Skip to content

Commit

Permalink
AMDGPU: Fix crashing on calls to C functions from graphics contexts
Browse files Browse the repository at this point in the history 
If we had one of the shader calling conventions calling a default
calling convention callee, this would crash when the caller did not
have anything to pass to the workitem ID.

This is illegal, but we still need to produce something
sensible. llvm-reduce likes to replace calls to intrinsics with calls
to null or undef, so this does appear and is helpful to avoid hard
erroring.

Pass undef in this case, as already happened for the other implicit
arguments. It might make sense to define the behavior here and pass
null for the pointers, and -1 for the workitem ID. We do have extra
bits in the workitem ID, so this wouldn't conflict with a valid value.
  • Loading branch information
@arsenm
arsenm committed Jan 17, 2022
1 parent 9392b40 commit dc2457c
Show file tree
Hide file tree
Showing 5 changed files with 229 additions and 17 deletions.
24 changes: 16 additions & 8 deletions llvm/lib/Target/AMDGPU/AMDGPUCallLowering.cpp
View file
Expand Up @@ -908,16 +908,24 @@ bool AMDGPUCallLowering::passSpecialInputs(MachineIRBuilder &MIRBuilder,
InputReg = InputReg ? MIRBuilder.buildOr(S32, InputReg, Z).getReg(0) : Z;
}

if (!InputReg && (NeedWorkItemIDX || NeedWorkItemIDY || NeedWorkItemIDZ)) {
if (!InputReg &&
(NeedWorkItemIDX || NeedWorkItemIDY || NeedWorkItemIDZ)) {
InputReg = MRI.createGenericVirtualRegister(S32);

// Workitem ids are already packed, any of present incoming arguments will
// carry all required fields.
ArgDescriptor IncomingArg = ArgDescriptor::createArg(
IncomingArgX ? *IncomingArgX :
if (!IncomingArgX && !IncomingArgY && !IncomingArgZ) {
// We're in a situation where the outgoing function requires the workitem
// ID, but the calling function does not have it (e.g a graphics function
// calling a C calling convention function). This is illegal, but we need
// to produce something.
MIRBuilder.buildUndef(InputReg);
} else {
// Workitem ids are already packed, any of present incoming arguments will
// carry all required fields.
ArgDescriptor IncomingArg = ArgDescriptor::createArg(
IncomingArgX ? *IncomingArgX :
IncomingArgY ? *IncomingArgY : *IncomingArgZ, ~0u);
LI->loadInputValue(InputReg, MIRBuilder, &IncomingArg,
&AMDGPU::VGPR_32RegClass, S32);
LI->loadInputValue(InputReg, MIRBuilder, &IncomingArg,
&AMDGPU::VGPR_32RegClass, S32);
}
}

if (OutgoingArg->isRegister()) {
Expand Down
22 changes: 15 additions & 7 deletions llvm/lib/Target/AMDGPU/SIISelLowering.cpp
View file
Expand Up @@ -2911,13 +2911,21 @@ void SITargetLowering::passSpecialInputs(
}

if (!InputReg && (NeedWorkItemIDX || NeedWorkItemIDY || NeedWorkItemIDZ)) {
// Workitem ids are already packed, any of present incoming arguments
// will carry all required fields.
ArgDescriptor IncomingArg = ArgDescriptor::createArg(
IncomingArgX ? *IncomingArgX :
IncomingArgY ? *IncomingArgY :
*IncomingArgZ, ~0u);
InputReg = loadInputValue(DAG, ArgRC, MVT::i32, DL, IncomingArg);
if (!IncomingArgX && !IncomingArgY && !IncomingArgZ) {
// We're in a situation where the outgoing function requires the workitem
// ID, but the calling function does not have it (e.g a graphics function
// calling a C calling convention function). This is illegal, but we need
// to produce something.
InputReg = DAG.getUNDEF(MVT::i32);
} else {
// Workitem ids are already packed, any of present incoming arguments
// will carry all required fields.
ArgDescriptor IncomingArg = ArgDescriptor::createArg(
IncomingArgX ? *IncomingArgX :
IncomingArgY ? *IncomingArgY :
*IncomingArgZ, ~0u);
InputReg = loadInputValue(DAG, ArgRC, MVT::i32, DL, IncomingArg);
}
}

if (OutgoingArg->isRegister()) {
Expand Down
33 changes: 33 additions & 0 deletions llvm/test/CodeGen/AMDGPU/GlobalISel/irtranslator-call.ll
View file
Expand Up @@ -5175,6 +5175,39 @@ entry:
ret void
}

define amdgpu_ps void @amdgpu_ps_call_default_cc() {
; CHECK-LABEL: name: amdgpu_ps_call_default_cc
; CHECK: bb.1.main_body:
; CHECK-NEXT: [[C:%[0-9]+]]:_(p0) = G_CONSTANT i64 0
; CHECK-NEXT: ADJCALLSTACKUP 0, 0, implicit-def $scc
; CHECK-NEXT: [[DEF:%[0-9]+]]:_(p4) = G_IMPLICIT_DEF
; CHECK-NEXT: [[COPY:%[0-9]+]]:_(p4) = COPY [[DEF]](p4)
; CHECK-NEXT: [[C1:%[0-9]+]]:_(p4) = G_CONSTANT i64 0
; CHECK-NEXT: [[C2:%[0-9]+]]:_(s64) = G_CONSTANT i64 36
; CHECK-NEXT: [[PTR_ADD:%[0-9]+]]:_(p4) = G_PTR_ADD [[C1]], [[C2]](s64)
; CHECK-NEXT: [[DEF1:%[0-9]+]]:_(s64) = G_IMPLICIT_DEF
; CHECK-NEXT: [[DEF2:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
; CHECK-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY [[DEF2]](s32)
; CHECK-NEXT: [[COPY2:%[0-9]+]]:_(s32) = COPY [[DEF2]](s32)
; CHECK-NEXT: [[COPY3:%[0-9]+]]:_(s32) = COPY [[DEF2]](s32)
; CHECK-NEXT: [[COPY4:%[0-9]+]]:_(<4 x s32>) = COPY $private_rsrc_reg
; CHECK-NEXT: $sgpr0_sgpr1_sgpr2_sgpr3 = COPY [[COPY4]](<4 x s32>)
; CHECK-NEXT: $sgpr4_sgpr5 = COPY [[DEF]](p4)
; CHECK-NEXT: $sgpr6_sgpr7 = COPY [[COPY]](p4)
; CHECK-NEXT: $sgpr8_sgpr9 = COPY [[PTR_ADD]](p4)
; CHECK-NEXT: $sgpr10_sgpr11 = COPY [[DEF1]](s64)
; CHECK-NEXT: $sgpr12 = COPY [[DEF2]](s32)
; CHECK-NEXT: $sgpr13 = COPY [[COPY1]](s32)
; CHECK-NEXT: $sgpr14 = COPY [[COPY2]](s32)
; CHECK-NEXT: $vgpr31 = COPY [[COPY3]](s32)
; CHECK-NEXT: $sgpr30_sgpr31 = G_SI_CALL [[C]](p0), 0, csr_amdgpu_highregs, implicit $sgpr0_sgpr1_sgpr2_sgpr3, implicit $sgpr4_sgpr5, implicit $sgpr6_sgpr7, implicit $sgpr8_sgpr9, implicit $sgpr10_sgpr11, implicit $sgpr12, implicit $sgpr13, implicit $sgpr14, implicit $vgpr31
; CHECK-NEXT: ADJCALLSTACKDOWN 0, 0, implicit-def $scc
; CHECK-NEXT: S_ENDPGM 0
main_body:
call void null()
ret void
}

attributes #0 = { nounwind }
attributes #1 = { nounwind readnone }
attributes #2 = { nounwind noinline }
4 changes: 2 additions & 2 deletions llvm/test/CodeGen/AMDGPU/amdpal-callable.ll
View file
Expand Up @@ -2,8 +2,8 @@
; RUN: llc -mtriple=amdgcn--amdpal -mcpu=gfx900 -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,SDAG,GFX9 -enable-var-scope %s
; RUN: llc -global-isel -mtriple=amdgcn--amdpal -mcpu=gfx900 -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,GISEL,GFX9 -enable-var-scope %s

declare float @extern_func(float) #0
declare float @extern_func_many_args(<64 x float>) #0
declare amdgpu_gfx float @extern_func(float) #0
declare amdgpu_gfx float @extern_func_many_args(<64 x float>) #0

@funcptr = external hidden unnamed_addr addrspace(4) constant void()*, align 4

Expand Down
163 changes: 163 additions & 0 deletions llvm/test/CodeGen/AMDGPU/gfx-call-non-gfx-func.ll
View file
@@ -0,0 +1,163 @@
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -mtriple=amdgcn--amdpal -mcpu=gfx900 -verify-machineinstrs < %s | FileCheck -check-prefix=SDAG -enable-var-scope %s
; RUN: llc -global-isel -mtriple=amdgcn--amdpal -mcpu=gfx900 -verify-machineinstrs < %s | FileCheck -check-prefix=GISEL -enable-var-scope %s

declare void @extern_c_func()

define amdgpu_gfx void @gfx_func() {
; SDAG-LABEL: gfx_func:
; SDAG: ; %bb.0:
; SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; SDAG-NEXT: s_or_saveexec_b64 s[34:35], -1
; SDAG-NEXT: buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
; SDAG-NEXT: s_mov_b64 exec, s[34:35]
; SDAG-NEXT: v_writelane_b32 v40, s33, 26
; SDAG-NEXT: v_writelane_b32 v40, s4, 0
; SDAG-NEXT: v_writelane_b32 v40, s5, 1
; SDAG-NEXT: v_writelane_b32 v40, s6, 2
; SDAG-NEXT: v_writelane_b32 v40, s7, 3
; SDAG-NEXT: v_writelane_b32 v40, s8, 4
; SDAG-NEXT: v_writelane_b32 v40, s9, 5
; SDAG-NEXT: v_writelane_b32 v40, s10, 6
; SDAG-NEXT: v_writelane_b32 v40, s11, 7
; SDAG-NEXT: v_writelane_b32 v40, s12, 8
; SDAG-NEXT: v_writelane_b32 v40, s13, 9
; SDAG-NEXT: v_writelane_b32 v40, s14, 10
; SDAG-NEXT: v_writelane_b32 v40, s15, 11
; SDAG-NEXT: v_writelane_b32 v40, s16, 12
; SDAG-NEXT: v_writelane_b32 v40, s17, 13
; SDAG-NEXT: v_writelane_b32 v40, s18, 14
; SDAG-NEXT: v_writelane_b32 v40, s19, 15
; SDAG-NEXT: v_writelane_b32 v40, s20, 16
; SDAG-NEXT: v_writelane_b32 v40, s21, 17
; SDAG-NEXT: s_mov_b32 s33, s32
; SDAG-NEXT: s_addk_i32 s32, 0x400
; SDAG-NEXT: v_writelane_b32 v40, s22, 18
; SDAG-NEXT: v_writelane_b32 v40, s23, 19
; SDAG-NEXT: s_mov_b64 s[36:37], s[30:31]
; SDAG-NEXT: s_getpc_b64 s[30:31]
; SDAG-NEXT: s_add_u32 s30, s30, extern_c_func@gotpcrel32@lo+4
; SDAG-NEXT: s_addc_u32 s31, s31, extern_c_func@gotpcrel32@hi+12
; SDAG-NEXT: v_writelane_b32 v40, s24, 20
; SDAG-NEXT: s_load_dwordx2 s[30:31], s[30:31], 0x0
; SDAG-NEXT: v_writelane_b32 v40, s25, 21
; SDAG-NEXT: v_writelane_b32 v40, s26, 22
; SDAG-NEXT: v_writelane_b32 v40, s27, 23
; SDAG-NEXT: v_writelane_b32 v40, s28, 24
; SDAG-NEXT: s_mov_b64 s[8:9], 0
; SDAG-NEXT: v_writelane_b32 v40, s29, 25
; SDAG-NEXT: s_waitcnt lgkmcnt(0)
; SDAG-NEXT: s_swappc_b64 s[30:31], s[30:31]
; SDAG-NEXT: v_readlane_b32 s29, v40, 25
; SDAG-NEXT: v_readlane_b32 s28, v40, 24
; SDAG-NEXT: v_readlane_b32 s27, v40, 23
; SDAG-NEXT: v_readlane_b32 s26, v40, 22
; SDAG-NEXT: v_readlane_b32 s25, v40, 21
; SDAG-NEXT: v_readlane_b32 s24, v40, 20
; SDAG-NEXT: v_readlane_b32 s23, v40, 19
; SDAG-NEXT: v_readlane_b32 s22, v40, 18
; SDAG-NEXT: v_readlane_b32 s21, v40, 17
; SDAG-NEXT: v_readlane_b32 s20, v40, 16
; SDAG-NEXT: v_readlane_b32 s19, v40, 15
; SDAG-NEXT: v_readlane_b32 s18, v40, 14
; SDAG-NEXT: v_readlane_b32 s17, v40, 13
; SDAG-NEXT: v_readlane_b32 s16, v40, 12
; SDAG-NEXT: v_readlane_b32 s15, v40, 11
; SDAG-NEXT: v_readlane_b32 s14, v40, 10
; SDAG-NEXT: v_readlane_b32 s13, v40, 9
; SDAG-NEXT: v_readlane_b32 s12, v40, 8
; SDAG-NEXT: v_readlane_b32 s11, v40, 7
; SDAG-NEXT: v_readlane_b32 s10, v40, 6
; SDAG-NEXT: v_readlane_b32 s9, v40, 5
; SDAG-NEXT: v_readlane_b32 s8, v40, 4
; SDAG-NEXT: v_readlane_b32 s7, v40, 3
; SDAG-NEXT: v_readlane_b32 s6, v40, 2
; SDAG-NEXT: v_readlane_b32 s5, v40, 1
; SDAG-NEXT: v_readlane_b32 s4, v40, 0
; SDAG-NEXT: s_addk_i32 s32, 0xfc00
; SDAG-NEXT: v_readlane_b32 s33, v40, 26
; SDAG-NEXT: s_or_saveexec_b64 s[30:31], -1
; SDAG-NEXT: buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
; SDAG-NEXT: s_mov_b64 exec, s[30:31]
; SDAG-NEXT: s_waitcnt vmcnt(0)
; SDAG-NEXT: s_setpc_b64 s[36:37]
;
; GISEL-LABEL: gfx_func:
; GISEL: ; %bb.0:
; GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GISEL-NEXT: s_or_saveexec_b64 s[34:35], -1
; GISEL-NEXT: buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill
; GISEL-NEXT: s_mov_b64 exec, s[34:35]
; GISEL-NEXT: v_writelane_b32 v40, s33, 26
; GISEL-NEXT: v_writelane_b32 v40, s4, 0
; GISEL-NEXT: v_writelane_b32 v40, s5, 1
; GISEL-NEXT: v_writelane_b32 v40, s6, 2
; GISEL-NEXT: v_writelane_b32 v40, s7, 3
; GISEL-NEXT: v_writelane_b32 v40, s8, 4
; GISEL-NEXT: v_writelane_b32 v40, s9, 5
; GISEL-NEXT: v_writelane_b32 v40, s10, 6
; GISEL-NEXT: v_writelane_b32 v40, s11, 7
; GISEL-NEXT: v_writelane_b32 v40, s12, 8
; GISEL-NEXT: v_writelane_b32 v40, s13, 9
; GISEL-NEXT: v_writelane_b32 v40, s14, 10
; GISEL-NEXT: v_writelane_b32 v40, s15, 11
; GISEL-NEXT: v_writelane_b32 v40, s16, 12
; GISEL-NEXT: v_writelane_b32 v40, s17, 13
; GISEL-NEXT: v_writelane_b32 v40, s18, 14
; GISEL-NEXT: v_writelane_b32 v40, s19, 15
; GISEL-NEXT: v_writelane_b32 v40, s20, 16
; GISEL-NEXT: v_writelane_b32 v40, s21, 17
; GISEL-NEXT: s_mov_b32 s33, s32
; GISEL-NEXT: s_addk_i32 s32, 0x400
; GISEL-NEXT: v_writelane_b32 v40, s22, 18
; GISEL-NEXT: v_writelane_b32 v40, s23, 19
; GISEL-NEXT: s_mov_b64 s[36:37], s[30:31]
; GISEL-NEXT: s_getpc_b64 s[30:31]
; GISEL-NEXT: s_add_u32 s30, s30, extern_c_func@gotpcrel32@lo+4
; GISEL-NEXT: s_addc_u32 s31, s31, extern_c_func@gotpcrel32@hi+12
; GISEL-NEXT: v_writelane_b32 v40, s24, 20
; GISEL-NEXT: s_load_dwordx2 s[30:31], s[30:31], 0x0
; GISEL-NEXT: v_writelane_b32 v40, s25, 21
; GISEL-NEXT: v_writelane_b32 v40, s26, 22
; GISEL-NEXT: v_writelane_b32 v40, s27, 23
; GISEL-NEXT: v_writelane_b32 v40, s28, 24
; GISEL-NEXT: s_mov_b64 s[8:9], s[4:5]
; GISEL-NEXT: v_writelane_b32 v40, s29, 25
; GISEL-NEXT: s_waitcnt lgkmcnt(0)
; GISEL-NEXT: s_swappc_b64 s[30:31], s[30:31]
; GISEL-NEXT: v_readlane_b32 s29, v40, 25
; GISEL-NEXT: v_readlane_b32 s28, v40, 24
; GISEL-NEXT: v_readlane_b32 s27, v40, 23
; GISEL-NEXT: v_readlane_b32 s26, v40, 22
; GISEL-NEXT: v_readlane_b32 s25, v40, 21
; GISEL-NEXT: v_readlane_b32 s24, v40, 20
; GISEL-NEXT: v_readlane_b32 s23, v40, 19
; GISEL-NEXT: v_readlane_b32 s22, v40, 18
; GISEL-NEXT: v_readlane_b32 s21, v40, 17
; GISEL-NEXT: v_readlane_b32 s20, v40, 16
; GISEL-NEXT: v_readlane_b32 s19, v40, 15
; GISEL-NEXT: v_readlane_b32 s18, v40, 14
; GISEL-NEXT: v_readlane_b32 s17, v40, 13
; GISEL-NEXT: v_readlane_b32 s16, v40, 12
; GISEL-NEXT: v_readlane_b32 s15, v40, 11
; GISEL-NEXT: v_readlane_b32 s14, v40, 10
; GISEL-NEXT: v_readlane_b32 s13, v40, 9
; GISEL-NEXT: v_readlane_b32 s12, v40, 8
; GISEL-NEXT: v_readlane_b32 s11, v40, 7
; GISEL-NEXT: v_readlane_b32 s10, v40, 6
; GISEL-NEXT: v_readlane_b32 s9, v40, 5
; GISEL-NEXT: v_readlane_b32 s8, v40, 4
; GISEL-NEXT: v_readlane_b32 s7, v40, 3
; GISEL-NEXT: v_readlane_b32 s6, v40, 2
; GISEL-NEXT: v_readlane_b32 s5, v40, 1
; GISEL-NEXT: v_readlane_b32 s4, v40, 0
; GISEL-NEXT: s_addk_i32 s32, 0xfc00
; GISEL-NEXT: v_readlane_b32 s33, v40, 26
; GISEL-NEXT: s_or_saveexec_b64 s[30:31], -1
; GISEL-NEXT: buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
; GISEL-NEXT: s_mov_b64 exec, s[30:31]
; GISEL-NEXT: s_waitcnt vmcnt(0)
; GISEL-NEXT: s_setpc_b64 s[36:37]
call void @extern_c_func()
ret void
}

0 comments on commit dc2457c

Please sign in to comment.