maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity
llvm-project/llvm/test/CodeGen/Mips/hf16call32.ll

1031 lines
49 KiB
LLVM
Raw Normal View History

[Mips] Adjust float ABI settings in case of MIPS16 mode. Hard float for mips16 means essentially to compile as soft float but to use a runtime library for soft float that is written with native mips32 floating point instructions (those runtime routines run in mips32 hard float mode). The patch reviewed by Reed Kotler. llvm-svn: 195123
2013-11-19 20:20:17 +08:00
; RUN: llc -mtriple=mipsel-linux-gnu -march=mipsel -mcpu=mips16 -relocation-model=static < %s | FileCheck %s -check-prefix=stel
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
@x = common global float 0.000000e+00, align 4
@y = common global float 0.000000e+00, align 4
@xd = common global double 0.000000e+00, align 8
@yd = common global double 0.000000e+00, align 8
@xy = common global { float, float } zeroinitializer, align 4
@xyd = common global { double, double } zeroinitializer, align 8
@ret_sf = common global float 0.000000e+00, align 4
@ret_df = common global double 0.000000e+00, align 8
@ret_sc = common global { float, float } zeroinitializer, align 4
@ret_dc = common global { double, double } zeroinitializer, align 8
@lx = common global float 0.000000e+00, align 4
@ly = common global float 0.000000e+00, align 4
@lxd = common global double 0.000000e+00, align 8
@lyd = common global double 0.000000e+00, align 8
@lxy = common global { float, float } zeroinitializer, align 4
@lxyd = common global { double, double } zeroinitializer, align 8
@lret_sf = common global float 0.000000e+00, align 4
@lret_df = common global double 0.000000e+00, align 8
@lret_sc = common global { float, float } zeroinitializer, align 4
@lret_dc = common global { double, double } zeroinitializer, align 8
@.str = private unnamed_addr constant [10 x i8] c"%f %f %i\0A\00", align 1
@.str1 = private unnamed_addr constant [16 x i8] c"%f=%f %f=%f %i\0A\00", align 1
@.str2 = private unnamed_addr constant [22 x i8] c"%f=%f %f=%f %f=%f %i\0A\00", align 1
@.str3 = private unnamed_addr constant [18 x i8] c"%f+%fi=%f+%fi %i\0A\00", align 1
@.str4 = private unnamed_addr constant [24 x i8] c"%f+%fi=%f+%fi %f=%f %i\0A\00", align 1
; Function Attrs: nounwind
define void @clear() #0 {
entry:
store float 1.000000e+00, float* @x, align 4
store float 1.000000e+00, float* @y, align 4
store double 1.000000e+00, double* @xd, align 8
store double 1.000000e+00, double* @yd, align 8
[opaque pointer type] Add textual IR support for explicit type parameter to gep operator Similar to gep (r230786) and load (r230794) changes. Similar migration script can be used to update test cases, which successfully migrated all of LLVM and Polly, but about 4 test cases needed manually changes in Clang. (this script will read the contents of stdin and massage it into stdout - wrap it in the 'apply.sh' script shown in previous commits + xargs to apply it over a large set of test cases) import fileinput import sys import re rep = re.compile(r"(getelementptr(?:\s+inbounds)?\s*\()((<\d*\s+x\s+)?([^@]*?)(|\s*addrspace\(\d+\))\s*\*(?(3)>)\s*)(?=$|%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|zeroinitializer|<|\[\[[a-zA-Z]|\{\{)", re.MULTILINE | re.DOTALL) def conv(match): line = match.group(1) line += match.group(4) line += ", " line += match.group(2) return line line = sys.stdin.read() off = 0 for match in re.finditer(rep, line): sys.stdout.write(line[off:match.start()]) sys.stdout.write(conv(match)) off = match.end() sys.stdout.write(line[off:]) llvm-svn: 232184
2015-03-14 02:20:45 +08:00
store float 1.000000e+00, float* getelementptr inbounds ({ float, float }, { float, float }* @xy, i32 0, i32 0)
store float 0.000000e+00, float* getelementptr inbounds ({ float, float }, { float, float }* @xy, i32 0, i32 1)
store double 1.000000e+00, double* getelementptr inbounds ({ double, double }, { double, double }* @xyd, i32 0, i32 0)
store double 0.000000e+00, double* getelementptr inbounds ({ double, double }, { double, double }* @xyd, i32 0, i32 1)
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
store float 1.000000e+00, float* @ret_sf, align 4
store double 1.000000e+00, double* @ret_df, align 8
[opaque pointer type] Add textual IR support for explicit type parameter to gep operator Similar to gep (r230786) and load (r230794) changes. Similar migration script can be used to update test cases, which successfully migrated all of LLVM and Polly, but about 4 test cases needed manually changes in Clang. (this script will read the contents of stdin and massage it into stdout - wrap it in the 'apply.sh' script shown in previous commits + xargs to apply it over a large set of test cases) import fileinput import sys import re rep = re.compile(r"(getelementptr(?:\s+inbounds)?\s*\()((<\d*\s+x\s+)?([^@]*?)(|\s*addrspace\(\d+\))\s*\*(?(3)>)\s*)(?=$|%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|zeroinitializer|<|\[\[[a-zA-Z]|\{\{)", re.MULTILINE | re.DOTALL) def conv(match): line = match.group(1) line += match.group(4) line += ", " line += match.group(2) return line line = sys.stdin.read() off = 0 for match in re.finditer(rep, line): sys.stdout.write(line[off:match.start()]) sys.stdout.write(conv(match)) off = match.end() sys.stdout.write(line[off:]) llvm-svn: 232184
2015-03-14 02:20:45 +08:00
store float 1.000000e+00, float* getelementptr inbounds ({ float, float }, { float, float }* @ret_sc, i32 0, i32 0)
store float 0.000000e+00, float* getelementptr inbounds ({ float, float }, { float, float }* @ret_sc, i32 0, i32 1)
store double 1.000000e+00, double* getelementptr inbounds ({ double, double }, { double, double }* @ret_dc, i32 0, i32 0)
store double 0.000000e+00, double* getelementptr inbounds ({ double, double }, { double, double }* @ret_dc, i32 0, i32 1)
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
store float 0.000000e+00, float* @lx, align 4
store float 0.000000e+00, float* @ly, align 4
store double 0.000000e+00, double* @lxd, align 8
store double 0.000000e+00, double* @lyd, align 8
[opaque pointer type] Add textual IR support for explicit type parameter to gep operator Similar to gep (r230786) and load (r230794) changes. Similar migration script can be used to update test cases, which successfully migrated all of LLVM and Polly, but about 4 test cases needed manually changes in Clang. (this script will read the contents of stdin and massage it into stdout - wrap it in the 'apply.sh' script shown in previous commits + xargs to apply it over a large set of test cases) import fileinput import sys import re rep = re.compile(r"(getelementptr(?:\s+inbounds)?\s*\()((<\d*\s+x\s+)?([^@]*?)(|\s*addrspace\(\d+\))\s*\*(?(3)>)\s*)(?=$|%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|zeroinitializer|<|\[\[[a-zA-Z]|\{\{)", re.MULTILINE | re.DOTALL) def conv(match): line = match.group(1) line += match.group(4) line += ", " line += match.group(2) return line line = sys.stdin.read() off = 0 for match in re.finditer(rep, line): sys.stdout.write(line[off:match.start()]) sys.stdout.write(conv(match)) off = match.end() sys.stdout.write(line[off:]) llvm-svn: 232184
2015-03-14 02:20:45 +08:00
store float 0.000000e+00, float* getelementptr inbounds ({ float, float }, { float, float }* @lxy, i32 0, i32 0)
store float 0.000000e+00, float* getelementptr inbounds ({ float, float }, { float, float }* @lxy, i32 0, i32 1)
store double 0.000000e+00, double* getelementptr inbounds ({ double, double }, { double, double }* @lxyd, i32 0, i32 0)
store double 0.000000e+00, double* getelementptr inbounds ({ double, double }, { double, double }* @lxyd, i32 0, i32 1)
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
store float 0.000000e+00, float* @lret_sf, align 4
store double 0.000000e+00, double* @lret_df, align 8
[opaque pointer type] Add textual IR support for explicit type parameter to gep operator Similar to gep (r230786) and load (r230794) changes. Similar migration script can be used to update test cases, which successfully migrated all of LLVM and Polly, but about 4 test cases needed manually changes in Clang. (this script will read the contents of stdin and massage it into stdout - wrap it in the 'apply.sh' script shown in previous commits + xargs to apply it over a large set of test cases) import fileinput import sys import re rep = re.compile(r"(getelementptr(?:\s+inbounds)?\s*\()((<\d*\s+x\s+)?([^@]*?)(|\s*addrspace\(\d+\))\s*\*(?(3)>)\s*)(?=$|%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|zeroinitializer|<|\[\[[a-zA-Z]|\{\{)", re.MULTILINE | re.DOTALL) def conv(match): line = match.group(1) line += match.group(4) line += ", " line += match.group(2) return line line = sys.stdin.read() off = 0 for match in re.finditer(rep, line): sys.stdout.write(line[off:match.start()]) sys.stdout.write(conv(match)) off = match.end() sys.stdout.write(line[off:]) llvm-svn: 232184
2015-03-14 02:20:45 +08:00
store float 0.000000e+00, float* getelementptr inbounds ({ float, float }, { float, float }* @lret_sc, i32 0, i32 0)
store float 0.000000e+00, float* getelementptr inbounds ({ float, float }, { float, float }* @lret_sc, i32 0, i32 1)
store double 0.000000e+00, double* getelementptr inbounds ({ double, double }, { double, double }* @lret_dc, i32 0, i32 0)
store double 0.000000e+00, double* getelementptr inbounds ({ double, double }, { double, double }* @lret_dc, i32 0, i32 1)
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
ret void
}
; Function Attrs: nounwind
define i32 @main() #0 {
entry:
%retval = alloca i32, align 4
store i32 0, i32* %retval
call void @clear()
store float 1.500000e+00, float* @lx, align 4
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%0 = load float, float* @lx, align 4
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
call void @v_sf(float %0)
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%1 = load float, float* @x, align 4
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%conv = fpext float %1 to double
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%2 = load float, float* @lx, align 4
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%conv1 = fpext float %2 to double
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%3 = load float, float* @x, align 4
%4 = load float, float* @lx, align 4
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%cmp = fcmp oeq float %3, %4
%conv2 = zext i1 %cmp to i32
[opaque pointer type] Add textual IR support for explicit type parameter to the call instruction See r230786 and r230794 for similar changes to gep and load respectively. Call is a bit different because it often doesn't have a single explicit type - usually the type is deduced from the arguments, and just the return type is explicit. In those cases there's no need to change the IR. When that's not the case, the IR usually contains the pointer type of the first operand - but since typed pointers are going away, that representation is insufficient so I'm just stripping the "pointerness" of the explicit type away. This does make the IR a bit weird - it /sort of/ reads like the type of the first operand: "call void () %x(" but %x is actually of type "void ()*" and will eventually be just of type "ptr". But this seems not too bad and I don't think it would benefit from repeating the type ("void (), void () * %x(" and then eventually "void (), ptr %x(") as has been done with gep and load. This also has a side benefit: since the explicit type is no longer a pointer, there's no ambiguity between an explicit type and a function that returns a function pointer. Previously this case needed an explicit type (eg: a function returning a void() function was written as "call void () () * @x(" rather than "call void () * @x(" because of the ambiguity between a function returning a pointer to a void() function and a function returning void). No ambiguity means even function pointer return types can just be written alone, without writing the whole function's type. This leaves /only/ the varargs case where the explicit type is required. Given the special type syntax in call instructions, the regex-fu used for migration was a bit more involved in its own unique way (as every one of these is) so here it is. Use it in conjunction with the apply.sh script and associated find/xargs commands I've provided in rr230786 to migrate your out of tree tests. Do let me know if any of this doesn't cover your cases & we can iterate on a more general script/regexes to help others with out of tree tests. About 9 test cases couldn't be automatically migrated - half of those were functions returning function pointers, where I just had to manually delete the function argument types now that we didn't need an explicit function type there. The other half were typedefs of function types used in calls - just had to manually drop the * from those. import fileinput import sys import re pat = re.compile(r'((?:=|:|^|\s)call\s(?:[^@]*?))(\s*$|\s*(?:(?:\[\[[a-zA-Z0-9_]+\]\]|[@%](?:(")?[\\\?@a-zA-Z0-9_.]*?(?(3)"|)|{{.*}}))(?:\(|$)|undef|inttoptr|bitcast|null|asm).*$)') addrspace_end = re.compile(r"addrspace\(\d+\)\s*\*$") func_end = re.compile("(?:void.*|\)\s*)\*$") def conv(match, line): if not match or re.search(addrspace_end, match.group(1)) or not re.search(func_end, match.group(1)): return line return line[:match.start()] + match.group(1)[:match.group(1).rfind('*')].rstrip() + match.group(2) + line[match.end():] for line in sys.stdin: sys.stdout.write(conv(re.search(pat, line), line)) llvm-svn: 235145
2015-04-17 07:24:18 +08:00
%call = call i32 (i8*, ...) @printf(i8* getelementptr inbounds ([10 x i8], [10 x i8]* @.str, i32 0, i32 0), double %conv, double %conv1, i32 %conv2)
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
call void @clear()
store double 0x41678C29C0000000, double* @lxd, align 8
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%5 = load double, double* @lxd, align 8
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
call void @v_df(double %5)
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%6 = load double, double* @xd, align 8
%7 = load double, double* @lxd, align 8
%8 = load double, double* @xd, align 8
%9 = load double, double* @lxd, align 8
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%cmp3 = fcmp oeq double %8, %9
%conv4 = zext i1 %cmp3 to i32
[opaque pointer type] Add textual IR support for explicit type parameter to the call instruction See r230786 and r230794 for similar changes to gep and load respectively. Call is a bit different because it often doesn't have a single explicit type - usually the type is deduced from the arguments, and just the return type is explicit. In those cases there's no need to change the IR. When that's not the case, the IR usually contains the pointer type of the first operand - but since typed pointers are going away, that representation is insufficient so I'm just stripping the "pointerness" of the explicit type away. This does make the IR a bit weird - it /sort of/ reads like the type of the first operand: "call void () %x(" but %x is actually of type "void ()*" and will eventually be just of type "ptr". But this seems not too bad and I don't think it would benefit from repeating the type ("void (), void () * %x(" and then eventually "void (), ptr %x(") as has been done with gep and load. This also has a side benefit: since the explicit type is no longer a pointer, there's no ambiguity between an explicit type and a function that returns a function pointer. Previously this case needed an explicit type (eg: a function returning a void() function was written as "call void () () * @x(" rather than "call void () * @x(" because of the ambiguity between a function returning a pointer to a void() function and a function returning void). No ambiguity means even function pointer return types can just be written alone, without writing the whole function's type. This leaves /only/ the varargs case where the explicit type is required. Given the special type syntax in call instructions, the regex-fu used for migration was a bit more involved in its own unique way (as every one of these is) so here it is. Use it in conjunction with the apply.sh script and associated find/xargs commands I've provided in rr230786 to migrate your out of tree tests. Do let me know if any of this doesn't cover your cases & we can iterate on a more general script/regexes to help others with out of tree tests. About 9 test cases couldn't be automatically migrated - half of those were functions returning function pointers, where I just had to manually delete the function argument types now that we didn't need an explicit function type there. The other half were typedefs of function types used in calls - just had to manually drop the * from those. import fileinput import sys import re pat = re.compile(r'((?:=|:|^|\s)call\s(?:[^@]*?))(\s*$|\s*(?:(?:\[\[[a-zA-Z0-9_]+\]\]|[@%](?:(")?[\\\?@a-zA-Z0-9_.]*?(?(3)"|)|{{.*}}))(?:\(|$)|undef|inttoptr|bitcast|null|asm).*$)') addrspace_end = re.compile(r"addrspace\(\d+\)\s*\*$") func_end = re.compile("(?:void.*|\)\s*)\*$") def conv(match, line): if not match or re.search(addrspace_end, match.group(1)) or not re.search(func_end, match.group(1)): return line return line[:match.start()] + match.group(1)[:match.group(1).rfind('*')].rstrip() + match.group(2) + line[match.end():] for line in sys.stdin: sys.stdout.write(conv(re.search(pat, line), line)) llvm-svn: 235145
2015-04-17 07:24:18 +08:00
%call5 = call i32 (i8*, ...) @printf(i8* getelementptr inbounds ([10 x i8], [10 x i8]* @.str, i32 0, i32 0), double %6, double %7, i32 %conv4)
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
call void @clear()
store float 9.000000e+00, float* @lx, align 4
store float 1.000000e+01, float* @ly, align 4
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%10 = load float, float* @lx, align 4
%11 = load float, float* @ly, align 4
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
call void @v_sf_sf(float %10, float %11)
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%12 = load float, float* @x, align 4
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%conv6 = fpext float %12 to double
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%13 = load float, float* @lx, align 4
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%conv7 = fpext float %13 to double
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%14 = load float, float* @y, align 4
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%conv8 = fpext float %14 to double
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%15 = load float, float* @ly, align 4
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%conv9 = fpext float %15 to double
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%16 = load float, float* @x, align 4
%17 = load float, float* @lx, align 4
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%cmp10 = fcmp oeq float %16, %17
br i1 %cmp10, label %land.rhs, label %land.end
land.rhs: ; preds = %entry
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%18 = load float, float* @y, align 4
%19 = load float, float* @ly, align 4
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%cmp12 = fcmp oeq float %18, %19
br label %land.end
land.end: ; preds = %land.rhs, %entry
%20 = phi i1 [ false, %entry ], [ %cmp12, %land.rhs ]
%land.ext = zext i1 %20 to i32
[opaque pointer type] Add textual IR support for explicit type parameter to the call instruction See r230786 and r230794 for similar changes to gep and load respectively. Call is a bit different because it often doesn't have a single explicit type - usually the type is deduced from the arguments, and just the return type is explicit. In those cases there's no need to change the IR. When that's not the case, the IR usually contains the pointer type of the first operand - but since typed pointers are going away, that representation is insufficient so I'm just stripping the "pointerness" of the explicit type away. This does make the IR a bit weird - it /sort of/ reads like the type of the first operand: "call void () %x(" but %x is actually of type "void ()*" and will eventually be just of type "ptr". But this seems not too bad and I don't think it would benefit from repeating the type ("void (), void () * %x(" and then eventually "void (), ptr %x(") as has been done with gep and load. This also has a side benefit: since the explicit type is no longer a pointer, there's no ambiguity between an explicit type and a function that returns a function pointer. Previously this case needed an explicit type (eg: a function returning a void() function was written as "call void () () * @x(" rather than "call void () * @x(" because of the ambiguity between a function returning a pointer to a void() function and a function returning void). No ambiguity means even function pointer return types can just be written alone, without writing the whole function's type. This leaves /only/ the varargs case where the explicit type is required. Given the special type syntax in call instructions, the regex-fu used for migration was a bit more involved in its own unique way (as every one of these is) so here it is. Use it in conjunction with the apply.sh script and associated find/xargs commands I've provided in rr230786 to migrate your out of tree tests. Do let me know if any of this doesn't cover your cases & we can iterate on a more general script/regexes to help others with out of tree tests. About 9 test cases couldn't be automatically migrated - half of those were functions returning function pointers, where I just had to manually delete the function argument types now that we didn't need an explicit function type there. The other half were typedefs of function types used in calls - just had to manually drop the * from those. import fileinput import sys import re pat = re.compile(r'((?:=|:|^|\s)call\s(?:[^@]*?))(\s*$|\s*(?:(?:\[\[[a-zA-Z0-9_]+\]\]|[@%](?:(")?[\\\?@a-zA-Z0-9_.]*?(?(3)"|)|{{.*}}))(?:\(|$)|undef|inttoptr|bitcast|null|asm).*$)') addrspace_end = re.compile(r"addrspace\(\d+\)\s*\*$") func_end = re.compile("(?:void.*|\)\s*)\*$") def conv(match, line): if not match or re.search(addrspace_end, match.group(1)) or not re.search(func_end, match.group(1)): return line return line[:match.start()] + match.group(1)[:match.group(1).rfind('*')].rstrip() + match.group(2) + line[match.end():] for line in sys.stdin: sys.stdout.write(conv(re.search(pat, line), line)) llvm-svn: 235145
2015-04-17 07:24:18 +08:00
%call14 = call i32 (i8*, ...) @printf(i8* getelementptr inbounds ([16 x i8], [16 x i8]* @.str1, i32 0, i32 0), double %conv6, double %conv7, double %conv8, double %conv9, i32 %land.ext)
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
call void @clear()
store float 0x3FFE666660000000, float* @lx, align 4
store double 0x4007E613249FF279, double* @lyd, align 8
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%21 = load float, float* @lx, align 4
%22 = load double, double* @lyd, align 8
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
call void @v_sf_df(float %21, double %22)
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%23 = load float, float* @x, align 4
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%conv15 = fpext float %23 to double
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%24 = load float, float* @lx, align 4
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%conv16 = fpext float %24 to double
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%25 = load double, double* @yd, align 8
%26 = load double, double* @lyd, align 8
%27 = load float, float* @x, align 4
%28 = load float, float* @lx, align 4
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%cmp17 = fcmp oeq float %27, %28
%conv18 = zext i1 %cmp17 to i32
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%29 = load double, double* @yd, align 8
%30 = load double, double* @lyd, align 8
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%cmp19 = fcmp oeq double %29, %30
%conv20 = zext i1 %cmp19 to i32
%and = and i32 %conv18, %conv20
[opaque pointer type] Add textual IR support for explicit type parameter to the call instruction See r230786 and r230794 for similar changes to gep and load respectively. Call is a bit different because it often doesn't have a single explicit type - usually the type is deduced from the arguments, and just the return type is explicit. In those cases there's no need to change the IR. When that's not the case, the IR usually contains the pointer type of the first operand - but since typed pointers are going away, that representation is insufficient so I'm just stripping the "pointerness" of the explicit type away. This does make the IR a bit weird - it /sort of/ reads like the type of the first operand: "call void () %x(" but %x is actually of type "void ()*" and will eventually be just of type "ptr". But this seems not too bad and I don't think it would benefit from repeating the type ("void (), void () * %x(" and then eventually "void (), ptr %x(") as has been done with gep and load. This also has a side benefit: since the explicit type is no longer a pointer, there's no ambiguity between an explicit type and a function that returns a function pointer. Previously this case needed an explicit type (eg: a function returning a void() function was written as "call void () () * @x(" rather than "call void () * @x(" because of the ambiguity between a function returning a pointer to a void() function and a function returning void). No ambiguity means even function pointer return types can just be written alone, without writing the whole function's type. This leaves /only/ the varargs case where the explicit type is required. Given the special type syntax in call instructions, the regex-fu used for migration was a bit more involved in its own unique way (as every one of these is) so here it is. Use it in conjunction with the apply.sh script and associated find/xargs commands I've provided in rr230786 to migrate your out of tree tests. Do let me know if any of this doesn't cover your cases & we can iterate on a more general script/regexes to help others with out of tree tests. About 9 test cases couldn't be automatically migrated - half of those were functions returning function pointers, where I just had to manually delete the function argument types now that we didn't need an explicit function type there. The other half were typedefs of function types used in calls - just had to manually drop the * from those. import fileinput import sys import re pat = re.compile(r'((?:=|:|^|\s)call\s(?:[^@]*?))(\s*$|\s*(?:(?:\[\[[a-zA-Z0-9_]+\]\]|[@%](?:(")?[\\\?@a-zA-Z0-9_.]*?(?(3)"|)|{{.*}}))(?:\(|$)|undef|inttoptr|bitcast|null|asm).*$)') addrspace_end = re.compile(r"addrspace\(\d+\)\s*\*$") func_end = re.compile("(?:void.*|\)\s*)\*$") def conv(match, line): if not match or re.search(addrspace_end, match.group(1)) or not re.search(func_end, match.group(1)): return line return line[:match.start()] + match.group(1)[:match.group(1).rfind('*')].rstrip() + match.group(2) + line[match.end():] for line in sys.stdin: sys.stdout.write(conv(re.search(pat, line), line)) llvm-svn: 235145
2015-04-17 07:24:18 +08:00
%call21 = call i32 (i8*, ...) @printf(i8* getelementptr inbounds ([16 x i8], [16 x i8]* @.str1, i32 0, i32 0), double %conv15, double %conv16, double %25, double %26, i32 %and)
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
call void @clear()
store double 0x4194E54F94000000, double* @lxd, align 8
store float 7.600000e+01, float* @ly, align 4
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%31 = load double, double* @lxd, align 8
%32 = load float, float* @ly, align 4
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
call void @v_df_sf(double %31, float %32)
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%33 = load double, double* @xd, align 8
%34 = load double, double* @lxd, align 8
%35 = load float, float* @y, align 4
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%conv22 = fpext float %35 to double
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%36 = load float, float* @ly, align 4
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%conv23 = fpext float %36 to double
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%37 = load double, double* @xd, align 8
%38 = load double, double* @lxd, align 8
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%cmp24 = fcmp oeq double %37, %38
%conv25 = zext i1 %cmp24 to i32
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%39 = load float, float* @y, align 4
%40 = load float, float* @ly, align 4
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%cmp26 = fcmp oeq float %39, %40
%conv27 = zext i1 %cmp26 to i32
%and28 = and i32 %conv25, %conv27
[opaque pointer type] Add textual IR support for explicit type parameter to the call instruction See r230786 and r230794 for similar changes to gep and load respectively. Call is a bit different because it often doesn't have a single explicit type - usually the type is deduced from the arguments, and just the return type is explicit. In those cases there's no need to change the IR. When that's not the case, the IR usually contains the pointer type of the first operand - but since typed pointers are going away, that representation is insufficient so I'm just stripping the "pointerness" of the explicit type away. This does make the IR a bit weird - it /sort of/ reads like the type of the first operand: "call void () %x(" but %x is actually of type "void ()*" and will eventually be just of type "ptr". But this seems not too bad and I don't think it would benefit from repeating the type ("void (), void () * %x(" and then eventually "void (), ptr %x(") as has been done with gep and load. This also has a side benefit: since the explicit type is no longer a pointer, there's no ambiguity between an explicit type and a function that returns a function pointer. Previously this case needed an explicit type (eg: a function returning a void() function was written as "call void () () * @x(" rather than "call void () * @x(" because of the ambiguity between a function returning a pointer to a void() function and a function returning void). No ambiguity means even function pointer return types can just be written alone, without writing the whole function's type. This leaves /only/ the varargs case where the explicit type is required. Given the special type syntax in call instructions, the regex-fu used for migration was a bit more involved in its own unique way (as every one of these is) so here it is. Use it in conjunction with the apply.sh script and associated find/xargs commands I've provided in rr230786 to migrate your out of tree tests. Do let me know if any of this doesn't cover your cases & we can iterate on a more general script/regexes to help others with out of tree tests. About 9 test cases couldn't be automatically migrated - half of those were functions returning function pointers, where I just had to manually delete the function argument types now that we didn't need an explicit function type there. The other half were typedefs of function types used in calls - just had to manually drop the * from those. import fileinput import sys import re pat = re.compile(r'((?:=|:|^|\s)call\s(?:[^@]*?))(\s*$|\s*(?:(?:\[\[[a-zA-Z0-9_]+\]\]|[@%](?:(")?[\\\?@a-zA-Z0-9_.]*?(?(3)"|)|{{.*}}))(?:\(|$)|undef|inttoptr|bitcast|null|asm).*$)') addrspace_end = re.compile(r"addrspace\(\d+\)\s*\*$") func_end = re.compile("(?:void.*|\)\s*)\*$") def conv(match, line): if not match or re.search(addrspace_end, match.group(1)) or not re.search(func_end, match.group(1)): return line return line[:match.start()] + match.group(1)[:match.group(1).rfind('*')].rstrip() + match.group(2) + line[match.end():] for line in sys.stdin: sys.stdout.write(conv(re.search(pat, line), line)) llvm-svn: 235145
2015-04-17 07:24:18 +08:00
%call29 = call i32 (i8*, ...) @printf(i8* getelementptr inbounds ([16 x i8], [16 x i8]* @.str1, i32 0, i32 0), double %33, double %34, double %conv22, double %conv23, i32 %and28)
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
call void @clear()
store double 7.365198e+07, double* @lxd, align 8
store double 0x416536CD80000000, double* @lyd, align 8
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%41 = load double, double* @lxd, align 8
%42 = load double, double* @lyd, align 8
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
call void @v_df_df(double %41, double %42)
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%43 = load double, double* @xd, align 8
%44 = load double, double* @lxd, align 8
%45 = load double, double* @yd, align 8
%46 = load double, double* @lyd, align 8
%47 = load double, double* @xd, align 8
%48 = load double, double* @lxd, align 8
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%cmp30 = fcmp oeq double %47, %48
%conv31 = zext i1 %cmp30 to i32
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%49 = load double, double* @yd, align 8
%50 = load double, double* @lyd, align 8
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%cmp32 = fcmp oeq double %49, %50
%conv33 = zext i1 %cmp32 to i32
%and34 = and i32 %conv31, %conv33
[opaque pointer type] Add textual IR support for explicit type parameter to the call instruction See r230786 and r230794 for similar changes to gep and load respectively. Call is a bit different because it often doesn't have a single explicit type - usually the type is deduced from the arguments, and just the return type is explicit. In those cases there's no need to change the IR. When that's not the case, the IR usually contains the pointer type of the first operand - but since typed pointers are going away, that representation is insufficient so I'm just stripping the "pointerness" of the explicit type away. This does make the IR a bit weird - it /sort of/ reads like the type of the first operand: "call void () %x(" but %x is actually of type "void ()*" and will eventually be just of type "ptr". But this seems not too bad and I don't think it would benefit from repeating the type ("void (), void () * %x(" and then eventually "void (), ptr %x(") as has been done with gep and load. This also has a side benefit: since the explicit type is no longer a pointer, there's no ambiguity between an explicit type and a function that returns a function pointer. Previously this case needed an explicit type (eg: a function returning a void() function was written as "call void () () * @x(" rather than "call void () * @x(" because of the ambiguity between a function returning a pointer to a void() function and a function returning void). No ambiguity means even function pointer return types can just be written alone, without writing the whole function's type. This leaves /only/ the varargs case where the explicit type is required. Given the special type syntax in call instructions, the regex-fu used for migration was a bit more involved in its own unique way (as every one of these is) so here it is. Use it in conjunction with the apply.sh script and associated find/xargs commands I've provided in rr230786 to migrate your out of tree tests. Do let me know if any of this doesn't cover your cases & we can iterate on a more general script/regexes to help others with out of tree tests. About 9 test cases couldn't be automatically migrated - half of those were functions returning function pointers, where I just had to manually delete the function argument types now that we didn't need an explicit function type there. The other half were typedefs of function types used in calls - just had to manually drop the * from those. import fileinput import sys import re pat = re.compile(r'((?:=|:|^|\s)call\s(?:[^@]*?))(\s*$|\s*(?:(?:\[\[[a-zA-Z0-9_]+\]\]|[@%](?:(")?[\\\?@a-zA-Z0-9_.]*?(?(3)"|)|{{.*}}))(?:\(|$)|undef|inttoptr|bitcast|null|asm).*$)') addrspace_end = re.compile(r"addrspace\(\d+\)\s*\*$") func_end = re.compile("(?:void.*|\)\s*)\*$") def conv(match, line): if not match or re.search(addrspace_end, match.group(1)) or not re.search(func_end, match.group(1)): return line return line[:match.start()] + match.group(1)[:match.group(1).rfind('*')].rstrip() + match.group(2) + line[match.end():] for line in sys.stdin: sys.stdout.write(conv(re.search(pat, line), line)) llvm-svn: 235145
2015-04-17 07:24:18 +08:00
%call35 = call i32 (i8*, ...) @printf(i8* getelementptr inbounds ([16 x i8], [16 x i8]* @.str1, i32 0, i32 0), double %43, double %44, double %45, double %46, i32 %and34)
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
call void @clear()
store float 0x4016666660000000, float* @ret_sf, align 4
%call36 = call float @sf_v()
store float %call36, float* @lret_sf, align 4
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%51 = load float, float* @ret_sf, align 4
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%conv37 = fpext float %51 to double
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%52 = load float, float* @lret_sf, align 4
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%conv38 = fpext float %52 to double
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%53 = load float, float* @ret_sf, align 4
%54 = load float, float* @lret_sf, align 4
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%cmp39 = fcmp oeq float %53, %54
%conv40 = zext i1 %cmp39 to i32
[opaque pointer type] Add textual IR support for explicit type parameter to the call instruction See r230786 and r230794 for similar changes to gep and load respectively. Call is a bit different because it often doesn't have a single explicit type - usually the type is deduced from the arguments, and just the return type is explicit. In those cases there's no need to change the IR. When that's not the case, the IR usually contains the pointer type of the first operand - but since typed pointers are going away, that representation is insufficient so I'm just stripping the "pointerness" of the explicit type away. This does make the IR a bit weird - it /sort of/ reads like the type of the first operand: "call void () %x(" but %x is actually of type "void ()*" and will eventually be just of type "ptr". But this seems not too bad and I don't think it would benefit from repeating the type ("void (), void () * %x(" and then eventually "void (), ptr %x(") as has been done with gep and load. This also has a side benefit: since the explicit type is no longer a pointer, there's no ambiguity between an explicit type and a function that returns a function pointer. Previously this case needed an explicit type (eg: a function returning a void() function was written as "call void () () * @x(" rather than "call void () * @x(" because of the ambiguity between a function returning a pointer to a void() function and a function returning void). No ambiguity means even function pointer return types can just be written alone, without writing the whole function's type. This leaves /only/ the varargs case where the explicit type is required. Given the special type syntax in call instructions, the regex-fu used for migration was a bit more involved in its own unique way (as every one of these is) so here it is. Use it in conjunction with the apply.sh script and associated find/xargs commands I've provided in rr230786 to migrate your out of tree tests. Do let me know if any of this doesn't cover your cases & we can iterate on a more general script/regexes to help others with out of tree tests. About 9 test cases couldn't be automatically migrated - half of those were functions returning function pointers, where I just had to manually delete the function argument types now that we didn't need an explicit function type there. The other half were typedefs of function types used in calls - just had to manually drop the * from those. import fileinput import sys import re pat = re.compile(r'((?:=|:|^|\s)call\s(?:[^@]*?))(\s*$|\s*(?:(?:\[\[[a-zA-Z0-9_]+\]\]|[@%](?:(")?[\\\?@a-zA-Z0-9_.]*?(?(3)"|)|{{.*}}))(?:\(|$)|undef|inttoptr|bitcast|null|asm).*$)') addrspace_end = re.compile(r"addrspace\(\d+\)\s*\*$") func_end = re.compile("(?:void.*|\)\s*)\*$") def conv(match, line): if not match or re.search(addrspace_end, match.group(1)) or not re.search(func_end, match.group(1)): return line return line[:match.start()] + match.group(1)[:match.group(1).rfind('*')].rstrip() + match.group(2) + line[match.end():] for line in sys.stdin: sys.stdout.write(conv(re.search(pat, line), line)) llvm-svn: 235145
2015-04-17 07:24:18 +08:00
%call41 = call i32 (i8*, ...) @printf(i8* getelementptr inbounds ([10 x i8], [10 x i8]* @.str, i32 0, i32 0), double %conv37, double %conv38, i32 %conv40)
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
call void @clear()
store float 4.587300e+06, float* @ret_sf, align 4
store float 3.420000e+02, float* @lx, align 4
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%55 = load float, float* @lx, align 4
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%call42 = call float @sf_sf(float %55)
store float %call42, float* @lret_sf, align 4
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%56 = load float, float* @ret_sf, align 4
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%conv43 = fpext float %56 to double
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%57 = load float, float* @lret_sf, align 4
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%conv44 = fpext float %57 to double
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%58 = load float, float* @x, align 4
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%conv45 = fpext float %58 to double
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%59 = load float, float* @lx, align 4
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%conv46 = fpext float %59 to double
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%60 = load float, float* @ret_sf, align 4
%61 = load float, float* @lret_sf, align 4
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%cmp47 = fcmp oeq float %60, %61
%conv48 = zext i1 %cmp47 to i32
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%62 = load float, float* @x, align 4
%63 = load float, float* @lx, align 4
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%cmp49 = fcmp oeq float %62, %63
%conv50 = zext i1 %cmp49 to i32
%and51 = and i32 %conv48, %conv50
[opaque pointer type] Add textual IR support for explicit type parameter to the call instruction See r230786 and r230794 for similar changes to gep and load respectively. Call is a bit different because it often doesn't have a single explicit type - usually the type is deduced from the arguments, and just the return type is explicit. In those cases there's no need to change the IR. When that's not the case, the IR usually contains the pointer type of the first operand - but since typed pointers are going away, that representation is insufficient so I'm just stripping the "pointerness" of the explicit type away. This does make the IR a bit weird - it /sort of/ reads like the type of the first operand: "call void () %x(" but %x is actually of type "void ()*" and will eventually be just of type "ptr". But this seems not too bad and I don't think it would benefit from repeating the type ("void (), void () * %x(" and then eventually "void (), ptr %x(") as has been done with gep and load. This also has a side benefit: since the explicit type is no longer a pointer, there's no ambiguity between an explicit type and a function that returns a function pointer. Previously this case needed an explicit type (eg: a function returning a void() function was written as "call void () () * @x(" rather than "call void () * @x(" because of the ambiguity between a function returning a pointer to a void() function and a function returning void). No ambiguity means even function pointer return types can just be written alone, without writing the whole function's type. This leaves /only/ the varargs case where the explicit type is required. Given the special type syntax in call instructions, the regex-fu used for migration was a bit more involved in its own unique way (as every one of these is) so here it is. Use it in conjunction with the apply.sh script and associated find/xargs commands I've provided in rr230786 to migrate your out of tree tests. Do let me know if any of this doesn't cover your cases & we can iterate on a more general script/regexes to help others with out of tree tests. About 9 test cases couldn't be automatically migrated - half of those were functions returning function pointers, where I just had to manually delete the function argument types now that we didn't need an explicit function type there. The other half were typedefs of function types used in calls - just had to manually drop the * from those. import fileinput import sys import re pat = re.compile(r'((?:=|:|^|\s)call\s(?:[^@]*?))(\s*$|\s*(?:(?:\[\[[a-zA-Z0-9_]+\]\]|[@%](?:(")?[\\\?@a-zA-Z0-9_.]*?(?(3)"|)|{{.*}}))(?:\(|$)|undef|inttoptr|bitcast|null|asm).*$)') addrspace_end = re.compile(r"addrspace\(\d+\)\s*\*$") func_end = re.compile("(?:void.*|\)\s*)\*$") def conv(match, line): if not match or re.search(addrspace_end, match.group(1)) or not re.search(func_end, match.group(1)): return line return line[:match.start()] + match.group(1)[:match.group(1).rfind('*')].rstrip() + match.group(2) + line[match.end():] for line in sys.stdin: sys.stdout.write(conv(re.search(pat, line), line)) llvm-svn: 235145
2015-04-17 07:24:18 +08:00
%call52 = call i32 (i8*, ...) @printf(i8* getelementptr inbounds ([16 x i8], [16 x i8]* @.str1, i32 0, i32 0), double %conv43, double %conv44, double %conv45, double %conv46, i32 %and51)
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
call void @clear()
store float 4.445910e+06, float* @ret_sf, align 4
store double 0x419A7DB294000000, double* @lxd, align 8
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%64 = load double, double* @lxd, align 8
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%call53 = call float @sf_df(double %64)
store float %call53, float* @lret_sf, align 4
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%65 = load float, float* @ret_sf, align 4
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%conv54 = fpext float %65 to double
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%66 = load float, float* @lret_sf, align 4
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%conv55 = fpext float %66 to double
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%67 = load double, double* @xd, align 8
%68 = load double, double* @lxd, align 8
%69 = load float, float* @ret_sf, align 4
%70 = load float, float* @lret_sf, align 4
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%cmp56 = fcmp oeq float %69, %70
%conv57 = zext i1 %cmp56 to i32
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%71 = load double, double* @xd, align 8
%72 = load double, double* @lxd, align 8
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%cmp58 = fcmp oeq double %71, %72
%conv59 = zext i1 %cmp58 to i32
%and60 = and i32 %conv57, %conv59
[opaque pointer type] Add textual IR support for explicit type parameter to the call instruction See r230786 and r230794 for similar changes to gep and load respectively. Call is a bit different because it often doesn't have a single explicit type - usually the type is deduced from the arguments, and just the return type is explicit. In those cases there's no need to change the IR. When that's not the case, the IR usually contains the pointer type of the first operand - but since typed pointers are going away, that representation is insufficient so I'm just stripping the "pointerness" of the explicit type away. This does make the IR a bit weird - it /sort of/ reads like the type of the first operand: "call void () %x(" but %x is actually of type "void ()*" and will eventually be just of type "ptr". But this seems not too bad and I don't think it would benefit from repeating the type ("void (), void () * %x(" and then eventually "void (), ptr %x(") as has been done with gep and load. This also has a side benefit: since the explicit type is no longer a pointer, there's no ambiguity between an explicit type and a function that returns a function pointer. Previously this case needed an explicit type (eg: a function returning a void() function was written as "call void () () * @x(" rather than "call void () * @x(" because of the ambiguity between a function returning a pointer to a void() function and a function returning void). No ambiguity means even function pointer return types can just be written alone, without writing the whole function's type. This leaves /only/ the varargs case where the explicit type is required. Given the special type syntax in call instructions, the regex-fu used for migration was a bit more involved in its own unique way (as every one of these is) so here it is. Use it in conjunction with the apply.sh script and associated find/xargs commands I've provided in rr230786 to migrate your out of tree tests. Do let me know if any of this doesn't cover your cases & we can iterate on a more general script/regexes to help others with out of tree tests. About 9 test cases couldn't be automatically migrated - half of those were functions returning function pointers, where I just had to manually delete the function argument types now that we didn't need an explicit function type there. The other half were typedefs of function types used in calls - just had to manually drop the * from those. import fileinput import sys import re pat = re.compile(r'((?:=|:|^|\s)call\s(?:[^@]*?))(\s*$|\s*(?:(?:\[\[[a-zA-Z0-9_]+\]\]|[@%](?:(")?[\\\?@a-zA-Z0-9_.]*?(?(3)"|)|{{.*}}))(?:\(|$)|undef|inttoptr|bitcast|null|asm).*$)') addrspace_end = re.compile(r"addrspace\(\d+\)\s*\*$") func_end = re.compile("(?:void.*|\)\s*)\*$") def conv(match, line): if not match or re.search(addrspace_end, match.group(1)) or not re.search(func_end, match.group(1)): return line return line[:match.start()] + match.group(1)[:match.group(1).rfind('*')].rstrip() + match.group(2) + line[match.end():] for line in sys.stdin: sys.stdout.write(conv(re.search(pat, line), line)) llvm-svn: 235145
2015-04-17 07:24:18 +08:00
%call61 = call i32 (i8*, ...) @printf(i8* getelementptr inbounds ([16 x i8], [16 x i8]* @.str1, i32 0, i32 0), double %conv54, double %conv55, double %67, double %68, i32 %and60)
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
call void @clear()
store float 0x3FFF4BC6A0000000, float* @ret_sf, align 4
store float 4.445500e+03, float* @lx, align 4
store float 0x4068ACCCC0000000, float* @ly, align 4
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%73 = load float, float* @lx, align 4
%74 = load float, float* @ly, align 4
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%call62 = call float @sf_sf_sf(float %73, float %74)
store float %call62, float* @lret_sf, align 4
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%75 = load float, float* @ret_sf, align 4
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%conv63 = fpext float %75 to double
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%76 = load float, float* @lret_sf, align 4
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%conv64 = fpext float %76 to double
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%77 = load float, float* @x, align 4
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%conv65 = fpext float %77 to double
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%78 = load float, float* @lx, align 4
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%conv66 = fpext float %78 to double
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%79 = load float, float* @y, align 4
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%conv67 = fpext float %79 to double
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%80 = load float, float* @ly, align 4
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%conv68 = fpext float %80 to double
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%81 = load float, float* @ret_sf, align 4
%82 = load float, float* @lret_sf, align 4
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%cmp69 = fcmp oeq float %81, %82
br i1 %cmp69, label %land.lhs.true, label %land.end76
land.lhs.true: ; preds = %land.end
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%83 = load float, float* @x, align 4
%84 = load float, float* @lx, align 4
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%cmp71 = fcmp oeq float %83, %84
br i1 %cmp71, label %land.rhs73, label %land.end76
land.rhs73: ; preds = %land.lhs.true
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%85 = load float, float* @y, align 4
%86 = load float, float* @ly, align 4
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%cmp74 = fcmp oeq float %85, %86
br label %land.end76
land.end76: ; preds = %land.rhs73, %land.lhs.true, %land.end
%87 = phi i1 [ false, %land.lhs.true ], [ false, %land.end ], [ %cmp74, %land.rhs73 ]
%land.ext77 = zext i1 %87 to i32
[opaque pointer type] Add textual IR support for explicit type parameter to the call instruction See r230786 and r230794 for similar changes to gep and load respectively. Call is a bit different because it often doesn't have a single explicit type - usually the type is deduced from the arguments, and just the return type is explicit. In those cases there's no need to change the IR. When that's not the case, the IR usually contains the pointer type of the first operand - but since typed pointers are going away, that representation is insufficient so I'm just stripping the "pointerness" of the explicit type away. This does make the IR a bit weird - it /sort of/ reads like the type of the first operand: "call void () %x(" but %x is actually of type "void ()*" and will eventually be just of type "ptr". But this seems not too bad and I don't think it would benefit from repeating the type ("void (), void () * %x(" and then eventually "void (), ptr %x(") as has been done with gep and load. This also has a side benefit: since the explicit type is no longer a pointer, there's no ambiguity between an explicit type and a function that returns a function pointer. Previously this case needed an explicit type (eg: a function returning a void() function was written as "call void () () * @x(" rather than "call void () * @x(" because of the ambiguity between a function returning a pointer to a void() function and a function returning void). No ambiguity means even function pointer return types can just be written alone, without writing the whole function's type. This leaves /only/ the varargs case where the explicit type is required. Given the special type syntax in call instructions, the regex-fu used for migration was a bit more involved in its own unique way (as every one of these is) so here it is. Use it in conjunction with the apply.sh script and associated find/xargs commands I've provided in rr230786 to migrate your out of tree tests. Do let me know if any of this doesn't cover your cases & we can iterate on a more general script/regexes to help others with out of tree tests. About 9 test cases couldn't be automatically migrated - half of those were functions returning function pointers, where I just had to manually delete the function argument types now that we didn't need an explicit function type there. The other half were typedefs of function types used in calls - just had to manually drop the * from those. import fileinput import sys import re pat = re.compile(r'((?:=|:|^|\s)call\s(?:[^@]*?))(\s*$|\s*(?:(?:\[\[[a-zA-Z0-9_]+\]\]|[@%](?:(")?[\\\?@a-zA-Z0-9_.]*?(?(3)"|)|{{.*}}))(?:\(|$)|undef|inttoptr|bitcast|null|asm).*$)') addrspace_end = re.compile(r"addrspace\(\d+\)\s*\*$") func_end = re.compile("(?:void.*|\)\s*)\*$") def conv(match, line): if not match or re.search(addrspace_end, match.group(1)) or not re.search(func_end, match.group(1)): return line return line[:match.start()] + match.group(1)[:match.group(1).rfind('*')].rstrip() + match.group(2) + line[match.end():] for line in sys.stdin: sys.stdout.write(conv(re.search(pat, line), line)) llvm-svn: 235145
2015-04-17 07:24:18 +08:00
%call78 = call i32 (i8*, ...) @printf(i8* getelementptr inbounds ([22 x i8], [22 x i8]* @.str2, i32 0, i32 0), double %conv63, double %conv64, double %conv65, double %conv66, double %conv67, double %conv68, i32 %land.ext77)
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
call void @clear()
store float 9.991300e+04, float* @ret_sf, align 4
store float 1.114500e+04, float* @lx, align 4
store double 9.994445e+07, double* @lyd, align 8
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%88 = load float, float* @lx, align 4
%89 = load double, double* @lyd, align 8
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%call79 = call float @sf_sf_df(float %88, double %89)
store float %call79, float* @lret_sf, align 4
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%90 = load float, float* @ret_sf, align 4
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%conv80 = fpext float %90 to double
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%91 = load float, float* @lret_sf, align 4
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%conv81 = fpext float %91 to double
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%92 = load float, float* @x, align 4
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%conv82 = fpext float %92 to double
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%93 = load float, float* @lx, align 4
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%conv83 = fpext float %93 to double
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%94 = load double, double* @yd, align 8
%95 = load double, double* @lyd, align 8
%96 = load float, float* @ret_sf, align 4
%97 = load float, float* @lret_sf, align 4
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%cmp84 = fcmp oeq float %96, %97
br i1 %cmp84, label %land.lhs.true86, label %land.end92
land.lhs.true86: ; preds = %land.end76
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%98 = load float, float* @x, align 4
%99 = load float, float* @lx, align 4
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%cmp87 = fcmp oeq float %98, %99
br i1 %cmp87, label %land.rhs89, label %land.end92
land.rhs89: ; preds = %land.lhs.true86
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%100 = load double, double* @yd, align 8
%101 = load double, double* @lyd, align 8
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%cmp90 = fcmp oeq double %100, %101
br label %land.end92
land.end92: ; preds = %land.rhs89, %land.lhs.true86, %land.end76
%102 = phi i1 [ false, %land.lhs.true86 ], [ false, %land.end76 ], [ %cmp90, %land.rhs89 ]
%land.ext93 = zext i1 %102 to i32
[opaque pointer type] Add textual IR support for explicit type parameter to the call instruction See r230786 and r230794 for similar changes to gep and load respectively. Call is a bit different because it often doesn't have a single explicit type - usually the type is deduced from the arguments, and just the return type is explicit. In those cases there's no need to change the IR. When that's not the case, the IR usually contains the pointer type of the first operand - but since typed pointers are going away, that representation is insufficient so I'm just stripping the "pointerness" of the explicit type away. This does make the IR a bit weird - it /sort of/ reads like the type of the first operand: "call void () %x(" but %x is actually of type "void ()*" and will eventually be just of type "ptr". But this seems not too bad and I don't think it would benefit from repeating the type ("void (), void () * %x(" and then eventually "void (), ptr %x(") as has been done with gep and load. This also has a side benefit: since the explicit type is no longer a pointer, there's no ambiguity between an explicit type and a function that returns a function pointer. Previously this case needed an explicit type (eg: a function returning a void() function was written as "call void () () * @x(" rather than "call void () * @x(" because of the ambiguity between a function returning a pointer to a void() function and a function returning void). No ambiguity means even function pointer return types can just be written alone, without writing the whole function's type. This leaves /only/ the varargs case where the explicit type is required. Given the special type syntax in call instructions, the regex-fu used for migration was a bit more involved in its own unique way (as every one of these is) so here it is. Use it in conjunction with the apply.sh script and associated find/xargs commands I've provided in rr230786 to migrate your out of tree tests. Do let me know if any of this doesn't cover your cases & we can iterate on a more general script/regexes to help others with out of tree tests. About 9 test cases couldn't be automatically migrated - half of those were functions returning function pointers, where I just had to manually delete the function argument types now that we didn't need an explicit function type there. The other half were typedefs of function types used in calls - just had to manually drop the * from those. import fileinput import sys import re pat = re.compile(r'((?:=|:|^|\s)call\s(?:[^@]*?))(\s*$|\s*(?:(?:\[\[[a-zA-Z0-9_]+\]\]|[@%](?:(")?[\\\?@a-zA-Z0-9_.]*?(?(3)"|)|{{.*}}))(?:\(|$)|undef|inttoptr|bitcast|null|asm).*$)') addrspace_end = re.compile(r"addrspace\(\d+\)\s*\*$") func_end = re.compile("(?:void.*|\)\s*)\*$") def conv(match, line): if not match or re.search(addrspace_end, match.group(1)) or not re.search(func_end, match.group(1)): return line return line[:match.start()] + match.group(1)[:match.group(1).rfind('*')].rstrip() + match.group(2) + line[match.end():] for line in sys.stdin: sys.stdout.write(conv(re.search(pat, line), line)) llvm-svn: 235145
2015-04-17 07:24:18 +08:00
%call94 = call i32 (i8*, ...) @printf(i8* getelementptr inbounds ([22 x i8], [22 x i8]* @.str2, i32 0, i32 0), double %conv80, double %conv81, double %conv82, double %conv83, double %94, double %95, i32 %land.ext93)
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
call void @clear()
store float 0x417CCC7A00000000, float* @ret_sf, align 4
store double 0x4172034530000000, double* @lxd, align 8
store float 4.456200e+04, float* @ly, align 4
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%103 = load double, double* @lxd, align 8
%104 = load float, float* @ly, align 4
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%call95 = call float @sf_df_sf(double %103, float %104)
store float %call95, float* @lret_sf, align 4
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%105 = load float, float* @ret_sf, align 4
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%conv96 = fpext float %105 to double
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%106 = load float, float* @lret_sf, align 4
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%conv97 = fpext float %106 to double
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%107 = load double, double* @xd, align 8
%108 = load double, double* @lxd, align 8
%109 = load float, float* @y, align 4
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%conv98 = fpext float %109 to double
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%110 = load float, float* @ly, align 4
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%conv99 = fpext float %110 to double
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%111 = load float, float* @ret_sf, align 4
%112 = load float, float* @lret_sf, align 4
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%cmp100 = fcmp oeq float %111, %112
br i1 %cmp100, label %land.lhs.true102, label %land.end108
land.lhs.true102: ; preds = %land.end92
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%113 = load double, double* @xd, align 8
%114 = load double, double* @lxd, align 8
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%cmp103 = fcmp oeq double %113, %114
br i1 %cmp103, label %land.rhs105, label %land.end108
land.rhs105: ; preds = %land.lhs.true102
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%115 = load float, float* @y, align 4
%116 = load float, float* @ly, align 4
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%cmp106 = fcmp oeq float %115, %116
br label %land.end108
land.end108: ; preds = %land.rhs105, %land.lhs.true102, %land.end92
%117 = phi i1 [ false, %land.lhs.true102 ], [ false, %land.end92 ], [ %cmp106, %land.rhs105 ]
%land.ext109 = zext i1 %117 to i32
[opaque pointer type] Add textual IR support for explicit type parameter to the call instruction See r230786 and r230794 for similar changes to gep and load respectively. Call is a bit different because it often doesn't have a single explicit type - usually the type is deduced from the arguments, and just the return type is explicit. In those cases there's no need to change the IR. When that's not the case, the IR usually contains the pointer type of the first operand - but since typed pointers are going away, that representation is insufficient so I'm just stripping the "pointerness" of the explicit type away. This does make the IR a bit weird - it /sort of/ reads like the type of the first operand: "call void () %x(" but %x is actually of type "void ()*" and will eventually be just of type "ptr". But this seems not too bad and I don't think it would benefit from repeating the type ("void (), void () * %x(" and then eventually "void (), ptr %x(") as has been done with gep and load. This also has a side benefit: since the explicit type is no longer a pointer, there's no ambiguity between an explicit type and a function that returns a function pointer. Previously this case needed an explicit type (eg: a function returning a void() function was written as "call void () () * @x(" rather than "call void () * @x(" because of the ambiguity between a function returning a pointer to a void() function and a function returning void). No ambiguity means even function pointer return types can just be written alone, without writing the whole function's type. This leaves /only/ the varargs case where the explicit type is required. Given the special type syntax in call instructions, the regex-fu used for migration was a bit more involved in its own unique way (as every one of these is) so here it is. Use it in conjunction with the apply.sh script and associated find/xargs commands I've provided in rr230786 to migrate your out of tree tests. Do let me know if any of this doesn't cover your cases & we can iterate on a more general script/regexes to help others with out of tree tests. About 9 test cases couldn't be automatically migrated - half of those were functions returning function pointers, where I just had to manually delete the function argument types now that we didn't need an explicit function type there. The other half were typedefs of function types used in calls - just had to manually drop the * from those. import fileinput import sys import re pat = re.compile(r'((?:=|:|^|\s)call\s(?:[^@]*?))(\s*$|\s*(?:(?:\[\[[a-zA-Z0-9_]+\]\]|[@%](?:(")?[\\\?@a-zA-Z0-9_.]*?(?(3)"|)|{{.*}}))(?:\(|$)|undef|inttoptr|bitcast|null|asm).*$)') addrspace_end = re.compile(r"addrspace\(\d+\)\s*\*$") func_end = re.compile("(?:void.*|\)\s*)\*$") def conv(match, line): if not match or re.search(addrspace_end, match.group(1)) or not re.search(func_end, match.group(1)): return line return line[:match.start()] + match.group(1)[:match.group(1).rfind('*')].rstrip() + match.group(2) + line[match.end():] for line in sys.stdin: sys.stdout.write(conv(re.search(pat, line), line)) llvm-svn: 235145
2015-04-17 07:24:18 +08:00
%call110 = call i32 (i8*, ...) @printf(i8* getelementptr inbounds ([22 x i8], [22 x i8]* @.str2, i32 0, i32 0), double %conv96, double %conv97, double %107, double %108, double %conv98, double %conv99, i32 %land.ext109)
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
call void @clear()
store float 3.987721e+06, float* @ret_sf, align 4
store double 0x3FF1F49F6DDDC2D8, double* @lxd, align 8
store double 0x409129F306A2B170, double* @lyd, align 8
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%118 = load double, double* @lxd, align 8
%119 = load double, double* @lyd, align 8
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%call111 = call float @sf_df_df(double %118, double %119)
store float %call111, float* @lret_sf, align 4
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%120 = load float, float* @ret_sf, align 4
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%conv112 = fpext float %120 to double
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%121 = load float, float* @lret_sf, align 4
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%conv113 = fpext float %121 to double
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%122 = load double, double* @xd, align 8
%123 = load double, double* @lxd, align 8
%124 = load double, double* @yd, align 8
%125 = load double, double* @lyd, align 8
%126 = load float, float* @ret_sf, align 4
%127 = load float, float* @lret_sf, align 4
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%cmp114 = fcmp oeq float %126, %127
br i1 %cmp114, label %land.lhs.true116, label %land.end122
land.lhs.true116: ; preds = %land.end108
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%128 = load double, double* @xd, align 8
%129 = load double, double* @lxd, align 8
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%cmp117 = fcmp oeq double %128, %129
br i1 %cmp117, label %land.rhs119, label %land.end122
land.rhs119: ; preds = %land.lhs.true116
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%130 = load double, double* @yd, align 8
%131 = load double, double* @lyd, align 8
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%cmp120 = fcmp oeq double %130, %131
br label %land.end122
land.end122: ; preds = %land.rhs119, %land.lhs.true116, %land.end108
%132 = phi i1 [ false, %land.lhs.true116 ], [ false, %land.end108 ], [ %cmp120, %land.rhs119 ]
%land.ext123 = zext i1 %132 to i32
[opaque pointer type] Add textual IR support for explicit type parameter to the call instruction See r230786 and r230794 for similar changes to gep and load respectively. Call is a bit different because it often doesn't have a single explicit type - usually the type is deduced from the arguments, and just the return type is explicit. In those cases there's no need to change the IR. When that's not the case, the IR usually contains the pointer type of the first operand - but since typed pointers are going away, that representation is insufficient so I'm just stripping the "pointerness" of the explicit type away. This does make the IR a bit weird - it /sort of/ reads like the type of the first operand: "call void () %x(" but %x is actually of type "void ()*" and will eventually be just of type "ptr". But this seems not too bad and I don't think it would benefit from repeating the type ("void (), void () * %x(" and then eventually "void (), ptr %x(") as has been done with gep and load. This also has a side benefit: since the explicit type is no longer a pointer, there's no ambiguity between an explicit type and a function that returns a function pointer. Previously this case needed an explicit type (eg: a function returning a void() function was written as "call void () () * @x(" rather than "call void () * @x(" because of the ambiguity between a function returning a pointer to a void() function and a function returning void). No ambiguity means even function pointer return types can just be written alone, without writing the whole function's type. This leaves /only/ the varargs case where the explicit type is required. Given the special type syntax in call instructions, the regex-fu used for migration was a bit more involved in its own unique way (as every one of these is) so here it is. Use it in conjunction with the apply.sh script and associated find/xargs commands I've provided in rr230786 to migrate your out of tree tests. Do let me know if any of this doesn't cover your cases & we can iterate on a more general script/regexes to help others with out of tree tests. About 9 test cases couldn't be automatically migrated - half of those were functions returning function pointers, where I just had to manually delete the function argument types now that we didn't need an explicit function type there. The other half were typedefs of function types used in calls - just had to manually drop the * from those. import fileinput import sys import re pat = re.compile(r'((?:=|:|^|\s)call\s(?:[^@]*?))(\s*$|\s*(?:(?:\[\[[a-zA-Z0-9_]+\]\]|[@%](?:(")?[\\\?@a-zA-Z0-9_.]*?(?(3)"|)|{{.*}}))(?:\(|$)|undef|inttoptr|bitcast|null|asm).*$)') addrspace_end = re.compile(r"addrspace\(\d+\)\s*\*$") func_end = re.compile("(?:void.*|\)\s*)\*$") def conv(match, line): if not match or re.search(addrspace_end, match.group(1)) or not re.search(func_end, match.group(1)): return line return line[:match.start()] + match.group(1)[:match.group(1).rfind('*')].rstrip() + match.group(2) + line[match.end():] for line in sys.stdin: sys.stdout.write(conv(re.search(pat, line), line)) llvm-svn: 235145
2015-04-17 07:24:18 +08:00
%call124 = call i32 (i8*, ...) @printf(i8* getelementptr inbounds ([22 x i8], [22 x i8]* @.str2, i32 0, i32 0), double %conv112, double %conv113, double %122, double %123, double %124, double %125, i32 %land.ext123)
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
call void @clear()
store double 1.561234e+01, double* @ret_df, align 8
%call125 = call double @df_v()
store double %call125, double* @lret_df, align 8
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%133 = load double, double* @ret_df, align 8
%134 = load double, double* @lret_df, align 8
%135 = load double, double* @ret_df, align 8
%136 = load double, double* @lret_df, align 8
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%cmp126 = fcmp oeq double %135, %136
%conv127 = zext i1 %cmp126 to i32
[opaque pointer type] Add textual IR support for explicit type parameter to the call instruction See r230786 and r230794 for similar changes to gep and load respectively. Call is a bit different because it often doesn't have a single explicit type - usually the type is deduced from the arguments, and just the return type is explicit. In those cases there's no need to change the IR. When that's not the case, the IR usually contains the pointer type of the first operand - but since typed pointers are going away, that representation is insufficient so I'm just stripping the "pointerness" of the explicit type away. This does make the IR a bit weird - it /sort of/ reads like the type of the first operand: "call void () %x(" but %x is actually of type "void ()*" and will eventually be just of type "ptr". But this seems not too bad and I don't think it would benefit from repeating the type ("void (), void () * %x(" and then eventually "void (), ptr %x(") as has been done with gep and load. This also has a side benefit: since the explicit type is no longer a pointer, there's no ambiguity between an explicit type and a function that returns a function pointer. Previously this case needed an explicit type (eg: a function returning a void() function was written as "call void () () * @x(" rather than "call void () * @x(" because of the ambiguity between a function returning a pointer to a void() function and a function returning void). No ambiguity means even function pointer return types can just be written alone, without writing the whole function's type. This leaves /only/ the varargs case where the explicit type is required. Given the special type syntax in call instructions, the regex-fu used for migration was a bit more involved in its own unique way (as every one of these is) so here it is. Use it in conjunction with the apply.sh script and associated find/xargs commands I've provided in rr230786 to migrate your out of tree tests. Do let me know if any of this doesn't cover your cases & we can iterate on a more general script/regexes to help others with out of tree tests. About 9 test cases couldn't be automatically migrated - half of those were functions returning function pointers, where I just had to manually delete the function argument types now that we didn't need an explicit function type there. The other half were typedefs of function types used in calls - just had to manually drop the * from those. import fileinput import sys import re pat = re.compile(r'((?:=|:|^|\s)call\s(?:[^@]*?))(\s*$|\s*(?:(?:\[\[[a-zA-Z0-9_]+\]\]|[@%](?:(")?[\\\?@a-zA-Z0-9_.]*?(?(3)"|)|{{.*}}))(?:\(|$)|undef|inttoptr|bitcast|null|asm).*$)') addrspace_end = re.compile(r"addrspace\(\d+\)\s*\*$") func_end = re.compile("(?:void.*|\)\s*)\*$") def conv(match, line): if not match or re.search(addrspace_end, match.group(1)) or not re.search(func_end, match.group(1)): return line return line[:match.start()] + match.group(1)[:match.group(1).rfind('*')].rstrip() + match.group(2) + line[match.end():] for line in sys.stdin: sys.stdout.write(conv(re.search(pat, line), line)) llvm-svn: 235145
2015-04-17 07:24:18 +08:00
%call128 = call i32 (i8*, ...) @printf(i8* getelementptr inbounds ([10 x i8], [10 x i8]* @.str, i32 0, i32 0), double %133, double %134, i32 %conv127)
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
call void @clear()
store double 1.345873e+01, double* @ret_df, align 8
store float 3.434520e+05, float* @lx, align 4
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%137 = load float, float* @lx, align 4
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%call129 = call double @df_sf(float %137)
store double %call129, double* @lret_df, align 8
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%138 = load double, double* @ret_df, align 8
%139 = load double, double* @lret_df, align 8
%140 = load float, float* @x, align 4
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%conv130 = fpext float %140 to double
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%141 = load float, float* @lx, align 4
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%conv131 = fpext float %141 to double
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%142 = load double, double* @ret_df, align 8
%143 = load double, double* @lret_df, align 8
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%cmp132 = fcmp oeq double %142, %143
%conv133 = zext i1 %cmp132 to i32
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%144 = load float, float* @x, align 4
%145 = load float, float* @lx, align 4
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%cmp134 = fcmp oeq float %144, %145
%conv135 = zext i1 %cmp134 to i32
%and136 = and i32 %conv133, %conv135
[opaque pointer type] Add textual IR support for explicit type parameter to the call instruction See r230786 and r230794 for similar changes to gep and load respectively. Call is a bit different because it often doesn't have a single explicit type - usually the type is deduced from the arguments, and just the return type is explicit. In those cases there's no need to change the IR. When that's not the case, the IR usually contains the pointer type of the first operand - but since typed pointers are going away, that representation is insufficient so I'm just stripping the "pointerness" of the explicit type away. This does make the IR a bit weird - it /sort of/ reads like the type of the first operand: "call void () %x(" but %x is actually of type "void ()*" and will eventually be just of type "ptr". But this seems not too bad and I don't think it would benefit from repeating the type ("void (), void () * %x(" and then eventually "void (), ptr %x(") as has been done with gep and load. This also has a side benefit: since the explicit type is no longer a pointer, there's no ambiguity between an explicit type and a function that returns a function pointer. Previously this case needed an explicit type (eg: a function returning a void() function was written as "call void () () * @x(" rather than "call void () * @x(" because of the ambiguity between a function returning a pointer to a void() function and a function returning void). No ambiguity means even function pointer return types can just be written alone, without writing the whole function's type. This leaves /only/ the varargs case where the explicit type is required. Given the special type syntax in call instructions, the regex-fu used for migration was a bit more involved in its own unique way (as every one of these is) so here it is. Use it in conjunction with the apply.sh script and associated find/xargs commands I've provided in rr230786 to migrate your out of tree tests. Do let me know if any of this doesn't cover your cases & we can iterate on a more general script/regexes to help others with out of tree tests. About 9 test cases couldn't be automatically migrated - half of those were functions returning function pointers, where I just had to manually delete the function argument types now that we didn't need an explicit function type there. The other half were typedefs of function types used in calls - just had to manually drop the * from those. import fileinput import sys import re pat = re.compile(r'((?:=|:|^|\s)call\s(?:[^@]*?))(\s*$|\s*(?:(?:\[\[[a-zA-Z0-9_]+\]\]|[@%](?:(")?[\\\?@a-zA-Z0-9_.]*?(?(3)"|)|{{.*}}))(?:\(|$)|undef|inttoptr|bitcast|null|asm).*$)') addrspace_end = re.compile(r"addrspace\(\d+\)\s*\*$") func_end = re.compile("(?:void.*|\)\s*)\*$") def conv(match, line): if not match or re.search(addrspace_end, match.group(1)) or not re.search(func_end, match.group(1)): return line return line[:match.start()] + match.group(1)[:match.group(1).rfind('*')].rstrip() + match.group(2) + line[match.end():] for line in sys.stdin: sys.stdout.write(conv(re.search(pat, line), line)) llvm-svn: 235145
2015-04-17 07:24:18 +08:00
%call137 = call i32 (i8*, ...) @printf(i8* getelementptr inbounds ([16 x i8], [16 x i8]* @.str1, i32 0, i32 0), double %138, double %139, double %conv130, double %conv131, i32 %and136)
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
call void @clear()
store double 0x4084F3AB7AA25D8D, double* @ret_df, align 8
store double 0x4114F671D2F1A9FC, double* @lxd, align 8
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%146 = load double, double* @lxd, align 8
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%call138 = call double @df_df(double %146)
store double %call138, double* @lret_df, align 8
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%147 = load double, double* @ret_df, align 8
%148 = load double, double* @lret_df, align 8
%149 = load double, double* @xd, align 8
%150 = load double, double* @lxd, align 8
%151 = load double, double* @ret_df, align 8
%152 = load double, double* @lret_df, align 8
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%cmp139 = fcmp oeq double %151, %152
%conv140 = zext i1 %cmp139 to i32
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%153 = load double, double* @xd, align 8
%154 = load double, double* @lxd, align 8
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%cmp141 = fcmp oeq double %153, %154
%conv142 = zext i1 %cmp141 to i32
%and143 = and i32 %conv140, %conv142
[opaque pointer type] Add textual IR support for explicit type parameter to the call instruction See r230786 and r230794 for similar changes to gep and load respectively. Call is a bit different because it often doesn't have a single explicit type - usually the type is deduced from the arguments, and just the return type is explicit. In those cases there's no need to change the IR. When that's not the case, the IR usually contains the pointer type of the first operand - but since typed pointers are going away, that representation is insufficient so I'm just stripping the "pointerness" of the explicit type away. This does make the IR a bit weird - it /sort of/ reads like the type of the first operand: "call void () %x(" but %x is actually of type "void ()*" and will eventually be just of type "ptr". But this seems not too bad and I don't think it would benefit from repeating the type ("void (), void () * %x(" and then eventually "void (), ptr %x(") as has been done with gep and load. This also has a side benefit: since the explicit type is no longer a pointer, there's no ambiguity between an explicit type and a function that returns a function pointer. Previously this case needed an explicit type (eg: a function returning a void() function was written as "call void () () * @x(" rather than "call void () * @x(" because of the ambiguity between a function returning a pointer to a void() function and a function returning void). No ambiguity means even function pointer return types can just be written alone, without writing the whole function's type. This leaves /only/ the varargs case where the explicit type is required. Given the special type syntax in call instructions, the regex-fu used for migration was a bit more involved in its own unique way (as every one of these is) so here it is. Use it in conjunction with the apply.sh script and associated find/xargs commands I've provided in rr230786 to migrate your out of tree tests. Do let me know if any of this doesn't cover your cases & we can iterate on a more general script/regexes to help others with out of tree tests. About 9 test cases couldn't be automatically migrated - half of those were functions returning function pointers, where I just had to manually delete the function argument types now that we didn't need an explicit function type there. The other half were typedefs of function types used in calls - just had to manually drop the * from those. import fileinput import sys import re pat = re.compile(r'((?:=|:|^|\s)call\s(?:[^@]*?))(\s*$|\s*(?:(?:\[\[[a-zA-Z0-9_]+\]\]|[@%](?:(")?[\\\?@a-zA-Z0-9_.]*?(?(3)"|)|{{.*}}))(?:\(|$)|undef|inttoptr|bitcast|null|asm).*$)') addrspace_end = re.compile(r"addrspace\(\d+\)\s*\*$") func_end = re.compile("(?:void.*|\)\s*)\*$") def conv(match, line): if not match or re.search(addrspace_end, match.group(1)) or not re.search(func_end, match.group(1)): return line return line[:match.start()] + match.group(1)[:match.group(1).rfind('*')].rstrip() + match.group(2) + line[match.end():] for line in sys.stdin: sys.stdout.write(conv(re.search(pat, line), line)) llvm-svn: 235145
2015-04-17 07:24:18 +08:00
%call144 = call i32 (i8*, ...) @printf(i8* getelementptr inbounds ([16 x i8], [16 x i8]* @.str1, i32 0, i32 0), double %147, double %148, double %149, double %150, i32 %and143)
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
call void @clear()
store double 6.781956e+03, double* @ret_df, align 8
store float 4.445500e+03, float* @lx, align 4
store float 0x4068ACCCC0000000, float* @ly, align 4
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%155 = load float, float* @lx, align 4
%156 = load float, float* @ly, align 4
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%call145 = call double @df_sf_sf(float %155, float %156)
store double %call145, double* @lret_df, align 8
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%157 = load double, double* @ret_df, align 8
%158 = load double, double* @lret_df, align 8
%159 = load float, float* @x, align 4
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%conv146 = fpext float %159 to double
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%160 = load float, float* @lx, align 4
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%conv147 = fpext float %160 to double
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%161 = load float, float* @y, align 4
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%conv148 = fpext float %161 to double
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%162 = load float, float* @ly, align 4
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%conv149 = fpext float %162 to double
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%163 = load double, double* @ret_df, align 8
%164 = load double, double* @lret_df, align 8
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%cmp150 = fcmp oeq double %163, %164
br i1 %cmp150, label %land.lhs.true152, label %land.end158
land.lhs.true152: ; preds = %land.end122
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%165 = load float, float* @x, align 4
%166 = load float, float* @lx, align 4
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%cmp153 = fcmp oeq float %165, %166
br i1 %cmp153, label %land.rhs155, label %land.end158
land.rhs155: ; preds = %land.lhs.true152
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%167 = load float, float* @y, align 4
%168 = load float, float* @ly, align 4
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%cmp156 = fcmp oeq float %167, %168
br label %land.end158
land.end158: ; preds = %land.rhs155, %land.lhs.true152, %land.end122
%169 = phi i1 [ false, %land.lhs.true152 ], [ false, %land.end122 ], [ %cmp156, %land.rhs155 ]
%land.ext159 = zext i1 %169 to i32
[opaque pointer type] Add textual IR support for explicit type parameter to the call instruction See r230786 and r230794 for similar changes to gep and load respectively. Call is a bit different because it often doesn't have a single explicit type - usually the type is deduced from the arguments, and just the return type is explicit. In those cases there's no need to change the IR. When that's not the case, the IR usually contains the pointer type of the first operand - but since typed pointers are going away, that representation is insufficient so I'm just stripping the "pointerness" of the explicit type away. This does make the IR a bit weird - it /sort of/ reads like the type of the first operand: "call void () %x(" but %x is actually of type "void ()*" and will eventually be just of type "ptr". But this seems not too bad and I don't think it would benefit from repeating the type ("void (), void () * %x(" and then eventually "void (), ptr %x(") as has been done with gep and load. This also has a side benefit: since the explicit type is no longer a pointer, there's no ambiguity between an explicit type and a function that returns a function pointer. Previously this case needed an explicit type (eg: a function returning a void() function was written as "call void () () * @x(" rather than "call void () * @x(" because of the ambiguity between a function returning a pointer to a void() function and a function returning void). No ambiguity means even function pointer return types can just be written alone, without writing the whole function's type. This leaves /only/ the varargs case where the explicit type is required. Given the special type syntax in call instructions, the regex-fu used for migration was a bit more involved in its own unique way (as every one of these is) so here it is. Use it in conjunction with the apply.sh script and associated find/xargs commands I've provided in rr230786 to migrate your out of tree tests. Do let me know if any of this doesn't cover your cases & we can iterate on a more general script/regexes to help others with out of tree tests. About 9 test cases couldn't be automatically migrated - half of those were functions returning function pointers, where I just had to manually delete the function argument types now that we didn't need an explicit function type there. The other half were typedefs of function types used in calls - just had to manually drop the * from those. import fileinput import sys import re pat = re.compile(r'((?:=|:|^|\s)call\s(?:[^@]*?))(\s*$|\s*(?:(?:\[\[[a-zA-Z0-9_]+\]\]|[@%](?:(")?[\\\?@a-zA-Z0-9_.]*?(?(3)"|)|{{.*}}))(?:\(|$)|undef|inttoptr|bitcast|null|asm).*$)') addrspace_end = re.compile(r"addrspace\(\d+\)\s*\*$") func_end = re.compile("(?:void.*|\)\s*)\*$") def conv(match, line): if not match or re.search(addrspace_end, match.group(1)) or not re.search(func_end, match.group(1)): return line return line[:match.start()] + match.group(1)[:match.group(1).rfind('*')].rstrip() + match.group(2) + line[match.end():] for line in sys.stdin: sys.stdout.write(conv(re.search(pat, line), line)) llvm-svn: 235145
2015-04-17 07:24:18 +08:00
%call160 = call i32 (i8*, ...) @printf(i8* getelementptr inbounds ([22 x i8], [22 x i8]* @.str2, i32 0, i32 0), double %157, double %158, double %conv146, double %conv147, double %conv148, double %conv149, i32 %land.ext159)
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
call void @clear()
store double 1.889130e+05, double* @ret_df, align 8
store float 9.111450e+05, float* @lx, align 4
store double 0x4185320A58000000, double* @lyd, align 8
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%170 = load float, float* @lx, align 4
%171 = load double, double* @lyd, align 8
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%call161 = call double @df_sf_df(float %170, double %171)
store double %call161, double* @lret_df, align 8
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%172 = load double, double* @ret_df, align 8
%173 = load double, double* @lret_df, align 8
%174 = load float, float* @x, align 4
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%conv162 = fpext float %174 to double
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%175 = load float, float* @lx, align 4
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%conv163 = fpext float %175 to double
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%176 = load double, double* @yd, align 8
%177 = load double, double* @lyd, align 8
%178 = load double, double* @ret_df, align 8
%179 = load double, double* @lret_df, align 8
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%cmp164 = fcmp oeq double %178, %179
br i1 %cmp164, label %land.lhs.true166, label %land.end172
land.lhs.true166: ; preds = %land.end158
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%180 = load float, float* @x, align 4
%181 = load float, float* @lx, align 4
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%cmp167 = fcmp oeq float %180, %181
br i1 %cmp167, label %land.rhs169, label %land.end172
land.rhs169: ; preds = %land.lhs.true166
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%182 = load double, double* @yd, align 8
%183 = load double, double* @lyd, align 8
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%cmp170 = fcmp oeq double %182, %183
br label %land.end172
land.end172: ; preds = %land.rhs169, %land.lhs.true166, %land.end158
%184 = phi i1 [ false, %land.lhs.true166 ], [ false, %land.end158 ], [ %cmp170, %land.rhs169 ]
%land.ext173 = zext i1 %184 to i32
[opaque pointer type] Add textual IR support for explicit type parameter to the call instruction See r230786 and r230794 for similar changes to gep and load respectively. Call is a bit different because it often doesn't have a single explicit type - usually the type is deduced from the arguments, and just the return type is explicit. In those cases there's no need to change the IR. When that's not the case, the IR usually contains the pointer type of the first operand - but since typed pointers are going away, that representation is insufficient so I'm just stripping the "pointerness" of the explicit type away. This does make the IR a bit weird - it /sort of/ reads like the type of the first operand: "call void () %x(" but %x is actually of type "void ()*" and will eventually be just of type "ptr". But this seems not too bad and I don't think it would benefit from repeating the type ("void (), void () * %x(" and then eventually "void (), ptr %x(") as has been done with gep and load. This also has a side benefit: since the explicit type is no longer a pointer, there's no ambiguity between an explicit type and a function that returns a function pointer. Previously this case needed an explicit type (eg: a function returning a void() function was written as "call void () () * @x(" rather than "call void () * @x(" because of the ambiguity between a function returning a pointer to a void() function and a function returning void). No ambiguity means even function pointer return types can just be written alone, without writing the whole function's type. This leaves /only/ the varargs case where the explicit type is required. Given the special type syntax in call instructions, the regex-fu used for migration was a bit more involved in its own unique way (as every one of these is) so here it is. Use it in conjunction with the apply.sh script and associated find/xargs commands I've provided in rr230786 to migrate your out of tree tests. Do let me know if any of this doesn't cover your cases & we can iterate on a more general script/regexes to help others with out of tree tests. About 9 test cases couldn't be automatically migrated - half of those were functions returning function pointers, where I just had to manually delete the function argument types now that we didn't need an explicit function type there. The other half were typedefs of function types used in calls - just had to manually drop the * from those. import fileinput import sys import re pat = re.compile(r'((?:=|:|^|\s)call\s(?:[^@]*?))(\s*$|\s*(?:(?:\[\[[a-zA-Z0-9_]+\]\]|[@%](?:(")?[\\\?@a-zA-Z0-9_.]*?(?(3)"|)|{{.*}}))(?:\(|$)|undef|inttoptr|bitcast|null|asm).*$)') addrspace_end = re.compile(r"addrspace\(\d+\)\s*\*$") func_end = re.compile("(?:void.*|\)\s*)\*$") def conv(match, line): if not match or re.search(addrspace_end, match.group(1)) or not re.search(func_end, match.group(1)): return line return line[:match.start()] + match.group(1)[:match.group(1).rfind('*')].rstrip() + match.group(2) + line[match.end():] for line in sys.stdin: sys.stdout.write(conv(re.search(pat, line), line)) llvm-svn: 235145
2015-04-17 07:24:18 +08:00
%call174 = call i32 (i8*, ...) @printf(i8* getelementptr inbounds ([22 x i8], [22 x i8]* @.str2, i32 0, i32 0), double %172, double %173, double %conv162, double %conv163, double %176, double %177, i32 %land.ext173)
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
call void @clear()
store double 0x418B2DB900000000, double* @ret_df, align 8
store double 0x41B1EF2ED3000000, double* @lxd, align 8
store float 1.244562e+06, float* @ly, align 4
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%185 = load double, double* @lxd, align 8
%186 = load float, float* @ly, align 4
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%call175 = call double @df_df_sf(double %185, float %186)
store double %call175, double* @lret_df, align 8
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%187 = load double, double* @ret_df, align 8
%188 = load double, double* @lret_df, align 8
%189 = load double, double* @xd, align 8
%190 = load double, double* @lxd, align 8
%191 = load float, float* @y, align 4
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%conv176 = fpext float %191 to double
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%192 = load float, float* @ly, align 4
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%conv177 = fpext float %192 to double
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%193 = load double, double* @ret_df, align 8
%194 = load double, double* @lret_df, align 8
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%cmp178 = fcmp oeq double %193, %194
br i1 %cmp178, label %land.lhs.true180, label %land.end186
land.lhs.true180: ; preds = %land.end172
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%195 = load double, double* @xd, align 8
%196 = load double, double* @lxd, align 8
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%cmp181 = fcmp oeq double %195, %196
br i1 %cmp181, label %land.rhs183, label %land.end186
land.rhs183: ; preds = %land.lhs.true180
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%197 = load float, float* @y, align 4
%198 = load float, float* @ly, align 4
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%cmp184 = fcmp oeq float %197, %198
br label %land.end186
land.end186: ; preds = %land.rhs183, %land.lhs.true180, %land.end172
%199 = phi i1 [ false, %land.lhs.true180 ], [ false, %land.end172 ], [ %cmp184, %land.rhs183 ]
%land.ext187 = zext i1 %199 to i32
[opaque pointer type] Add textual IR support for explicit type parameter to the call instruction See r230786 and r230794 for similar changes to gep and load respectively. Call is a bit different because it often doesn't have a single explicit type - usually the type is deduced from the arguments, and just the return type is explicit. In those cases there's no need to change the IR. When that's not the case, the IR usually contains the pointer type of the first operand - but since typed pointers are going away, that representation is insufficient so I'm just stripping the "pointerness" of the explicit type away. This does make the IR a bit weird - it /sort of/ reads like the type of the first operand: "call void () %x(" but %x is actually of type "void ()*" and will eventually be just of type "ptr". But this seems not too bad and I don't think it would benefit from repeating the type ("void (), void () * %x(" and then eventually "void (), ptr %x(") as has been done with gep and load. This also has a side benefit: since the explicit type is no longer a pointer, there's no ambiguity between an explicit type and a function that returns a function pointer. Previously this case needed an explicit type (eg: a function returning a void() function was written as "call void () () * @x(" rather than "call void () * @x(" because of the ambiguity between a function returning a pointer to a void() function and a function returning void). No ambiguity means even function pointer return types can just be written alone, without writing the whole function's type. This leaves /only/ the varargs case where the explicit type is required. Given the special type syntax in call instructions, the regex-fu used for migration was a bit more involved in its own unique way (as every one of these is) so here it is. Use it in conjunction with the apply.sh script and associated find/xargs commands I've provided in rr230786 to migrate your out of tree tests. Do let me know if any of this doesn't cover your cases & we can iterate on a more general script/regexes to help others with out of tree tests. About 9 test cases couldn't be automatically migrated - half of those were functions returning function pointers, where I just had to manually delete the function argument types now that we didn't need an explicit function type there. The other half were typedefs of function types used in calls - just had to manually drop the * from those. import fileinput import sys import re pat = re.compile(r'((?:=|:|^|\s)call\s(?:[^@]*?))(\s*$|\s*(?:(?:\[\[[a-zA-Z0-9_]+\]\]|[@%](?:(")?[\\\?@a-zA-Z0-9_.]*?(?(3)"|)|{{.*}}))(?:\(|$)|undef|inttoptr|bitcast|null|asm).*$)') addrspace_end = re.compile(r"addrspace\(\d+\)\s*\*$") func_end = re.compile("(?:void.*|\)\s*)\*$") def conv(match, line): if not match or re.search(addrspace_end, match.group(1)) or not re.search(func_end, match.group(1)): return line return line[:match.start()] + match.group(1)[:match.group(1).rfind('*')].rstrip() + match.group(2) + line[match.end():] for line in sys.stdin: sys.stdout.write(conv(re.search(pat, line), line)) llvm-svn: 235145
2015-04-17 07:24:18 +08:00
%call188 = call i32 (i8*, ...) @printf(i8* getelementptr inbounds ([22 x i8], [22 x i8]* @.str2, i32 0, i32 0), double %187, double %188, double %189, double %190, double %conv176, double %conv177, i32 %land.ext187)
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
call void @clear()
store double 3.987721e+06, double* @ret_df, align 8
store double 5.223560e+00, double* @lxd, align 8
store double 0x40B7D37CC1A8AC5C, double* @lyd, align 8
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%200 = load double, double* @lxd, align 8
%201 = load double, double* @lyd, align 8
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%call189 = call double @df_df_df(double %200, double %201)
store double %call189, double* @lret_df, align 8
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%202 = load double, double* @ret_df, align 8
%203 = load double, double* @lret_df, align 8
%204 = load double, double* @xd, align 8
%205 = load double, double* @lxd, align 8
%206 = load double, double* @yd, align 8
%207 = load double, double* @lyd, align 8
%208 = load double, double* @ret_df, align 8
%209 = load double, double* @lret_df, align 8
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%cmp190 = fcmp oeq double %208, %209
br i1 %cmp190, label %land.lhs.true192, label %land.end198
land.lhs.true192: ; preds = %land.end186
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%210 = load double, double* @xd, align 8
%211 = load double, double* @lxd, align 8
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%cmp193 = fcmp oeq double %210, %211
br i1 %cmp193, label %land.rhs195, label %land.end198
land.rhs195: ; preds = %land.lhs.true192
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%212 = load double, double* @yd, align 8
%213 = load double, double* @lyd, align 8
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%cmp196 = fcmp oeq double %212, %213
br label %land.end198
land.end198: ; preds = %land.rhs195, %land.lhs.true192, %land.end186
%214 = phi i1 [ false, %land.lhs.true192 ], [ false, %land.end186 ], [ %cmp196, %land.rhs195 ]
%land.ext199 = zext i1 %214 to i32
[opaque pointer type] Add textual IR support for explicit type parameter to the call instruction See r230786 and r230794 for similar changes to gep and load respectively. Call is a bit different because it often doesn't have a single explicit type - usually the type is deduced from the arguments, and just the return type is explicit. In those cases there's no need to change the IR. When that's not the case, the IR usually contains the pointer type of the first operand - but since typed pointers are going away, that representation is insufficient so I'm just stripping the "pointerness" of the explicit type away. This does make the IR a bit weird - it /sort of/ reads like the type of the first operand: "call void () %x(" but %x is actually of type "void ()*" and will eventually be just of type "ptr". But this seems not too bad and I don't think it would benefit from repeating the type ("void (), void () * %x(" and then eventually "void (), ptr %x(") as has been done with gep and load. This also has a side benefit: since the explicit type is no longer a pointer, there's no ambiguity between an explicit type and a function that returns a function pointer. Previously this case needed an explicit type (eg: a function returning a void() function was written as "call void () () * @x(" rather than "call void () * @x(" because of the ambiguity between a function returning a pointer to a void() function and a function returning void). No ambiguity means even function pointer return types can just be written alone, without writing the whole function's type. This leaves /only/ the varargs case where the explicit type is required. Given the special type syntax in call instructions, the regex-fu used for migration was a bit more involved in its own unique way (as every one of these is) so here it is. Use it in conjunction with the apply.sh script and associated find/xargs commands I've provided in rr230786 to migrate your out of tree tests. Do let me know if any of this doesn't cover your cases & we can iterate on a more general script/regexes to help others with out of tree tests. About 9 test cases couldn't be automatically migrated - half of those were functions returning function pointers, where I just had to manually delete the function argument types now that we didn't need an explicit function type there. The other half were typedefs of function types used in calls - just had to manually drop the * from those. import fileinput import sys import re pat = re.compile(r'((?:=|:|^|\s)call\s(?:[^@]*?))(\s*$|\s*(?:(?:\[\[[a-zA-Z0-9_]+\]\]|[@%](?:(")?[\\\?@a-zA-Z0-9_.]*?(?(3)"|)|{{.*}}))(?:\(|$)|undef|inttoptr|bitcast|null|asm).*$)') addrspace_end = re.compile(r"addrspace\(\d+\)\s*\*$") func_end = re.compile("(?:void.*|\)\s*)\*$") def conv(match, line): if not match or re.search(addrspace_end, match.group(1)) or not re.search(func_end, match.group(1)): return line return line[:match.start()] + match.group(1)[:match.group(1).rfind('*')].rstrip() + match.group(2) + line[match.end():] for line in sys.stdin: sys.stdout.write(conv(re.search(pat, line), line)) llvm-svn: 235145
2015-04-17 07:24:18 +08:00
%call200 = call i32 (i8*, ...) @printf(i8* getelementptr inbounds ([22 x i8], [22 x i8]* @.str2, i32 0, i32 0), double %202, double %203, double %204, double %205, double %206, double %207, i32 %land.ext199)
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
call void @clear()
[opaque pointer type] Add textual IR support for explicit type parameter to gep operator Similar to gep (r230786) and load (r230794) changes. Similar migration script can be used to update test cases, which successfully migrated all of LLVM and Polly, but about 4 test cases needed manually changes in Clang. (this script will read the contents of stdin and massage it into stdout - wrap it in the 'apply.sh' script shown in previous commits + xargs to apply it over a large set of test cases) import fileinput import sys import re rep = re.compile(r"(getelementptr(?:\s+inbounds)?\s*\()((<\d*\s+x\s+)?([^@]*?)(|\s*addrspace\(\d+\))\s*\*(?(3)>)\s*)(?=$|%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|zeroinitializer|<|\[\[[a-zA-Z]|\{\{)", re.MULTILINE | re.DOTALL) def conv(match): line = match.group(1) line += match.group(4) line += ", " line += match.group(2) return line line = sys.stdin.read() off = 0 for match in re.finditer(rep, line): sys.stdout.write(line[off:match.start()]) sys.stdout.write(conv(match)) off = match.end() sys.stdout.write(line[off:]) llvm-svn: 232184
2015-03-14 02:20:45 +08:00
store float 4.500000e+00, float* getelementptr inbounds ({ float, float }, { float, float }* @ret_sc, i32 0, i32 0)
store float 7.000000e+00, float* getelementptr inbounds ({ float, float }, { float, float }* @ret_sc, i32 0, i32 1)
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%call201 = call { float, float } @sc_v()
%215 = extractvalue { float, float } %call201, 0
%216 = extractvalue { float, float } %call201, 1
[opaque pointer type] Add textual IR support for explicit type parameter to gep operator Similar to gep (r230786) and load (r230794) changes. Similar migration script can be used to update test cases, which successfully migrated all of LLVM and Polly, but about 4 test cases needed manually changes in Clang. (this script will read the contents of stdin and massage it into stdout - wrap it in the 'apply.sh' script shown in previous commits + xargs to apply it over a large set of test cases) import fileinput import sys import re rep = re.compile(r"(getelementptr(?:\s+inbounds)?\s*\()((<\d*\s+x\s+)?([^@]*?)(|\s*addrspace\(\d+\))\s*\*(?(3)>)\s*)(?=$|%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|zeroinitializer|<|\[\[[a-zA-Z]|\{\{)", re.MULTILINE | re.DOTALL) def conv(match): line = match.group(1) line += match.group(4) line += ", " line += match.group(2) return line line = sys.stdin.read() off = 0 for match in re.finditer(rep, line): sys.stdout.write(line[off:match.start()]) sys.stdout.write(conv(match)) off = match.end() sys.stdout.write(line[off:]) llvm-svn: 232184
2015-03-14 02:20:45 +08:00
store float %215, float* getelementptr inbounds ({ float, float }, { float, float }* @lret_sc, i32 0, i32 0)
store float %216, float* getelementptr inbounds ({ float, float }, { float, float }* @lret_sc, i32 0, i32 1)
%ret_sc.real = load float, float* getelementptr inbounds ({ float, float }, { float, float }* @ret_sc, i32 0, i32 0)
%ret_sc.imag = load float, float* getelementptr inbounds ({ float, float }, { float, float }* @ret_sc, i32 0, i32 1)
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%conv202 = fpext float %ret_sc.real to double
%conv203 = fpext float %ret_sc.imag to double
[opaque pointer type] Add textual IR support for explicit type parameter to gep operator Similar to gep (r230786) and load (r230794) changes. Similar migration script can be used to update test cases, which successfully migrated all of LLVM and Polly, but about 4 test cases needed manually changes in Clang. (this script will read the contents of stdin and massage it into stdout - wrap it in the 'apply.sh' script shown in previous commits + xargs to apply it over a large set of test cases) import fileinput import sys import re rep = re.compile(r"(getelementptr(?:\s+inbounds)?\s*\()((<\d*\s+x\s+)?([^@]*?)(|\s*addrspace\(\d+\))\s*\*(?(3)>)\s*)(?=$|%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|zeroinitializer|<|\[\[[a-zA-Z]|\{\{)", re.MULTILINE | re.DOTALL) def conv(match): line = match.group(1) line += match.group(4) line += ", " line += match.group(2) return line line = sys.stdin.read() off = 0 for match in re.finditer(rep, line): sys.stdout.write(line[off:match.start()]) sys.stdout.write(conv(match)) off = match.end() sys.stdout.write(line[off:]) llvm-svn: 232184
2015-03-14 02:20:45 +08:00
%ret_sc.real204 = load float, float* getelementptr inbounds ({ float, float }, { float, float }* @ret_sc, i32 0, i32 0)
%ret_sc.imag205 = load float, float* getelementptr inbounds ({ float, float }, { float, float }* @ret_sc, i32 0, i32 1)
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%conv206 = fpext float %ret_sc.real204 to double
%conv207 = fpext float %ret_sc.imag205 to double
[opaque pointer type] Add textual IR support for explicit type parameter to gep operator Similar to gep (r230786) and load (r230794) changes. Similar migration script can be used to update test cases, which successfully migrated all of LLVM and Polly, but about 4 test cases needed manually changes in Clang. (this script will read the contents of stdin and massage it into stdout - wrap it in the 'apply.sh' script shown in previous commits + xargs to apply it over a large set of test cases) import fileinput import sys import re rep = re.compile(r"(getelementptr(?:\s+inbounds)?\s*\()((<\d*\s+x\s+)?([^@]*?)(|\s*addrspace\(\d+\))\s*\*(?(3)>)\s*)(?=$|%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|zeroinitializer|<|\[\[[a-zA-Z]|\{\{)", re.MULTILINE | re.DOTALL) def conv(match): line = match.group(1) line += match.group(4) line += ", " line += match.group(2) return line line = sys.stdin.read() off = 0 for match in re.finditer(rep, line): sys.stdout.write(line[off:match.start()]) sys.stdout.write(conv(match)) off = match.end() sys.stdout.write(line[off:]) llvm-svn: 232184
2015-03-14 02:20:45 +08:00
%lret_sc.real = load float, float* getelementptr inbounds ({ float, float }, { float, float }* @lret_sc, i32 0, i32 0)
%lret_sc.imag = load float, float* getelementptr inbounds ({ float, float }, { float, float }* @lret_sc, i32 0, i32 1)
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%conv208 = fpext float %lret_sc.real to double
%conv209 = fpext float %lret_sc.imag to double
[opaque pointer type] Add textual IR support for explicit type parameter to gep operator Similar to gep (r230786) and load (r230794) changes. Similar migration script can be used to update test cases, which successfully migrated all of LLVM and Polly, but about 4 test cases needed manually changes in Clang. (this script will read the contents of stdin and massage it into stdout - wrap it in the 'apply.sh' script shown in previous commits + xargs to apply it over a large set of test cases) import fileinput import sys import re rep = re.compile(r"(getelementptr(?:\s+inbounds)?\s*\()((<\d*\s+x\s+)?([^@]*?)(|\s*addrspace\(\d+\))\s*\*(?(3)>)\s*)(?=$|%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|zeroinitializer|<|\[\[[a-zA-Z]|\{\{)", re.MULTILINE | re.DOTALL) def conv(match): line = match.group(1) line += match.group(4) line += ", " line += match.group(2) return line line = sys.stdin.read() off = 0 for match in re.finditer(rep, line): sys.stdout.write(line[off:match.start()]) sys.stdout.write(conv(match)) off = match.end() sys.stdout.write(line[off:]) llvm-svn: 232184
2015-03-14 02:20:45 +08:00
%lret_sc.real210 = load float, float* getelementptr inbounds ({ float, float }, { float, float }* @lret_sc, i32 0, i32 0)
%lret_sc.imag211 = load float, float* getelementptr inbounds ({ float, float }, { float, float }* @lret_sc, i32 0, i32 1)
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%conv212 = fpext float %lret_sc.real210 to double
%conv213 = fpext float %lret_sc.imag211 to double
[opaque pointer type] Add textual IR support for explicit type parameter to gep operator Similar to gep (r230786) and load (r230794) changes. Similar migration script can be used to update test cases, which successfully migrated all of LLVM and Polly, but about 4 test cases needed manually changes in Clang. (this script will read the contents of stdin and massage it into stdout - wrap it in the 'apply.sh' script shown in previous commits + xargs to apply it over a large set of test cases) import fileinput import sys import re rep = re.compile(r"(getelementptr(?:\s+inbounds)?\s*\()((<\d*\s+x\s+)?([^@]*?)(|\s*addrspace\(\d+\))\s*\*(?(3)>)\s*)(?=$|%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|zeroinitializer|<|\[\[[a-zA-Z]|\{\{)", re.MULTILINE | re.DOTALL) def conv(match): line = match.group(1) line += match.group(4) line += ", " line += match.group(2) return line line = sys.stdin.read() off = 0 for match in re.finditer(rep, line): sys.stdout.write(line[off:match.start()]) sys.stdout.write(conv(match)) off = match.end() sys.stdout.write(line[off:]) llvm-svn: 232184
2015-03-14 02:20:45 +08:00
%ret_sc.real214 = load float, float* getelementptr inbounds ({ float, float }, { float, float }* @ret_sc, i32 0, i32 0)
%ret_sc.imag215 = load float, float* getelementptr inbounds ({ float, float }, { float, float }* @ret_sc, i32 0, i32 1)
%lret_sc.real216 = load float, float* getelementptr inbounds ({ float, float }, { float, float }* @lret_sc, i32 0, i32 0)
%lret_sc.imag217 = load float, float* getelementptr inbounds ({ float, float }, { float, float }* @lret_sc, i32 0, i32 1)
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%cmp.r = fcmp oeq float %ret_sc.real214, %lret_sc.real216
%cmp.i = fcmp oeq float %ret_sc.imag215, %lret_sc.imag217
%and.ri = and i1 %cmp.r, %cmp.i
%conv218 = zext i1 %and.ri to i32
[opaque pointer type] Add textual IR support for explicit type parameter to the call instruction See r230786 and r230794 for similar changes to gep and load respectively. Call is a bit different because it often doesn't have a single explicit type - usually the type is deduced from the arguments, and just the return type is explicit. In those cases there's no need to change the IR. When that's not the case, the IR usually contains the pointer type of the first operand - but since typed pointers are going away, that representation is insufficient so I'm just stripping the "pointerness" of the explicit type away. This does make the IR a bit weird - it /sort of/ reads like the type of the first operand: "call void () %x(" but %x is actually of type "void ()*" and will eventually be just of type "ptr". But this seems not too bad and I don't think it would benefit from repeating the type ("void (), void () * %x(" and then eventually "void (), ptr %x(") as has been done with gep and load. This also has a side benefit: since the explicit type is no longer a pointer, there's no ambiguity between an explicit type and a function that returns a function pointer. Previously this case needed an explicit type (eg: a function returning a void() function was written as "call void () () * @x(" rather than "call void () * @x(" because of the ambiguity between a function returning a pointer to a void() function and a function returning void). No ambiguity means even function pointer return types can just be written alone, without writing the whole function's type. This leaves /only/ the varargs case where the explicit type is required. Given the special type syntax in call instructions, the regex-fu used for migration was a bit more involved in its own unique way (as every one of these is) so here it is. Use it in conjunction with the apply.sh script and associated find/xargs commands I've provided in rr230786 to migrate your out of tree tests. Do let me know if any of this doesn't cover your cases & we can iterate on a more general script/regexes to help others with out of tree tests. About 9 test cases couldn't be automatically migrated - half of those were functions returning function pointers, where I just had to manually delete the function argument types now that we didn't need an explicit function type there. The other half were typedefs of function types used in calls - just had to manually drop the * from those. import fileinput import sys import re pat = re.compile(r'((?:=|:|^|\s)call\s(?:[^@]*?))(\s*$|\s*(?:(?:\[\[[a-zA-Z0-9_]+\]\]|[@%](?:(")?[\\\?@a-zA-Z0-9_.]*?(?(3)"|)|{{.*}}))(?:\(|$)|undef|inttoptr|bitcast|null|asm).*$)') addrspace_end = re.compile(r"addrspace\(\d+\)\s*\*$") func_end = re.compile("(?:void.*|\)\s*)\*$") def conv(match, line): if not match or re.search(addrspace_end, match.group(1)) or not re.search(func_end, match.group(1)): return line return line[:match.start()] + match.group(1)[:match.group(1).rfind('*')].rstrip() + match.group(2) + line[match.end():] for line in sys.stdin: sys.stdout.write(conv(re.search(pat, line), line)) llvm-svn: 235145
2015-04-17 07:24:18 +08:00
%call219 = call i32 (i8*, ...) @printf(i8* getelementptr inbounds ([18 x i8], [18 x i8]* @.str3, i32 0, i32 0), double %conv202, double %conv207, double %conv208, double %conv213, i32 %conv218)
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
call void @clear()
store float 0x3FF7A99300000000, float* @lx, align 4
[opaque pointer type] Add textual IR support for explicit type parameter to gep operator Similar to gep (r230786) and load (r230794) changes. Similar migration script can be used to update test cases, which successfully migrated all of LLVM and Polly, but about 4 test cases needed manually changes in Clang. (this script will read the contents of stdin and massage it into stdout - wrap it in the 'apply.sh' script shown in previous commits + xargs to apply it over a large set of test cases) import fileinput import sys import re rep = re.compile(r"(getelementptr(?:\s+inbounds)?\s*\()((<\d*\s+x\s+)?([^@]*?)(|\s*addrspace\(\d+\))\s*\*(?(3)>)\s*)(?=$|%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|zeroinitializer|<|\[\[[a-zA-Z]|\{\{)", re.MULTILINE | re.DOTALL) def conv(match): line = match.group(1) line += match.group(4) line += ", " line += match.group(2) return line line = sys.stdin.read() off = 0 for match in re.finditer(rep, line): sys.stdout.write(line[off:match.start()]) sys.stdout.write(conv(match)) off = match.end() sys.stdout.write(line[off:]) llvm-svn: 232184
2015-03-14 02:20:45 +08:00
store float 4.500000e+00, float* getelementptr inbounds ({ float, float }, { float, float }* @ret_sc, i32 0, i32 0)
store float 7.000000e+00, float* getelementptr inbounds ({ float, float }, { float, float }* @ret_sc, i32 0, i32 1)
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%217 = load float, float* @lx, align 4
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%call220 = call { float, float } @sc_sf(float %217)
%218 = extractvalue { float, float } %call220, 0
%219 = extractvalue { float, float } %call220, 1
[opaque pointer type] Add textual IR support for explicit type parameter to gep operator Similar to gep (r230786) and load (r230794) changes. Similar migration script can be used to update test cases, which successfully migrated all of LLVM and Polly, but about 4 test cases needed manually changes in Clang. (this script will read the contents of stdin and massage it into stdout - wrap it in the 'apply.sh' script shown in previous commits + xargs to apply it over a large set of test cases) import fileinput import sys import re rep = re.compile(r"(getelementptr(?:\s+inbounds)?\s*\()((<\d*\s+x\s+)?([^@]*?)(|\s*addrspace\(\d+\))\s*\*(?(3)>)\s*)(?=$|%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|zeroinitializer|<|\[\[[a-zA-Z]|\{\{)", re.MULTILINE | re.DOTALL) def conv(match): line = match.group(1) line += match.group(4) line += ", " line += match.group(2) return line line = sys.stdin.read() off = 0 for match in re.finditer(rep, line): sys.stdout.write(line[off:match.start()]) sys.stdout.write(conv(match)) off = match.end() sys.stdout.write(line[off:]) llvm-svn: 232184
2015-03-14 02:20:45 +08:00
store float %218, float* getelementptr inbounds ({ float, float }, { float, float }* @lret_sc, i32 0, i32 0)
store float %219, float* getelementptr inbounds ({ float, float }, { float, float }* @lret_sc, i32 0, i32 1)
%ret_sc.real221 = load float, float* getelementptr inbounds ({ float, float }, { float, float }* @ret_sc, i32 0, i32 0)
%ret_sc.imag222 = load float, float* getelementptr inbounds ({ float, float }, { float, float }* @ret_sc, i32 0, i32 1)
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%conv223 = fpext float %ret_sc.real221 to double
%conv224 = fpext float %ret_sc.imag222 to double
[opaque pointer type] Add textual IR support for explicit type parameter to gep operator Similar to gep (r230786) and load (r230794) changes. Similar migration script can be used to update test cases, which successfully migrated all of LLVM and Polly, but about 4 test cases needed manually changes in Clang. (this script will read the contents of stdin and massage it into stdout - wrap it in the 'apply.sh' script shown in previous commits + xargs to apply it over a large set of test cases) import fileinput import sys import re rep = re.compile(r"(getelementptr(?:\s+inbounds)?\s*\()((<\d*\s+x\s+)?([^@]*?)(|\s*addrspace\(\d+\))\s*\*(?(3)>)\s*)(?=$|%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|zeroinitializer|<|\[\[[a-zA-Z]|\{\{)", re.MULTILINE | re.DOTALL) def conv(match): line = match.group(1) line += match.group(4) line += ", " line += match.group(2) return line line = sys.stdin.read() off = 0 for match in re.finditer(rep, line): sys.stdout.write(line[off:match.start()]) sys.stdout.write(conv(match)) off = match.end() sys.stdout.write(line[off:]) llvm-svn: 232184
2015-03-14 02:20:45 +08:00
%ret_sc.real225 = load float, float* getelementptr inbounds ({ float, float }, { float, float }* @ret_sc, i32 0, i32 0)
%ret_sc.imag226 = load float, float* getelementptr inbounds ({ float, float }, { float, float }* @ret_sc, i32 0, i32 1)
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%conv227 = fpext float %ret_sc.real225 to double
%conv228 = fpext float %ret_sc.imag226 to double
[opaque pointer type] Add textual IR support for explicit type parameter to gep operator Similar to gep (r230786) and load (r230794) changes. Similar migration script can be used to update test cases, which successfully migrated all of LLVM and Polly, but about 4 test cases needed manually changes in Clang. (this script will read the contents of stdin and massage it into stdout - wrap it in the 'apply.sh' script shown in previous commits + xargs to apply it over a large set of test cases) import fileinput import sys import re rep = re.compile(r"(getelementptr(?:\s+inbounds)?\s*\()((<\d*\s+x\s+)?([^@]*?)(|\s*addrspace\(\d+\))\s*\*(?(3)>)\s*)(?=$|%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|zeroinitializer|<|\[\[[a-zA-Z]|\{\{)", re.MULTILINE | re.DOTALL) def conv(match): line = match.group(1) line += match.group(4) line += ", " line += match.group(2) return line line = sys.stdin.read() off = 0 for match in re.finditer(rep, line): sys.stdout.write(line[off:match.start()]) sys.stdout.write(conv(match)) off = match.end() sys.stdout.write(line[off:]) llvm-svn: 232184
2015-03-14 02:20:45 +08:00
%lret_sc.real229 = load float, float* getelementptr inbounds ({ float, float }, { float, float }* @lret_sc, i32 0, i32 0)
%lret_sc.imag230 = load float, float* getelementptr inbounds ({ float, float }, { float, float }* @lret_sc, i32 0, i32 1)
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%conv231 = fpext float %lret_sc.real229 to double
%conv232 = fpext float %lret_sc.imag230 to double
[opaque pointer type] Add textual IR support for explicit type parameter to gep operator Similar to gep (r230786) and load (r230794) changes. Similar migration script can be used to update test cases, which successfully migrated all of LLVM and Polly, but about 4 test cases needed manually changes in Clang. (this script will read the contents of stdin and massage it into stdout - wrap it in the 'apply.sh' script shown in previous commits + xargs to apply it over a large set of test cases) import fileinput import sys import re rep = re.compile(r"(getelementptr(?:\s+inbounds)?\s*\()((<\d*\s+x\s+)?([^@]*?)(|\s*addrspace\(\d+\))\s*\*(?(3)>)\s*)(?=$|%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|zeroinitializer|<|\[\[[a-zA-Z]|\{\{)", re.MULTILINE | re.DOTALL) def conv(match): line = match.group(1) line += match.group(4) line += ", " line += match.group(2) return line line = sys.stdin.read() off = 0 for match in re.finditer(rep, line): sys.stdout.write(line[off:match.start()]) sys.stdout.write(conv(match)) off = match.end() sys.stdout.write(line[off:]) llvm-svn: 232184
2015-03-14 02:20:45 +08:00
%lret_sc.real233 = load float, float* getelementptr inbounds ({ float, float }, { float, float }* @lret_sc, i32 0, i32 0)
%lret_sc.imag234 = load float, float* getelementptr inbounds ({ float, float }, { float, float }* @lret_sc, i32 0, i32 1)
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%conv235 = fpext float %lret_sc.real233 to double
%conv236 = fpext float %lret_sc.imag234 to double
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%220 = load float, float* @x, align 4
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%conv237 = fpext float %220 to double
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%221 = load float, float* @lx, align 4
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%conv238 = fpext float %221 to double
[opaque pointer type] Add textual IR support for explicit type parameter to gep operator Similar to gep (r230786) and load (r230794) changes. Similar migration script can be used to update test cases, which successfully migrated all of LLVM and Polly, but about 4 test cases needed manually changes in Clang. (this script will read the contents of stdin and massage it into stdout - wrap it in the 'apply.sh' script shown in previous commits + xargs to apply it over a large set of test cases) import fileinput import sys import re rep = re.compile(r"(getelementptr(?:\s+inbounds)?\s*\()((<\d*\s+x\s+)?([^@]*?)(|\s*addrspace\(\d+\))\s*\*(?(3)>)\s*)(?=$|%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|zeroinitializer|<|\[\[[a-zA-Z]|\{\{)", re.MULTILINE | re.DOTALL) def conv(match): line = match.group(1) line += match.group(4) line += ", " line += match.group(2) return line line = sys.stdin.read() off = 0 for match in re.finditer(rep, line): sys.stdout.write(line[off:match.start()]) sys.stdout.write(conv(match)) off = match.end() sys.stdout.write(line[off:]) llvm-svn: 232184
2015-03-14 02:20:45 +08:00
%ret_sc.real239 = load float, float* getelementptr inbounds ({ float, float }, { float, float }* @ret_sc, i32 0, i32 0)
%ret_sc.imag240 = load float, float* getelementptr inbounds ({ float, float }, { float, float }* @ret_sc, i32 0, i32 1)
%lret_sc.real241 = load float, float* getelementptr inbounds ({ float, float }, { float, float }* @lret_sc, i32 0, i32 0)
%lret_sc.imag242 = load float, float* getelementptr inbounds ({ float, float }, { float, float }* @lret_sc, i32 0, i32 1)
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%cmp.r243 = fcmp oeq float %ret_sc.real239, %lret_sc.real241
%cmp.i244 = fcmp oeq float %ret_sc.imag240, %lret_sc.imag242
%and.ri245 = and i1 %cmp.r243, %cmp.i244
br i1 %and.ri245, label %land.rhs247, label %land.end250
land.rhs247: ; preds = %land.end198
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%222 = load float, float* @x, align 4
%223 = load float, float* @lx, align 4
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%cmp248 = fcmp oeq float %222, %223
br label %land.end250
land.end250: ; preds = %land.rhs247, %land.end198
%224 = phi i1 [ false, %land.end198 ], [ %cmp248, %land.rhs247 ]
%land.ext251 = zext i1 %224 to i32
[opaque pointer type] Add textual IR support for explicit type parameter to the call instruction See r230786 and r230794 for similar changes to gep and load respectively. Call is a bit different because it often doesn't have a single explicit type - usually the type is deduced from the arguments, and just the return type is explicit. In those cases there's no need to change the IR. When that's not the case, the IR usually contains the pointer type of the first operand - but since typed pointers are going away, that representation is insufficient so I'm just stripping the "pointerness" of the explicit type away. This does make the IR a bit weird - it /sort of/ reads like the type of the first operand: "call void () %x(" but %x is actually of type "void ()*" and will eventually be just of type "ptr". But this seems not too bad and I don't think it would benefit from repeating the type ("void (), void () * %x(" and then eventually "void (), ptr %x(") as has been done with gep and load. This also has a side benefit: since the explicit type is no longer a pointer, there's no ambiguity between an explicit type and a function that returns a function pointer. Previously this case needed an explicit type (eg: a function returning a void() function was written as "call void () () * @x(" rather than "call void () * @x(" because of the ambiguity between a function returning a pointer to a void() function and a function returning void). No ambiguity means even function pointer return types can just be written alone, without writing the whole function's type. This leaves /only/ the varargs case where the explicit type is required. Given the special type syntax in call instructions, the regex-fu used for migration was a bit more involved in its own unique way (as every one of these is) so here it is. Use it in conjunction with the apply.sh script and associated find/xargs commands I've provided in rr230786 to migrate your out of tree tests. Do let me know if any of this doesn't cover your cases & we can iterate on a more general script/regexes to help others with out of tree tests. About 9 test cases couldn't be automatically migrated - half of those were functions returning function pointers, where I just had to manually delete the function argument types now that we didn't need an explicit function type there. The other half were typedefs of function types used in calls - just had to manually drop the * from those. import fileinput import sys import re pat = re.compile(r'((?:=|:|^|\s)call\s(?:[^@]*?))(\s*$|\s*(?:(?:\[\[[a-zA-Z0-9_]+\]\]|[@%](?:(")?[\\\?@a-zA-Z0-9_.]*?(?(3)"|)|{{.*}}))(?:\(|$)|undef|inttoptr|bitcast|null|asm).*$)') addrspace_end = re.compile(r"addrspace\(\d+\)\s*\*$") func_end = re.compile("(?:void.*|\)\s*)\*$") def conv(match, line): if not match or re.search(addrspace_end, match.group(1)) or not re.search(func_end, match.group(1)): return line return line[:match.start()] + match.group(1)[:match.group(1).rfind('*')].rstrip() + match.group(2) + line[match.end():] for line in sys.stdin: sys.stdout.write(conv(re.search(pat, line), line)) llvm-svn: 235145
2015-04-17 07:24:18 +08:00
%call252 = call i32 (i8*, ...) @printf(i8* getelementptr inbounds ([24 x i8], [24 x i8]* @.str4, i32 0, i32 0), double %conv223, double %conv228, double %conv231, double %conv236, double %conv237, double %conv238, i32 %land.ext251)
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
call void @clear()
[opaque pointer type] Add textual IR support for explicit type parameter to gep operator Similar to gep (r230786) and load (r230794) changes. Similar migration script can be used to update test cases, which successfully migrated all of LLVM and Polly, but about 4 test cases needed manually changes in Clang. (this script will read the contents of stdin and massage it into stdout - wrap it in the 'apply.sh' script shown in previous commits + xargs to apply it over a large set of test cases) import fileinput import sys import re rep = re.compile(r"(getelementptr(?:\s+inbounds)?\s*\()((<\d*\s+x\s+)?([^@]*?)(|\s*addrspace\(\d+\))\s*\*(?(3)>)\s*)(?=$|%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|zeroinitializer|<|\[\[[a-zA-Z]|\{\{)", re.MULTILINE | re.DOTALL) def conv(match): line = match.group(1) line += match.group(4) line += ", " line += match.group(2) return line line = sys.stdin.read() off = 0 for match in re.finditer(rep, line): sys.stdout.write(line[off:match.start()]) sys.stdout.write(conv(match)) off = match.end() sys.stdout.write(line[off:]) llvm-svn: 232184
2015-03-14 02:20:45 +08:00
store double 1.234500e+03, double* getelementptr inbounds ({ double, double }, { double, double }* @ret_dc, i32 0, i32 0)
store double 7.677000e+03, double* getelementptr inbounds ({ double, double }, { double, double }* @ret_dc, i32 0, i32 1)
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%call253 = call { double, double } @dc_v()
%225 = extractvalue { double, double } %call253, 0
%226 = extractvalue { double, double } %call253, 1
[opaque pointer type] Add textual IR support for explicit type parameter to gep operator Similar to gep (r230786) and load (r230794) changes. Similar migration script can be used to update test cases, which successfully migrated all of LLVM and Polly, but about 4 test cases needed manually changes in Clang. (this script will read the contents of stdin and massage it into stdout - wrap it in the 'apply.sh' script shown in previous commits + xargs to apply it over a large set of test cases) import fileinput import sys import re rep = re.compile(r"(getelementptr(?:\s+inbounds)?\s*\()((<\d*\s+x\s+)?([^@]*?)(|\s*addrspace\(\d+\))\s*\*(?(3)>)\s*)(?=$|%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|zeroinitializer|<|\[\[[a-zA-Z]|\{\{)", re.MULTILINE | re.DOTALL) def conv(match): line = match.group(1) line += match.group(4) line += ", " line += match.group(2) return line line = sys.stdin.read() off = 0 for match in re.finditer(rep, line): sys.stdout.write(line[off:match.start()]) sys.stdout.write(conv(match)) off = match.end() sys.stdout.write(line[off:]) llvm-svn: 232184
2015-03-14 02:20:45 +08:00
store double %225, double* getelementptr inbounds ({ double, double }, { double, double }* @lret_dc, i32 0, i32 0)
store double %226, double* getelementptr inbounds ({ double, double }, { double, double }* @lret_dc, i32 0, i32 1)
%ret_dc.real = load double, double* getelementptr inbounds ({ double, double }, { double, double }* @ret_dc, i32 0, i32 0)
%ret_dc.imag = load double, double* getelementptr inbounds ({ double, double }, { double, double }* @ret_dc, i32 0, i32 1)
%ret_dc.real254 = load double, double* getelementptr inbounds ({ double, double }, { double, double }* @ret_dc, i32 0, i32 0)
%ret_dc.imag255 = load double, double* getelementptr inbounds ({ double, double }, { double, double }* @ret_dc, i32 0, i32 1)
%lret_dc.real = load double, double* getelementptr inbounds ({ double, double }, { double, double }* @lret_dc, i32 0, i32 0)
%lret_dc.imag = load double, double* getelementptr inbounds ({ double, double }, { double, double }* @lret_dc, i32 0, i32 1)
%lret_dc.real256 = load double, double* getelementptr inbounds ({ double, double }, { double, double }* @lret_dc, i32 0, i32 0)
%lret_dc.imag257 = load double, double* getelementptr inbounds ({ double, double }, { double, double }* @lret_dc, i32 0, i32 1)
%ret_dc.real258 = load double, double* getelementptr inbounds ({ double, double }, { double, double }* @ret_dc, i32 0, i32 0)
%ret_dc.imag259 = load double, double* getelementptr inbounds ({ double, double }, { double, double }* @ret_dc, i32 0, i32 1)
%lret_dc.real260 = load double, double* getelementptr inbounds ({ double, double }, { double, double }* @lret_dc, i32 0, i32 0)
%lret_dc.imag261 = load double, double* getelementptr inbounds ({ double, double }, { double, double }* @lret_dc, i32 0, i32 1)
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%cmp.r262 = fcmp oeq double %ret_dc.real258, %lret_dc.real260
%cmp.i263 = fcmp oeq double %ret_dc.imag259, %lret_dc.imag261
%and.ri264 = and i1 %cmp.r262, %cmp.i263
%conv265 = zext i1 %and.ri264 to i32
[opaque pointer type] Add textual IR support for explicit type parameter to the call instruction See r230786 and r230794 for similar changes to gep and load respectively. Call is a bit different because it often doesn't have a single explicit type - usually the type is deduced from the arguments, and just the return type is explicit. In those cases there's no need to change the IR. When that's not the case, the IR usually contains the pointer type of the first operand - but since typed pointers are going away, that representation is insufficient so I'm just stripping the "pointerness" of the explicit type away. This does make the IR a bit weird - it /sort of/ reads like the type of the first operand: "call void () %x(" but %x is actually of type "void ()*" and will eventually be just of type "ptr". But this seems not too bad and I don't think it would benefit from repeating the type ("void (), void () * %x(" and then eventually "void (), ptr %x(") as has been done with gep and load. This also has a side benefit: since the explicit type is no longer a pointer, there's no ambiguity between an explicit type and a function that returns a function pointer. Previously this case needed an explicit type (eg: a function returning a void() function was written as "call void () () * @x(" rather than "call void () * @x(" because of the ambiguity between a function returning a pointer to a void() function and a function returning void). No ambiguity means even function pointer return types can just be written alone, without writing the whole function's type. This leaves /only/ the varargs case where the explicit type is required. Given the special type syntax in call instructions, the regex-fu used for migration was a bit more involved in its own unique way (as every one of these is) so here it is. Use it in conjunction with the apply.sh script and associated find/xargs commands I've provided in rr230786 to migrate your out of tree tests. Do let me know if any of this doesn't cover your cases & we can iterate on a more general script/regexes to help others with out of tree tests. About 9 test cases couldn't be automatically migrated - half of those were functions returning function pointers, where I just had to manually delete the function argument types now that we didn't need an explicit function type there. The other half were typedefs of function types used in calls - just had to manually drop the * from those. import fileinput import sys import re pat = re.compile(r'((?:=|:|^|\s)call\s(?:[^@]*?))(\s*$|\s*(?:(?:\[\[[a-zA-Z0-9_]+\]\]|[@%](?:(")?[\\\?@a-zA-Z0-9_.]*?(?(3)"|)|{{.*}}))(?:\(|$)|undef|inttoptr|bitcast|null|asm).*$)') addrspace_end = re.compile(r"addrspace\(\d+\)\s*\*$") func_end = re.compile("(?:void.*|\)\s*)\*$") def conv(match, line): if not match or re.search(addrspace_end, match.group(1)) or not re.search(func_end, match.group(1)): return line return line[:match.start()] + match.group(1)[:match.group(1).rfind('*')].rstrip() + match.group(2) + line[match.end():] for line in sys.stdin: sys.stdout.write(conv(re.search(pat, line), line)) llvm-svn: 235145
2015-04-17 07:24:18 +08:00
%call266 = call i32 (i8*, ...) @printf(i8* getelementptr inbounds ([18 x i8], [18 x i8]* @.str3, i32 0, i32 0), double %ret_dc.real, double %ret_dc.imag255, double %lret_dc.real, double %lret_dc.imag257, i32 %conv265)
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
call void @clear()
store double 0x40AAF6F532617C1C, double* @lxd, align 8
[opaque pointer type] Add textual IR support for explicit type parameter to gep operator Similar to gep (r230786) and load (r230794) changes. Similar migration script can be used to update test cases, which successfully migrated all of LLVM and Polly, but about 4 test cases needed manually changes in Clang. (this script will read the contents of stdin and massage it into stdout - wrap it in the 'apply.sh' script shown in previous commits + xargs to apply it over a large set of test cases) import fileinput import sys import re rep = re.compile(r"(getelementptr(?:\s+inbounds)?\s*\()((<\d*\s+x\s+)?([^@]*?)(|\s*addrspace\(\d+\))\s*\*(?(3)>)\s*)(?=$|%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|zeroinitializer|<|\[\[[a-zA-Z]|\{\{)", re.MULTILINE | re.DOTALL) def conv(match): line = match.group(1) line += match.group(4) line += ", " line += match.group(2) return line line = sys.stdin.read() off = 0 for match in re.finditer(rep, line): sys.stdout.write(line[off:match.start()]) sys.stdout.write(conv(match)) off = match.end() sys.stdout.write(line[off:]) llvm-svn: 232184
2015-03-14 02:20:45 +08:00
store double 4.444500e+03, double* getelementptr inbounds ({ double, double }, { double, double }* @ret_dc, i32 0, i32 0)
store double 7.888000e+03, double* getelementptr inbounds ({ double, double }, { double, double }* @ret_dc, i32 0, i32 1)
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%227 = load float, float* @lx, align 4
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%call267 = call { double, double } @dc_sf(float %227)
%228 = extractvalue { double, double } %call267, 0
%229 = extractvalue { double, double } %call267, 1
[opaque pointer type] Add textual IR support for explicit type parameter to gep operator Similar to gep (r230786) and load (r230794) changes. Similar migration script can be used to update test cases, which successfully migrated all of LLVM and Polly, but about 4 test cases needed manually changes in Clang. (this script will read the contents of stdin and massage it into stdout - wrap it in the 'apply.sh' script shown in previous commits + xargs to apply it over a large set of test cases) import fileinput import sys import re rep = re.compile(r"(getelementptr(?:\s+inbounds)?\s*\()((<\d*\s+x\s+)?([^@]*?)(|\s*addrspace\(\d+\))\s*\*(?(3)>)\s*)(?=$|%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|zeroinitializer|<|\[\[[a-zA-Z]|\{\{)", re.MULTILINE | re.DOTALL) def conv(match): line = match.group(1) line += match.group(4) line += ", " line += match.group(2) return line line = sys.stdin.read() off = 0 for match in re.finditer(rep, line): sys.stdout.write(line[off:match.start()]) sys.stdout.write(conv(match)) off = match.end() sys.stdout.write(line[off:]) llvm-svn: 232184
2015-03-14 02:20:45 +08:00
store double %228, double* getelementptr inbounds ({ double, double }, { double, double }* @lret_dc, i32 0, i32 0)
store double %229, double* getelementptr inbounds ({ double, double }, { double, double }* @lret_dc, i32 0, i32 1)
%ret_dc.real268 = load double, double* getelementptr inbounds ({ double, double }, { double, double }* @ret_dc, i32 0, i32 0)
%ret_dc.imag269 = load double, double* getelementptr inbounds ({ double, double }, { double, double }* @ret_dc, i32 0, i32 1)
%ret_dc.real270 = load double, double* getelementptr inbounds ({ double, double }, { double, double }* @ret_dc, i32 0, i32 0)
%ret_dc.imag271 = load double, double* getelementptr inbounds ({ double, double }, { double, double }* @ret_dc, i32 0, i32 1)
%lret_dc.real272 = load double, double* getelementptr inbounds ({ double, double }, { double, double }* @lret_dc, i32 0, i32 0)
%lret_dc.imag273 = load double, double* getelementptr inbounds ({ double, double }, { double, double }* @lret_dc, i32 0, i32 1)
%lret_dc.real274 = load double, double* getelementptr inbounds ({ double, double }, { double, double }* @lret_dc, i32 0, i32 0)
%lret_dc.imag275 = load double, double* getelementptr inbounds ({ double, double }, { double, double }* @lret_dc, i32 0, i32 1)
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%230 = load float, float* @x, align 4
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%conv276 = fpext float %230 to double
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%231 = load float, float* @lx, align 4
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%conv277 = fpext float %231 to double
[opaque pointer type] Add textual IR support for explicit type parameter to gep operator Similar to gep (r230786) and load (r230794) changes. Similar migration script can be used to update test cases, which successfully migrated all of LLVM and Polly, but about 4 test cases needed manually changes in Clang. (this script will read the contents of stdin and massage it into stdout - wrap it in the 'apply.sh' script shown in previous commits + xargs to apply it over a large set of test cases) import fileinput import sys import re rep = re.compile(r"(getelementptr(?:\s+inbounds)?\s*\()((<\d*\s+x\s+)?([^@]*?)(|\s*addrspace\(\d+\))\s*\*(?(3)>)\s*)(?=$|%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|zeroinitializer|<|\[\[[a-zA-Z]|\{\{)", re.MULTILINE | re.DOTALL) def conv(match): line = match.group(1) line += match.group(4) line += ", " line += match.group(2) return line line = sys.stdin.read() off = 0 for match in re.finditer(rep, line): sys.stdout.write(line[off:match.start()]) sys.stdout.write(conv(match)) off = match.end() sys.stdout.write(line[off:]) llvm-svn: 232184
2015-03-14 02:20:45 +08:00
%ret_dc.real278 = load double, double* getelementptr inbounds ({ double, double }, { double, double }* @ret_dc, i32 0, i32 0)
%ret_dc.imag279 = load double, double* getelementptr inbounds ({ double, double }, { double, double }* @ret_dc, i32 0, i32 1)
%lret_dc.real280 = load double, double* getelementptr inbounds ({ double, double }, { double, double }* @lret_dc, i32 0, i32 0)
%lret_dc.imag281 = load double, double* getelementptr inbounds ({ double, double }, { double, double }* @lret_dc, i32 0, i32 1)
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%cmp.r282 = fcmp oeq double %ret_dc.real278, %lret_dc.real280
%cmp.i283 = fcmp oeq double %ret_dc.imag279, %lret_dc.imag281
%and.ri284 = and i1 %cmp.r282, %cmp.i283
br i1 %and.ri284, label %land.rhs286, label %land.end289
land.rhs286: ; preds = %land.end250
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%232 = load float, float* @x, align 4
%233 = load float, float* @lx, align 4
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
%cmp287 = fcmp oeq float %232, %233
br label %land.end289
land.end289: ; preds = %land.rhs286, %land.end250
%234 = phi i1 [ false, %land.end250 ], [ %cmp287, %land.rhs286 ]
%land.ext290 = zext i1 %234 to i32
[opaque pointer type] Add textual IR support for explicit type parameter to the call instruction See r230786 and r230794 for similar changes to gep and load respectively. Call is a bit different because it often doesn't have a single explicit type - usually the type is deduced from the arguments, and just the return type is explicit. In those cases there's no need to change the IR. When that's not the case, the IR usually contains the pointer type of the first operand - but since typed pointers are going away, that representation is insufficient so I'm just stripping the "pointerness" of the explicit type away. This does make the IR a bit weird - it /sort of/ reads like the type of the first operand: "call void () %x(" but %x is actually of type "void ()*" and will eventually be just of type "ptr". But this seems not too bad and I don't think it would benefit from repeating the type ("void (), void () * %x(" and then eventually "void (), ptr %x(") as has been done with gep and load. This also has a side benefit: since the explicit type is no longer a pointer, there's no ambiguity between an explicit type and a function that returns a function pointer. Previously this case needed an explicit type (eg: a function returning a void() function was written as "call void () () * @x(" rather than "call void () * @x(" because of the ambiguity between a function returning a pointer to a void() function and a function returning void). No ambiguity means even function pointer return types can just be written alone, without writing the whole function's type. This leaves /only/ the varargs case where the explicit type is required. Given the special type syntax in call instructions, the regex-fu used for migration was a bit more involved in its own unique way (as every one of these is) so here it is. Use it in conjunction with the apply.sh script and associated find/xargs commands I've provided in rr230786 to migrate your out of tree tests. Do let me know if any of this doesn't cover your cases & we can iterate on a more general script/regexes to help others with out of tree tests. About 9 test cases couldn't be automatically migrated - half of those were functions returning function pointers, where I just had to manually delete the function argument types now that we didn't need an explicit function type there. The other half were typedefs of function types used in calls - just had to manually drop the * from those. import fileinput import sys import re pat = re.compile(r'((?:=|:|^|\s)call\s(?:[^@]*?))(\s*$|\s*(?:(?:\[\[[a-zA-Z0-9_]+\]\]|[@%](?:(")?[\\\?@a-zA-Z0-9_.]*?(?(3)"|)|{{.*}}))(?:\(|$)|undef|inttoptr|bitcast|null|asm).*$)') addrspace_end = re.compile(r"addrspace\(\d+\)\s*\*$") func_end = re.compile("(?:void.*|\)\s*)\*$") def conv(match, line): if not match or re.search(addrspace_end, match.group(1)) or not re.search(func_end, match.group(1)): return line return line[:match.start()] + match.group(1)[:match.group(1).rfind('*')].rstrip() + match.group(2) + line[match.end():] for line in sys.stdin: sys.stdout.write(conv(re.search(pat, line), line)) llvm-svn: 235145
2015-04-17 07:24:18 +08:00
%call291 = call i32 (i8*, ...) @printf(i8* getelementptr inbounds ([24 x i8], [24 x i8]* @.str4, i32 0, i32 0), double %ret_dc.real268, double %ret_dc.imag271, double %lret_dc.real272, double %lret_dc.imag275, double %conv276, double %conv277, i32 %land.ext290)
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%235 = load i32, i32* %retval
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
ret i32 %235
}
declare void @v_sf(float) #1
; stel: .section .mips16.call.fp.v_sf,"ax",@progbits
Fix some misc. issues with Mips16 fp stubs. 1) They should never be inlined. 2) A naming inconsistency with gcc mips16 3) Stubs should not have the global attribute llvm-svn: 187555
2013-08-01 10:26:31 +08:00
; stel: .ent __call_stub_fp_v_sf
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
; stel: mtc1 $4,$f12
; stel: lui $25,%hi(v_sf)
; stel: addiu $25,$25,%lo(v_sf)
; stel: jr $25
Fix some misc. issues with Mips16 fp stubs. 1) They should never be inlined. 2) A naming inconsistency with gcc mips16 3) Stubs should not have the global attribute llvm-svn: 187555
2013-08-01 10:26:31 +08:00
; stel: .end __call_stub_fp_v_sf
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
declare i32 @printf(i8*, ...) #1
declare void @v_df(double) #1
; stel: .section .mips16.call.fp.v_df,"ax",@progbits
Fix some misc. issues with Mips16 fp stubs. 1) They should never be inlined. 2) A naming inconsistency with gcc mips16 3) Stubs should not have the global attribute llvm-svn: 187555
2013-08-01 10:26:31 +08:00
; stel: .ent __call_stub_fp_v_df
; stel: #APP
; setl: .set reorder
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
; stel: mtc1 $4,$f12
; stel: mtc1 $5,$f13
; stel: lui $25,%hi(v_df)
; stel: addiu $25,$25,%lo(v_df)
; stel: jr $25
Fix some misc. issues with Mips16 fp stubs. 1) They should never be inlined. 2) A naming inconsistency with gcc mips16 3) Stubs should not have the global attribute llvm-svn: 187555
2013-08-01 10:26:31 +08:00
; stel: .end __call_stub_fp_v_df
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
declare void @v_sf_sf(float, float) #1
; stel: .section .mips16.call.fp.v_sf_sf,"ax",@progbits
Fix some misc. issues with Mips16 fp stubs. 1) They should never be inlined. 2) A naming inconsistency with gcc mips16 3) Stubs should not have the global attribute llvm-svn: 187555
2013-08-01 10:26:31 +08:00
; stel: .ent __call_stub_fp_v_sf_sf
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
; stel: mtc1 $4,$f12
; stel: mtc1 $5,$f14
; stel: lui $25,%hi(v_sf_sf)
; stel: addiu $25,$25,%lo(v_sf_sf)
; stel: jr $25
Fix some misc. issues with Mips16 fp stubs. 1) They should never be inlined. 2) A naming inconsistency with gcc mips16 3) Stubs should not have the global attribute llvm-svn: 187555
2013-08-01 10:26:31 +08:00
; stel: .end __call_stub_fp_v_sf_sf
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
declare void @v_sf_df(float, double) #1
; stel: .section .mips16.call.fp.v_sf_df,"ax",@progbits
Fix some misc. issues with Mips16 fp stubs. 1) They should never be inlined. 2) A naming inconsistency with gcc mips16 3) Stubs should not have the global attribute llvm-svn: 187555
2013-08-01 10:26:31 +08:00
; stel: .ent __call_stub_fp_v_sf_df
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
; stel: mtc1 $4,$f12
; stel: mtc1 $6,$f14
; stel: mtc1 $7,$f15
; stel: lui $25,%hi(v_sf_df)
; stel: addiu $25,$25,%lo(v_sf_df)
; stel: jr $25
Fix some misc. issues with Mips16 fp stubs. 1) They should never be inlined. 2) A naming inconsistency with gcc mips16 3) Stubs should not have the global attribute llvm-svn: 187555
2013-08-01 10:26:31 +08:00
; stel: .end __call_stub_fp_v_sf_df
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
declare void @v_df_sf(double, float) #1
; stel: .section .mips16.call.fp.v_df_sf,"ax",@progbits
Fix some misc. issues with Mips16 fp stubs. 1) They should never be inlined. 2) A naming inconsistency with gcc mips16 3) Stubs should not have the global attribute llvm-svn: 187555
2013-08-01 10:26:31 +08:00
; stel: .ent __call_stub_fp_v_df_sf
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
; stel: mtc1 $4,$f12
; stel: mtc1 $5,$f13
; stel: mtc1 $6,$f14
; stel: lui $25,%hi(v_df_sf)
; stel: addiu $25,$25,%lo(v_df_sf)
; stel: jr $25
Fix some misc. issues with Mips16 fp stubs. 1) They should never be inlined. 2) A naming inconsistency with gcc mips16 3) Stubs should not have the global attribute llvm-svn: 187555
2013-08-01 10:26:31 +08:00
; stel: .end __call_stub_fp_v_df_sf
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
declare void @v_df_df(double, double) #1
; stel: .section .mips16.call.fp.v_df_df,"ax",@progbits
Fix some misc. issues with Mips16 fp stubs. 1) They should never be inlined. 2) A naming inconsistency with gcc mips16 3) Stubs should not have the global attribute llvm-svn: 187555
2013-08-01 10:26:31 +08:00
; stel: .ent __call_stub_fp_v_df_df
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
; stel: mtc1 $4,$f12
; stel: mtc1 $5,$f13
; stel: mtc1 $6,$f14
; stel: mtc1 $7,$f15
; stel: lui $25,%hi(v_df_df)
; stel: addiu $25,$25,%lo(v_df_df)
; stel: jr $25
Fix some misc. issues with Mips16 fp stubs. 1) They should never be inlined. 2) A naming inconsistency with gcc mips16 3) Stubs should not have the global attribute llvm-svn: 187555
2013-08-01 10:26:31 +08:00
; stel: .end __call_stub_fp_v_df_df
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
declare float @sf_v() #1
; stel: .section .mips16.call.fp.sf_v,"ax",@progbits
Fix some misc. issues with Mips16 fp stubs. 1) They should never be inlined. 2) A naming inconsistency with gcc mips16 3) Stubs should not have the global attribute llvm-svn: 187555
2013-08-01 10:26:31 +08:00
; stel: .ent __call_stub_fp_sf_v
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
; stel: move $18, $31
; stel: jal sf_v
; stel: mfc1 $2,$f0
; stel: jr $18
Fix some misc. issues with Mips16 fp stubs. 1) They should never be inlined. 2) A naming inconsistency with gcc mips16 3) Stubs should not have the global attribute llvm-svn: 187555
2013-08-01 10:26:31 +08:00
; stel: .end __call_stub_fp_sf_v
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
declare float @sf_sf(float) #1
; stel: .section .mips16.call.fp.sf_sf,"ax",@progbits
Fix some misc. issues with Mips16 fp stubs. 1) They should never be inlined. 2) A naming inconsistency with gcc mips16 3) Stubs should not have the global attribute llvm-svn: 187555
2013-08-01 10:26:31 +08:00
; stel: .ent __call_stub_fp_sf_sf
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
; stel: mtc1 $4,$f12
; stel: move $18, $31
; stel: jal sf_sf
; stel: mfc1 $2,$f0
; stel: jr $18
Fix some misc. issues with Mips16 fp stubs. 1) They should never be inlined. 2) A naming inconsistency with gcc mips16 3) Stubs should not have the global attribute llvm-svn: 187555
2013-08-01 10:26:31 +08:00
; stel: .end __call_stub_fp_sf_sf
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
declare float @sf_df(double) #1
; stel: .section .mips16.call.fp.sf_df,"ax",@progbits
Fix some misc. issues with Mips16 fp stubs. 1) They should never be inlined. 2) A naming inconsistency with gcc mips16 3) Stubs should not have the global attribute llvm-svn: 187555
2013-08-01 10:26:31 +08:00
; stel: .ent __call_stub_fp_sf_df
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
; stel: mtc1 $4,$f12
; stel: mtc1 $5,$f13
; stel: move $18, $31
; stel: jal sf_df
; stel: mfc1 $2,$f0
; stel: jr $18
Fix some misc. issues with Mips16 fp stubs. 1) They should never be inlined. 2) A naming inconsistency with gcc mips16 3) Stubs should not have the global attribute llvm-svn: 187555
2013-08-01 10:26:31 +08:00
; stel: .end __call_stub_fp_sf_df
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
declare float @sf_sf_sf(float, float) #1
; stel: .section .mips16.call.fp.sf_sf_sf,"ax",@progbits
Fix some misc. issues with Mips16 fp stubs. 1) They should never be inlined. 2) A naming inconsistency with gcc mips16 3) Stubs should not have the global attribute llvm-svn: 187555
2013-08-01 10:26:31 +08:00
; stel: .ent __call_stub_fp_sf_sf_sf
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
; stel: mtc1 $4,$f12
; stel: mtc1 $5,$f14
; stel: move $18, $31
; stel: jal sf_sf_sf
; stel: mfc1 $2,$f0
; stel: jr $18
Fix some misc. issues with Mips16 fp stubs. 1) They should never be inlined. 2) A naming inconsistency with gcc mips16 3) Stubs should not have the global attribute llvm-svn: 187555
2013-08-01 10:26:31 +08:00
; stel: .end __call_stub_fp_sf_sf_sf
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
declare float @sf_sf_df(float, double) #1
; stel: .section .mips16.call.fp.sf_sf_df,"ax",@progbits
Fix some misc. issues with Mips16 fp stubs. 1) They should never be inlined. 2) A naming inconsistency with gcc mips16 3) Stubs should not have the global attribute llvm-svn: 187555
2013-08-01 10:26:31 +08:00
; stel: .ent __call_stub_fp_sf_sf_df
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
; stel: mtc1 $4,$f12
; stel: mtc1 $6,$f14
; stel: mtc1 $7,$f15
; stel: move $18, $31
; stel: jal sf_sf_df
; stel: mfc1 $2,$f0
; stel: jr $18
Fix some misc. issues with Mips16 fp stubs. 1) They should never be inlined. 2) A naming inconsistency with gcc mips16 3) Stubs should not have the global attribute llvm-svn: 187555
2013-08-01 10:26:31 +08:00
; stel: .end __call_stub_fp_sf_sf_df
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
declare float @sf_df_sf(double, float) #1
; stel: .section .mips16.call.fp.sf_df_sf,"ax",@progbits
Fix some misc. issues with Mips16 fp stubs. 1) They should never be inlined. 2) A naming inconsistency with gcc mips16 3) Stubs should not have the global attribute llvm-svn: 187555
2013-08-01 10:26:31 +08:00
; stel: .ent __call_stub_fp_sf_df_sf
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
; stel: mtc1 $4,$f12
; stel: mtc1 $5,$f13
; stel: mtc1 $6,$f14
; stel: move $18, $31
; stel: jal sf_df_sf
; stel: mfc1 $2,$f0
; stel: jr $18
Fix some misc. issues with Mips16 fp stubs. 1) They should never be inlined. 2) A naming inconsistency with gcc mips16 3) Stubs should not have the global attribute llvm-svn: 187555
2013-08-01 10:26:31 +08:00
; stel: .end __call_stub_fp_sf_df_sf
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
declare float @sf_df_df(double, double) #1
; stel: .section .mips16.call.fp.sf_df_df,"ax",@progbits
Fix some misc. issues with Mips16 fp stubs. 1) They should never be inlined. 2) A naming inconsistency with gcc mips16 3) Stubs should not have the global attribute llvm-svn: 187555
2013-08-01 10:26:31 +08:00
; stel: .ent __call_stub_fp_sf_df_df
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
; stel: mtc1 $4,$f12
; stel: mtc1 $5,$f13
; stel: mtc1 $6,$f14
; stel: mtc1 $7,$f15
; stel: move $18, $31
; stel: jal sf_df_df
; stel: mfc1 $2,$f0
; stel: jr $18
Fix some misc. issues with Mips16 fp stubs. 1) They should never be inlined. 2) A naming inconsistency with gcc mips16 3) Stubs should not have the global attribute llvm-svn: 187555
2013-08-01 10:26:31 +08:00
; stel: .end __call_stub_fp_sf_df_df
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
declare double @df_v() #1
; stel: .section .mips16.call.fp.df_v,"ax",@progbits
Fix some misc. issues with Mips16 fp stubs. 1) They should never be inlined. 2) A naming inconsistency with gcc mips16 3) Stubs should not have the global attribute llvm-svn: 187555
2013-08-01 10:26:31 +08:00
; stel: .ent __call_stub_fp_df_v
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
; stel: move $18, $31
; stel: jal df_v
; stel: mfc1 $2,$f0
; stel: mfc1 $3,$f1
; stel: jr $18
Fix some misc. issues with Mips16 fp stubs. 1) They should never be inlined. 2) A naming inconsistency with gcc mips16 3) Stubs should not have the global attribute llvm-svn: 187555
2013-08-01 10:26:31 +08:00
; stel: .end __call_stub_fp_df_v
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
declare double @df_sf(float) #1
; stel: .section .mips16.call.fp.df_sf,"ax",@progbits
Fix some misc. issues with Mips16 fp stubs. 1) They should never be inlined. 2) A naming inconsistency with gcc mips16 3) Stubs should not have the global attribute llvm-svn: 187555
2013-08-01 10:26:31 +08:00
; stel: .ent __call_stub_fp_df_sf
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
; stel: mtc1 $4,$f12
; stel: move $18, $31
; stel: jal df_sf
; stel: mfc1 $2,$f0
; stel: mfc1 $3,$f1
; stel: jr $18
Fix some misc. issues with Mips16 fp stubs. 1) They should never be inlined. 2) A naming inconsistency with gcc mips16 3) Stubs should not have the global attribute llvm-svn: 187555
2013-08-01 10:26:31 +08:00
; stel: .end __call_stub_fp_df_sf
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
declare double @df_df(double) #1
; stel: .section .mips16.call.fp.df_df,"ax",@progbits
Fix some misc. issues with Mips16 fp stubs. 1) They should never be inlined. 2) A naming inconsistency with gcc mips16 3) Stubs should not have the global attribute llvm-svn: 187555
2013-08-01 10:26:31 +08:00
; stel: .ent __call_stub_fp_df_df
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
; stel: mtc1 $4,$f12
; stel: mtc1 $5,$f13
; stel: move $18, $31
; stel: jal df_df
; stel: mfc1 $2,$f0
; stel: mfc1 $3,$f1
; stel: jr $18
Fix some misc. issues with Mips16 fp stubs. 1) They should never be inlined. 2) A naming inconsistency with gcc mips16 3) Stubs should not have the global attribute llvm-svn: 187555
2013-08-01 10:26:31 +08:00
; stel: .end __call_stub_fp_df_df
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
declare double @df_sf_sf(float, float) #1
; stel: .section .mips16.call.fp.df_sf_sf,"ax",@progbits
Fix some misc. issues with Mips16 fp stubs. 1) They should never be inlined. 2) A naming inconsistency with gcc mips16 3) Stubs should not have the global attribute llvm-svn: 187555
2013-08-01 10:26:31 +08:00
; stel: .ent __call_stub_fp_df_sf_sf
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
; stel: mtc1 $4,$f12
; stel: mtc1 $5,$f14
; stel: move $18, $31
; stel: jal df_sf_sf
; stel: mfc1 $2,$f0
; stel: mfc1 $3,$f1
; stel: jr $18
Fix some misc. issues with Mips16 fp stubs. 1) They should never be inlined. 2) A naming inconsistency with gcc mips16 3) Stubs should not have the global attribute llvm-svn: 187555
2013-08-01 10:26:31 +08:00
; stel: .end __call_stub_fp_df_sf_sf
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
declare double @df_sf_df(float, double) #1
; stel: .section .mips16.call.fp.df_sf_df,"ax",@progbits
Fix some misc. issues with Mips16 fp stubs. 1) They should never be inlined. 2) A naming inconsistency with gcc mips16 3) Stubs should not have the global attribute llvm-svn: 187555
2013-08-01 10:26:31 +08:00
; stel: .ent __call_stub_fp_df_sf_df
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
; stel: mtc1 $4,$f12
; stel: mtc1 $6,$f14
; stel: mtc1 $7,$f15
; stel: move $18, $31
; stel: jal df_sf_df
; stel: mfc1 $2,$f0
; stel: mfc1 $3,$f1
; stel: jr $18
Fix some misc. issues with Mips16 fp stubs. 1) They should never be inlined. 2) A naming inconsistency with gcc mips16 3) Stubs should not have the global attribute llvm-svn: 187555
2013-08-01 10:26:31 +08:00
; stel: .end __call_stub_fp_df_sf_df
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
declare double @df_df_sf(double, float) #1
; stel: .section .mips16.call.fp.df_df_sf,"ax",@progbits
Fix some misc. issues with Mips16 fp stubs. 1) They should never be inlined. 2) A naming inconsistency with gcc mips16 3) Stubs should not have the global attribute llvm-svn: 187555
2013-08-01 10:26:31 +08:00
; stel: .ent __call_stub_fp_df_df_sf
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
; stel: mtc1 $4,$f12
; stel: mtc1 $5,$f13
; stel: mtc1 $6,$f14
; stel: move $18, $31
; stel: jal df_df_sf
; stel: mfc1 $2,$f0
; stel: mfc1 $3,$f1
; stel: jr $18
Fix some misc. issues with Mips16 fp stubs. 1) They should never be inlined. 2) A naming inconsistency with gcc mips16 3) Stubs should not have the global attribute llvm-svn: 187555
2013-08-01 10:26:31 +08:00
; stel: .end __call_stub_fp_df_df_sf
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
declare double @df_df_df(double, double) #1
; stel: .section .mips16.call.fp.df_df_df,"ax",@progbits
Fix some misc. issues with Mips16 fp stubs. 1) They should never be inlined. 2) A naming inconsistency with gcc mips16 3) Stubs should not have the global attribute llvm-svn: 187555
2013-08-01 10:26:31 +08:00
; stel: .ent __call_stub_fp_df_df_df
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
; stel: mtc1 $4,$f12
; stel: mtc1 $5,$f13
; stel: mtc1 $6,$f14
; stel: mtc1 $7,$f15
; stel: move $18, $31
; stel: jal df_df_df
; stel: mfc1 $2,$f0
; stel: mfc1 $3,$f1
; stel: jr $18
Fix some misc. issues with Mips16 fp stubs. 1) They should never be inlined. 2) A naming inconsistency with gcc mips16 3) Stubs should not have the global attribute llvm-svn: 187555
2013-08-01 10:26:31 +08:00
; stel: .end __call_stub_fp_df_df_df
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
declare { float, float } @sc_v() #1
; stel: .section .mips16.call.fp.sc_v,"ax",@progbits
Fix some misc. issues with Mips16 fp stubs. 1) They should never be inlined. 2) A naming inconsistency with gcc mips16 3) Stubs should not have the global attribute llvm-svn: 187555
2013-08-01 10:26:31 +08:00
; stel: .ent __call_stub_fp_sc_v
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
; stel: move $18, $31
; stel: jal sc_v
; stel: mfc1 $2,$f0
; stel: mfc1 $3,$f2
; stel: jr $18
Fix some misc. issues with Mips16 fp stubs. 1) They should never be inlined. 2) A naming inconsistency with gcc mips16 3) Stubs should not have the global attribute llvm-svn: 187555
2013-08-01 10:26:31 +08:00
; stel: .end __call_stub_fp_sc_v
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
declare { float, float } @sc_sf(float) #1
; stel: .section .mips16.call.fp.sc_sf,"ax",@progbits
Fix some misc. issues with Mips16 fp stubs. 1) They should never be inlined. 2) A naming inconsistency with gcc mips16 3) Stubs should not have the global attribute llvm-svn: 187555
2013-08-01 10:26:31 +08:00
; stel: .ent __call_stub_fp_sc_sf
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
; stel: mtc1 $4,$f12
; stel: move $18, $31
; stel: jal sc_sf
; stel: mfc1 $2,$f0
; stel: mfc1 $3,$f2
; stel: jr $18
Fix some misc. issues with Mips16 fp stubs. 1) They should never be inlined. 2) A naming inconsistency with gcc mips16 3) Stubs should not have the global attribute llvm-svn: 187555
2013-08-01 10:26:31 +08:00
; stel: .end __call_stub_fp_sc_sf
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
declare { double, double } @dc_v() #1
; stel: .section .mips16.call.fp.dc_v,"ax",@progbits
Fix some misc. issues with Mips16 fp stubs. 1) They should never be inlined. 2) A naming inconsistency with gcc mips16 3) Stubs should not have the global attribute llvm-svn: 187555
2013-08-01 10:26:31 +08:00
; stel: .ent __call_stub_fp_dc_v
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
; stel: move $18, $31
; stel: jal dc_v
; stel: mfc1 $4,$f2
; stel: mfc1 $5,$f3
; stel: mfc1 $2,$f0
; stel: mfc1 $3,$f1
; stel: jr $18
Fix some misc. issues with Mips16 fp stubs. 1) They should never be inlined. 2) A naming inconsistency with gcc mips16 3) Stubs should not have the global attribute llvm-svn: 187555
2013-08-01 10:26:31 +08:00
; stel: .end __call_stub_fp_dc_v
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
declare { double, double } @dc_sf(float) #1
; stel: .section .mips16.call.fp.dc_sf,"ax",@progbits
Fix some misc. issues with Mips16 fp stubs. 1) They should never be inlined. 2) A naming inconsistency with gcc mips16 3) Stubs should not have the global attribute llvm-svn: 187555
2013-08-01 10:26:31 +08:00
; stel: .ent __call_stub_fp_dc_sf
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
; stel: mtc1 $4,$f12
; stel: move $18, $31
; stel: jal dc_sf
; stel: mfc1 $4,$f2
; stel: mfc1 $5,$f3
; stel: mfc1 $2,$f0
; stel: mfc1 $3,$f1
; stel: jr $18
Fix some misc. issues with Mips16 fp stubs. 1) They should never be inlined. 2) A naming inconsistency with gcc mips16 3) Stubs should not have the global attribute llvm-svn: 187555
2013-08-01 10:26:31 +08:00
; stel: .end __call_stub_fp_dc_sf
This is the first of three patches which creates stubs used for Mips16/32 floating point interoperability. When Mips16 code calls external functions that would normally have some of its parameters or return values passed in floating point registers, it needs (Mips32) helper functions to do this because while in Mips16 mode there is no ability to access the floating point registers. In Pic mode, this is done with a set of predefined functions in libc. This case is already handled in llvm for Mips16. In static relocation mode, for efficiency reasons, the compiler generates stubs that the linker will use if it turns out that the external function is a Mips32 function. (If it's Mips16, then it does not need the helper stubs). These stubs are identically named and the linker knows about these tricks and will not create multiple copies and will delete them if they are not needed. llvm-svn: 181753
2013-05-14 10:00:24 +08:00
Update to remove the no-frame-pointer-elim-non-leaf flag if it was set to 'false'. llvm-svn: 189068
2013-08-23 05:28:54 +08:00
attributes #0 = { nounwind "less-precise-fpmad"="false" "no-frame-pointer-elim"="true" "no-frame-pointer-elim-non-leaf" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "unsafe-fp-math"="false" "use-soft-float"="false" }
attributes #1 = { "less-precise-fpmad"="false" "no-frame-pointer-elim"="true" "no-frame-pointer-elim-non-leaf" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "unsafe-fp-math"="false" "use-soft-float"="false" }