; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 6 ; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 < %s | FileCheck -check-prefix=GFX9 %s ; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1100 < %s | FileCheck -check-prefix=GFX11 %s ; Make sure that sub-dword arguments passed on the stack do not assert define i32 @stack_arg_i1(<8 x i32>, <8 x i32>, <8 x i32>, <4 x i32>, <3 x i32>, i1 %badarg) #0 { ; GFX9-LABEL: stack_arg_i1: ; GFX9: ; %bb.0: ; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX9-NEXT: buffer_load_ubyte v0, off, s[0:3], s32 ; GFX9-NEXT: s_waitcnt vmcnt(0) ; GFX9-NEXT: s_setpc_b64 s[30:31] ; ; GFX11-LABEL: stack_arg_i1: ; GFX11: ; %bb.0: ; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX11-NEXT: scratch_load_u8 v0, off, s32 ; GFX11-NEXT: s_waitcnt vmcnt(0) ; GFX11-NEXT: s_setpc_b64 s[30:31] %ext = zext i1 %badarg to i32 ret i32 %ext } define i32 @stack_arg_i1_zeroext(<8 x i32>, <8 x i32>, <8 x i32>, <4 x i32>, <3 x i32>, i1 zeroext %badarg) #0 { ; GFX9-LABEL: stack_arg_i1_zeroext: ; GFX9: ; %bb.0: ; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX9-NEXT: buffer_load_ubyte v0, off, s[0:3], s32 ; GFX9-NEXT: s_waitcnt vmcnt(0) ; GFX9-NEXT: s_setpc_b64 s[30:31] ; ; GFX11-LABEL: stack_arg_i1_zeroext: ; GFX11: ; %bb.0: ; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX11-NEXT: scratch_load_u8 v0, off, s32 ; GFX11-NEXT: s_waitcnt vmcnt(0) ; GFX11-NEXT: s_setpc_b64 s[30:31] %ext = zext i1 %badarg to i32 ret i32 %ext } define i32 @stack_arg_i1_signext(<8 x i32>, <8 x i32>, <8 x i32>, <4 x i32>, <3 x i32>, i1 signext %badarg) #0 { ; GFX9-LABEL: stack_arg_i1_signext: ; GFX9: ; %bb.0: ; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX9-NEXT: buffer_load_ubyte v0, off, s[0:3], s32 ; GFX9-NEXT: s_waitcnt vmcnt(0) ; GFX9-NEXT: v_bfe_i32 v0, v0, 0, 1 ; GFX9-NEXT: s_setpc_b64 s[30:31] ; ; GFX11-LABEL: stack_arg_i1_signext: ; GFX11: ; %bb.0: ; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX11-NEXT: scratch_load_u8 v0, off, s32 ; GFX11-NEXT: s_waitcnt vmcnt(0) ; GFX11-NEXT: v_bfe_i32 v0, v0, 0, 1 ; GFX11-NEXT: s_setpc_b64 s[30:31] %ext = sext i1 %badarg to i32 ret i32 %ext } define i32 @stack_arg_i8(<8 x i32>, <8 x i32>, <8 x i32>, <4 x i32>, <3 x i32>, i8 %badarg) #0 { ; GFX9-LABEL: stack_arg_i8: ; GFX9: ; %bb.0: ; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX9-NEXT: buffer_load_ubyte v0, off, s[0:3], s32 ; GFX9-NEXT: s_waitcnt vmcnt(0) ; GFX9-NEXT: s_setpc_b64 s[30:31] ; ; GFX11-LABEL: stack_arg_i8: ; GFX11: ; %bb.0: ; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX11-NEXT: scratch_load_u8 v0, off, s32 ; GFX11-NEXT: s_waitcnt vmcnt(0) ; GFX11-NEXT: s_setpc_b64 s[30:31] %ext = zext i8 %badarg to i32 ret i32 %ext } define i32 @stack_arg_i8_zeroext(<8 x i32>, <8 x i32>, <8 x i32>, <4 x i32>, <3 x i32>, i8 zeroext %badarg) #0 { ; GFX9-LABEL: stack_arg_i8_zeroext: ; GFX9: ; %bb.0: ; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX9-NEXT: buffer_load_ushort v0, off, s[0:3], s32 ; GFX9-NEXT: s_waitcnt vmcnt(0) ; GFX9-NEXT: s_setpc_b64 s[30:31] ; ; GFX11-LABEL: stack_arg_i8_zeroext: ; GFX11: ; %bb.0: ; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX11-NEXT: scratch_load_u16 v0, off, s32 ; GFX11-NEXT: s_waitcnt vmcnt(0) ; GFX11-NEXT: s_setpc_b64 s[30:31] %ext = zext i8 %badarg to i32 ret i32 %ext } define i32 @stack_arg_i8_signext(<8 x i32>, <8 x i32>, <8 x i32>, <4 x i32>, <3 x i32>, i8 signext %badarg) #0 { ; GFX9-LABEL: stack_arg_i8_signext: ; GFX9: ; %bb.0: ; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX9-NEXT: buffer_load_sshort v0, off, s[0:3], s32 ; GFX9-NEXT: s_waitcnt vmcnt(0) ; GFX9-NEXT: s_setpc_b64 s[30:31] ; ; GFX11-LABEL: stack_arg_i8_signext: ; GFX11: ; %bb.0: ; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX11-NEXT: scratch_load_i16 v0, off, s32 ; GFX11-NEXT: s_waitcnt vmcnt(0) ; GFX11-NEXT: s_setpc_b64 s[30:31] %ext = sext i8 %badarg to i32 ret i32 %ext } define i32 @stack_arg_i16(<8 x i32>, <8 x i32>, <8 x i32>, <4 x i32>, <3 x i32>, i16 %badarg) #0 { ; GFX9-LABEL: stack_arg_i16: ; GFX9: ; %bb.0: ; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX9-NEXT: buffer_load_ushort v0, off, s[0:3], s32 ; GFX9-NEXT: s_waitcnt vmcnt(0) ; GFX9-NEXT: s_setpc_b64 s[30:31] ; ; GFX11-LABEL: stack_arg_i16: ; GFX11: ; %bb.0: ; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX11-NEXT: scratch_load_u16 v0, off, s32 ; GFX11-NEXT: s_waitcnt vmcnt(0) ; GFX11-NEXT: s_setpc_b64 s[30:31] %ext = zext i16 %badarg to i32 ret i32 %ext } define i32 @stack_arg_i16_zeroext(<8 x i32>, <8 x i32>, <8 x i32>, <4 x i32>, <3 x i32>, i16 zeroext %badarg) #0 { ; GFX9-LABEL: stack_arg_i16_zeroext: ; GFX9: ; %bb.0: ; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX9-NEXT: buffer_load_ushort v0, off, s[0:3], s32 ; GFX9-NEXT: s_waitcnt vmcnt(0) ; GFX9-NEXT: s_setpc_b64 s[30:31] ; ; GFX11-LABEL: stack_arg_i16_zeroext: ; GFX11: ; %bb.0: ; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX11-NEXT: scratch_load_u16 v0, off, s32 ; GFX11-NEXT: s_waitcnt vmcnt(0) ; GFX11-NEXT: s_setpc_b64 s[30:31] %ext = zext i16 %badarg to i32 ret i32 %ext } define i32 @stack_arg_i16_signext(<8 x i32>, <8 x i32>, <8 x i32>, <4 x i32>, <3 x i32>, i16 signext %badarg) #0 { ; GFX9-LABEL: stack_arg_i16_signext: ; GFX9: ; %bb.0: ; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX9-NEXT: buffer_load_sshort v0, off, s[0:3], s32 ; GFX9-NEXT: s_waitcnt vmcnt(0) ; GFX9-NEXT: s_setpc_b64 s[30:31] ; ; GFX11-LABEL: stack_arg_i16_signext: ; GFX11: ; %bb.0: ; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX11-NEXT: scratch_load_i16 v0, off, s32 ; GFX11-NEXT: s_waitcnt vmcnt(0) ; GFX11-NEXT: s_setpc_b64 s[30:31] %ext = sext i16 %badarg to i32 ret i32 %ext } define i32 @stack_arg_i7(<8 x i32>, <8 x i32>, <8 x i32>, <4 x i32>, <3 x i32>, i7 %badarg) #0 { ; GFX9-LABEL: stack_arg_i7: ; GFX9: ; %bb.0: ; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX9-NEXT: buffer_load_ushort v0, off, s[0:3], s32 ; GFX9-NEXT: s_waitcnt vmcnt(0) ; GFX9-NEXT: v_and_b32_e32 v0, 0x7f, v0 ; GFX9-NEXT: s_setpc_b64 s[30:31] ; ; GFX11-LABEL: stack_arg_i7: ; GFX11: ; %bb.0: ; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX11-NEXT: scratch_load_u16 v0, off, s32 ; GFX11-NEXT: s_waitcnt vmcnt(0) ; GFX11-NEXT: v_and_b32_e32 v0, 0x7f, v0 ; GFX11-NEXT: s_setpc_b64 s[30:31] %ext = zext i7 %badarg to i32 ret i32 %ext } define i32 @stack_arg_i7_zeroext(<8 x i32>, <8 x i32>, <8 x i32>, <4 x i32>, <3 x i32>, i7 zeroext %badarg) #0 { ; GFX9-LABEL: stack_arg_i7_zeroext: ; GFX9: ; %bb.0: ; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX9-NEXT: buffer_load_ushort v0, off, s[0:3], s32 ; GFX9-NEXT: s_waitcnt vmcnt(0) ; GFX9-NEXT: s_setpc_b64 s[30:31] ; ; GFX11-LABEL: stack_arg_i7_zeroext: ; GFX11: ; %bb.0: ; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX11-NEXT: scratch_load_u16 v0, off, s32 ; GFX11-NEXT: s_waitcnt vmcnt(0) ; GFX11-NEXT: s_setpc_b64 s[30:31] %ext = zext i7 %badarg to i32 ret i32 %ext } define i32 @stack_arg_i7_signext(<8 x i32>, <8 x i32>, <8 x i32>, <4 x i32>, <3 x i32>, i7 signext %badarg) #0 { ; GFX9-LABEL: stack_arg_i7_signext: ; GFX9: ; %bb.0: ; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX9-NEXT: buffer_load_sshort v0, off, s[0:3], s32 ; GFX9-NEXT: s_waitcnt vmcnt(0) ; GFX9-NEXT: s_setpc_b64 s[30:31] ; ; GFX11-LABEL: stack_arg_i7_signext: ; GFX11: ; %bb.0: ; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX11-NEXT: scratch_load_i16 v0, off, s32 ; GFX11-NEXT: s_waitcnt vmcnt(0) ; GFX11-NEXT: s_setpc_b64 s[30:31] %ext = sext i7 %badarg to i32 ret i32 %ext } define i32 @stack_arg_i17(<8 x i32>, <8 x i32>, <8 x i32>, <4 x i32>, <3 x i32>, i17 %badarg) #0 { ; GFX9-LABEL: stack_arg_i17: ; GFX9: ; %bb.0: ; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX9-NEXT: buffer_load_dword v0, off, s[0:3], s32 ; GFX9-NEXT: s_waitcnt vmcnt(0) ; GFX9-NEXT: v_and_b32_e32 v0, 0x1ffff, v0 ; GFX9-NEXT: s_setpc_b64 s[30:31] ; ; GFX11-LABEL: stack_arg_i17: ; GFX11: ; %bb.0: ; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX11-NEXT: scratch_load_b32 v0, off, s32 ; GFX11-NEXT: s_waitcnt vmcnt(0) ; GFX11-NEXT: v_and_b32_e32 v0, 0x1ffff, v0 ; GFX11-NEXT: s_setpc_b64 s[30:31] %ext = zext i17 %badarg to i32 ret i32 %ext } define i32 @stack_arg_i17_zeroext(<8 x i32>, <8 x i32>, <8 x i32>, <4 x i32>, <3 x i32>, i17 zeroext %badarg) #0 { ; GFX9-LABEL: stack_arg_i17_zeroext: ; GFX9: ; %bb.0: ; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX9-NEXT: buffer_load_dword v0, off, s[0:3], s32 ; GFX9-NEXT: s_waitcnt vmcnt(0) ; GFX9-NEXT: s_setpc_b64 s[30:31] ; ; GFX11-LABEL: stack_arg_i17_zeroext: ; GFX11: ; %bb.0: ; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX11-NEXT: scratch_load_b32 v0, off, s32 ; GFX11-NEXT: s_waitcnt vmcnt(0) ; GFX11-NEXT: s_setpc_b64 s[30:31] %ext = zext i17 %badarg to i32 ret i32 %ext } define i32 @stack_arg_i17_signext(<8 x i32>, <8 x i32>, <8 x i32>, <4 x i32>, <3 x i32>, i17 signext %badarg) #0 { ; GFX9-LABEL: stack_arg_i17_signext: ; GFX9: ; %bb.0: ; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX9-NEXT: buffer_load_dword v0, off, s[0:3], s32 ; GFX9-NEXT: s_waitcnt vmcnt(0) ; GFX9-NEXT: s_setpc_b64 s[30:31] ; ; GFX11-LABEL: stack_arg_i17_signext: ; GFX11: ; %bb.0: ; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX11-NEXT: scratch_load_b32 v0, off, s32 ; GFX11-NEXT: s_waitcnt vmcnt(0) ; GFX11-NEXT: s_setpc_b64 s[30:31] %ext = sext i17 %badarg to i32 ret i32 %ext } attributes #0 = { nounwind }