Previously we allowed the store to be Custom. But without knowing for sure that the Custom handling won't split the store, we shouldn't convert a volatile store. We also probably shouldn't be creating a store the requires custom handling after LegalizeOps. This could lead to an infinite loop if the custom handling was to insert a bitcast. Though I guess isStoreBitCastBeneficial could be used to block such a loop. The test changes here are due to the volatile part of this. The stores in the test are all volatile and i32 stores are marked custom, So we are no longer converting them This is related to D50491 where I was trying to allow some bitcasting of volatile loads Differential Revision: https://reviews.llvm.org/D50578 llvm-svn: 340626
163 lines
7.3 KiB
LLVM
163 lines
7.3 KiB
LLVM
; RUN: llc -march=mips -relocation-model=static < %s | FileCheck -allow-deprecated-dag-overlap --check-prefixes=ALL,SYM32,O32,O32BE %s
|
|
; RUN: llc -march=mipsel -relocation-model=static < %s | FileCheck -allow-deprecated-dag-overlap --check-prefixes=ALL,SYM32,O32,O32LE %s
|
|
|
|
; RUN-TODO: llc -march=mips64 -relocation-model=static -target-abi o32 < %s | FileCheck --check-prefixes=ALL,SYM32,O32 %s
|
|
; RUN-TODO: llc -march=mips64el -relocation-model=static -target-abi o32 < %s | FileCheck --check-prefixes=ALL,SYM32,O32 %s
|
|
|
|
; RUN: llc -march=mips64 -relocation-model=static -target-abi n32 < %s | FileCheck -allow-deprecated-dag-overlap --check-prefixes=ALL,SYM32,N32,NEW,NEWBE %s
|
|
; RUN: llc -march=mips64el -relocation-model=static -target-abi n32 < %s | FileCheck -allow-deprecated-dag-overlap --check-prefixes=ALL,SYM32,N32,NEW,NEWLE %s
|
|
|
|
; RUN: llc -march=mips64 -relocation-model=static -target-abi n64 < %s | FileCheck -allow-deprecated-dag-overlap --check-prefixes=ALL,SYM64,N64,NEW,NEWBE %s
|
|
; RUN: llc -march=mips64el -relocation-model=static -target-abi n64 < %s | FileCheck -allow-deprecated-dag-overlap --check-prefixes=ALL,SYM64,N64,NEW,NEWLE %s
|
|
|
|
; Test the effect of varargs on floating point types in the non-variable part
|
|
; of the argument list as specified by section 2 of the MIPSpro N32 Handbook.
|
|
;
|
|
; N32/N64 are almost identical in this area so many of their checks have been
|
|
; combined into the 'NEW' prefix (the N stands for New).
|
|
;
|
|
; On O32, varargs prevents all FPU argument register usage. This contradicts
|
|
; the N32 handbook, but agrees with the SYSV ABI and GCC's behaviour.
|
|
|
|
@floats = global [11 x float] zeroinitializer
|
|
@doubles = global [11 x double] zeroinitializer
|
|
|
|
define void @double_args(double %a, ...)
|
|
nounwind {
|
|
entry:
|
|
%0 = getelementptr [11 x double], [11 x double]* @doubles, i32 0, i32 1
|
|
store volatile double %a, double* %0
|
|
|
|
%ap = alloca i8*
|
|
%ap2 = bitcast i8** %ap to i8*
|
|
call void @llvm.va_start(i8* %ap2)
|
|
%b = va_arg i8** %ap, double
|
|
%1 = getelementptr [11 x double], [11 x double]* @doubles, i32 0, i32 2
|
|
store volatile double %b, double* %1
|
|
call void @llvm.va_end(i8* %ap2)
|
|
ret void
|
|
}
|
|
|
|
; ALL-LABEL: double_args:
|
|
; We won't test the way the global address is calculated in this test. This is
|
|
; just to get the register number for the other checks.
|
|
; SYM32-DAG: addiu [[R2:\$[0-9]+]], ${{[0-9]+}}, %lo(doubles)
|
|
; SYM64-DAG: daddiu [[R2:\$[0-9]+]], ${{[0-9]+}}, %lo(doubles)
|
|
|
|
; O32 forbids using floating point registers for the non-variable portion.
|
|
; N32/N64 allow it.
|
|
; O32BE-DAG: mtc1 $5, [[FTMP1:\$f[0-9]*[02468]+]]
|
|
; O32BE-DAG: mtc1 $4, [[FTMP2:\$f[0-9]*[13579]+]]
|
|
; O32LE-DAG: mtc1 $4, [[FTMP1:\$f[0-9]*[02468]+]]
|
|
; O32LE-DAG: mtc1 $5, [[FTMP2:\$f[0-9]*[13579]+]]
|
|
; O32-DAG: sdc1 [[FTMP1]], 8([[R2]])
|
|
; NEW-DAG: sdc1 $f12, 8([[R2]])
|
|
|
|
; The varargs portion is dumped to stack
|
|
; O32-DAG: sw $6, 16($sp)
|
|
; O32-DAG: sw $7, 20($sp)
|
|
; NEW-DAG: sd $5, 8($sp)
|
|
; NEW-DAG: sd $6, 16($sp)
|
|
; NEW-DAG: sd $7, 24($sp)
|
|
; NEW-DAG: sd $8, 32($sp)
|
|
; NEW-DAG: sd $9, 40($sp)
|
|
; NEW-DAG: sd $10, 48($sp)
|
|
; NEW-DAG: sd $11, 56($sp)
|
|
|
|
; Get the varargs pointer
|
|
; O32 has 4 bytes padding, 4 bytes for the varargs pointer, and 8 bytes reserved
|
|
; for arguments 1 and 2.
|
|
; N32/N64 has 8 bytes for the varargs pointer, and no reserved area.
|
|
; O32-DAG: addiu [[VAPTR:\$[0-9]+]], $sp, 16
|
|
; O32-DAG: sw [[VAPTR]], 4($sp)
|
|
; N32-DAG: addiu [[VAPTR:\$[0-9]+]], $sp, 8
|
|
; N32-DAG: sw [[VAPTR]], 4($sp)
|
|
; N64-DAG: daddiu [[VAPTR:\$[0-9]+]], $sp, 8
|
|
; N64-DAG: sd [[VAPTR]], 0($sp)
|
|
|
|
; Increment the pointer then get the varargs arg
|
|
; LLVM will rebind the load to the stack pointer instead of the varargs pointer
|
|
; during lowering. This is fine and doesn't change the behaviour.
|
|
; O32-DAG: addiu [[VAPTR]], [[VAPTR]], 8
|
|
; O32-DAG: sw [[VAPTR]], 4($sp)
|
|
; N32-DAG: addiu [[VAPTR]], [[VAPTR]], 8
|
|
; N32-DAG: sw [[VAPTR]], 4($sp)
|
|
; N64-DAG: daddiu [[VAPTR]], [[VAPTR]], 8
|
|
; N64-DAG: sd [[VAPTR]], 0($sp)
|
|
; O32-DAG: ldc1 [[FTMP1:\$f[0-9]+]], 16($sp)
|
|
; NEW-DAG: ldc1 [[FTMP1:\$f[0-9]+]], 8($sp)
|
|
; ALL-DAG: sdc1 [[FTMP1]], 16([[R2]])
|
|
|
|
define void @float_args(float %a, ...) nounwind {
|
|
entry:
|
|
%0 = getelementptr [11 x float], [11 x float]* @floats, i32 0, i32 1
|
|
store volatile float %a, float* %0
|
|
|
|
%ap = alloca i8*
|
|
%ap2 = bitcast i8** %ap to i8*
|
|
call void @llvm.va_start(i8* %ap2)
|
|
%b = va_arg i8** %ap, float
|
|
%1 = getelementptr [11 x float], [11 x float]* @floats, i32 0, i32 2
|
|
store volatile float %b, float* %1
|
|
call void @llvm.va_end(i8* %ap2)
|
|
ret void
|
|
}
|
|
|
|
; ALL-LABEL: float_args:
|
|
; We won't test the way the global address is calculated in this test. This is
|
|
; just to get the register number for the other checks.
|
|
; SYM32-DAG: addiu [[R2:\$[0-9]+]], ${{[0-9]+}}, %lo(floats)
|
|
; SYM64-DAG: daddiu [[R2:\$[0-9]+]], ${{[0-9]+}}, %lo(floats)
|
|
|
|
; The first four arguments are the same in O32/N32/N64.
|
|
; The non-variable portion should be unaffected.
|
|
; O32-DAG: mtc1 $4, $f0
|
|
; O32-DAG: swc1 $f0, 4([[R2]])
|
|
; NEW-DAG: swc1 $f12, 4([[R2]])
|
|
|
|
; The varargs portion is dumped to stack
|
|
; O32-DAG: sw $5, 12($sp)
|
|
; O32-DAG: sw $6, 16($sp)
|
|
; O32-DAG: sw $7, 20($sp)
|
|
; NEW-DAG: sd $5, 8($sp)
|
|
; NEW-DAG: sd $6, 16($sp)
|
|
; NEW-DAG: sd $7, 24($sp)
|
|
; NEW-DAG: sd $8, 32($sp)
|
|
; NEW-DAG: sd $9, 40($sp)
|
|
; NEW-DAG: sd $10, 48($sp)
|
|
; NEW-DAG: sd $11, 56($sp)
|
|
|
|
; Get the varargs pointer
|
|
; O32 has 4 bytes padding, 4 bytes for the varargs pointer, and should have 8
|
|
; bytes reserved for arguments 1 and 2 (the first float arg) but as discussed in
|
|
; arguments-float.ll, GCC doesn't agree with MD00305 and treats floats as 4
|
|
; bytes so we only have 12 bytes total.
|
|
; N32/N64 has 8 bytes for the varargs pointer, and no reserved area.
|
|
; O32-DAG: addiu [[VAPTR:\$[0-9]+]], $sp, 12
|
|
; O32-DAG: sw [[VAPTR]], 4($sp)
|
|
; N32-DAG: addiu [[VAPTR:\$[0-9]+]], $sp, 8
|
|
; N32-DAG: sw [[VAPTR]], 4($sp)
|
|
; N64-DAG: daddiu [[VAPTR:\$[0-9]+]], $sp, 8
|
|
; N64-DAG: sd [[VAPTR]], 0($sp)
|
|
|
|
; Increment the pointer then get the varargs arg
|
|
; LLVM will rebind the load to the stack pointer instead of the varargs pointer
|
|
; during lowering. This is fine and doesn't change the behaviour.
|
|
; Also, in big-endian mode the offset must be increased by 4 to retrieve the
|
|
; correct half of the argument slot.
|
|
;
|
|
; O32-DAG: addiu [[VAPTR]], [[VAPTR]], 4
|
|
; O32-DAG: sw [[VAPTR]], 4($sp)
|
|
; N32-DAG: addiu [[VAPTR]], [[VAPTR]], 8
|
|
; N32-DAG: sw [[VAPTR]], 4($sp)
|
|
; N64-DAG: daddiu [[VAPTR]], [[VAPTR]], 8
|
|
; N64-DAG: sd [[VAPTR]], 0($sp)
|
|
; O32-DAG: lwc1 [[FTMP1:\$f[0-9]+]], 12($sp)
|
|
; NEWLE-DAG: lwc1 [[FTMP1:\$f[0-9]+]], 8($sp)
|
|
; NEWBE-DAG: lwc1 [[FTMP1:\$f[0-9]+]], 12($sp)
|
|
; ALL-DAG: swc1 [[FTMP1]], 8([[R2]])
|
|
|
|
declare void @llvm.va_start(i8*)
|
|
declare void @llvm.va_copy(i8*, i8*)
|
|
declare void @llvm.va_end(i8*)
|