert with negative ImmOffset
when getTargetConstant is used.

Fixes: SWDEV-506453

Cherry-pick from 62533f7 (amd-mainline)
jrbyrnes pushed a commit to jrbyrnes/llvm-project that referenced this pull request Apr 29, 2025
@rocm-devops
AMDGPU: Use getSignedTargetConstant for ImmOffset in SelectScratchSVA…
f1a78cc 
…ddr (llvm#121978) (llvm#1160)

Auto-submit by Jenkins
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@rampitec rampitec rampitec approved these changes

Assignees
No one assigned
Labels
backend:AMDGPU llvm:globalisel
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.
3 participants
@changpeng @llvmbot @rampitec