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/cmov.ll

766 lines
27 KiB
LLVM
Raw Normal View History

[mips][mips64r6] c.cond.fmt, mov[fntz], and mov[fntz].[ds] are not available on MIPS32r6/MIPS64r6 Summary: c.cond.fmt has been replaced by cmp.cond.fmt. Where c.cond.fmt wrote to dedicated condition registers, cmp.cond.fmt writes 1 or 0 to normal FGR's (like the GPR comparisons). mov[fntz] have been replaced by seleqz and selnez. These instructions conditionally zero a register based on a bool in a GPR. The results can then be or'd together to act as a select without, for example, requiring a third register read port. mov[fntz].[ds] have been replaced with sel.[ds] MIPS64r6 currently generates unnecessary sign-extensions for most selects. This is because the result of a SETCC is currently an i32. Bits 32-63 are undefined in i32 and the behaviour of seleqz/selnez would otherwise depend on undefined bits. Later, we will fix this by making the result of SETCC an i64 on MIPS64 targets. Depends on D3958 Reviewers: jkolek, vmedic, zoran.jovanovic Reviewed By: vmedic, zoran.jovanovic Differential Revision: http://reviews.llvm.org/D4003 llvm-svn: 210777
2014-06-12 21:39:06 +08:00
; RUN: llc -march=mips -mcpu=mips32 < %s | FileCheck %s -check-prefix=ALL -check-prefix=32-CMOV
; RUN: llc -march=mips -mcpu=mips32 -regalloc=basic < %s | FileCheck %s -check-prefix=ALL -check-prefix=32-CMOV
; RUN: llc -march=mips -mcpu=mips32r2 < %s | FileCheck %s -check-prefix=ALL -check-prefix=32-CMOV
; RUN: llc -march=mips -mcpu=mips32r6 < %s | FileCheck %s -check-prefix=ALL -check-prefix=32-CMP
; RUN: llc -march=mips64el -mcpu=mips4 < %s | FileCheck %s -check-prefix=ALL -check-prefix=64-CMOV
; RUN: llc -march=mips64el -mcpu=mips64 < %s | FileCheck %s -check-prefix=ALL -check-prefix=64-CMOV
; RUN: llc -march=mips64el -mcpu=mips64r6 < %s | FileCheck %s -check-prefix=ALL -check-prefix=64-CMP
Match a pattern generated by a dag combiner opt where: (select (load (load tga0)) (load tga1)) => (load (select (load tga0) tga1)) Thanks to Akira for pointing that. llvm-svn: 121163
2010-12-08 03:00:20 +08:00
@i1 = global [3 x i32] [i32 1, i32 2, i32 3], align 4
@i3 = common global i32* null, align 4
[mips][mips64r6] c.cond.fmt, mov[fntz], and mov[fntz].[ds] are not available on MIPS32r6/MIPS64r6 Summary: c.cond.fmt has been replaced by cmp.cond.fmt. Where c.cond.fmt wrote to dedicated condition registers, cmp.cond.fmt writes 1 or 0 to normal FGR's (like the GPR comparisons). mov[fntz] have been replaced by seleqz and selnez. These instructions conditionally zero a register based on a bool in a GPR. The results can then be or'd together to act as a select without, for example, requiring a third register read port. mov[fntz].[ds] have been replaced with sel.[ds] MIPS64r6 currently generates unnecessary sign-extensions for most selects. This is because the result of a SETCC is currently an i32. Bits 32-63 are undefined in i32 and the behaviour of seleqz/selnez would otherwise depend on undefined bits. Later, we will fix this by making the result of SETCC an i64 on MIPS64 targets. Depends on D3958 Reviewers: jkolek, vmedic, zoran.jovanovic Reviewed By: vmedic, zoran.jovanovic Differential Revision: http://reviews.llvm.org/D4003 llvm-svn: 210777
2014-06-12 21:39:06 +08:00
; ALL-LABEL: cmov1:
; 32-CMOV-DAG: lw $[[R0:[0-9]+]], %got(i3)
; 32-CMOV-DAG: addiu $[[R1:[0-9]+]], ${{[0-9]+}}, %got(i1)
; 32-CMOV-DAG: movn $[[R0]], $[[R1]], $4
; 32-CMOV-DAG: lw $2, 0($[[R0]])
; 32-CMP-DAG: lw $[[R0:[0-9]+]], %got(i3)
; 32-CMP-DAG: addiu $[[R1:[0-9]+]], ${{[0-9]+}}, %got(i1)
[mips][mips64r6] Correct select patterns that have the condition or true/false values backwards Summary: This bug caused SingleSource/Regression/C/uint64_to_float and SingleSource/UnitTests/2002-05-02-CastTest3 to fail (among others). Differential Revision: http://reviews.llvm.org/D4388 llvm-svn: 212608
2014-07-09 18:47:26 +08:00
; 32-CMP-DAG: selnez $[[T0:[0-9]+]], $[[R1]], $4
; 32-CMP-DAG: seleqz $[[T1:[0-9]+]], $[[R0]], $4
; 32-CMP-DAG: or $[[T2:[0-9]+]], $[[T0]], $[[T1]]
[mips][mips64r6] c.cond.fmt, mov[fntz], and mov[fntz].[ds] are not available on MIPS32r6/MIPS64r6 Summary: c.cond.fmt has been replaced by cmp.cond.fmt. Where c.cond.fmt wrote to dedicated condition registers, cmp.cond.fmt writes 1 or 0 to normal FGR's (like the GPR comparisons). mov[fntz] have been replaced by seleqz and selnez. These instructions conditionally zero a register based on a bool in a GPR. The results can then be or'd together to act as a select without, for example, requiring a third register read port. mov[fntz].[ds] have been replaced with sel.[ds] MIPS64r6 currently generates unnecessary sign-extensions for most selects. This is because the result of a SETCC is currently an i32. Bits 32-63 are undefined in i32 and the behaviour of seleqz/selnez would otherwise depend on undefined bits. Later, we will fix this by making the result of SETCC an i64 on MIPS64 targets. Depends on D3958 Reviewers: jkolek, vmedic, zoran.jovanovic Reviewed By: vmedic, zoran.jovanovic Differential Revision: http://reviews.llvm.org/D4003 llvm-svn: 210777
2014-06-12 21:39:06 +08:00
; 32-CMP-DAG: lw $2, 0($[[T2]])
; 64-CMOV-DAG: ldr $[[R0:[0-9]+]]
; 64-CMOV-DAG: ld $[[R1:[0-9]+]], %got_disp(i1)
; 64-CMOV-DAG: movn $[[R0]], $[[R1]], $4
; 64-CMP-DAG: ld $[[R0:[0-9]+]], %got_disp(i3)(
; 64-CMP-DAG: daddiu $[[R1:[0-9]+]], ${{[0-9]+}}, %got_disp(i1)
; FIXME: This sll works around an implementation detail in the code generator
; (setcc's result is i32 so bits 32-63 are undefined). It's not really
; needed.
; 64-CMP-DAG: sll $[[CC:[0-9]+]], $4, 0
[mips][mips64r6] Correct select patterns that have the condition or true/false values backwards Summary: This bug caused SingleSource/Regression/C/uint64_to_float and SingleSource/UnitTests/2002-05-02-CastTest3 to fail (among others). Differential Revision: http://reviews.llvm.org/D4388 llvm-svn: 212608
2014-07-09 18:47:26 +08:00
; 64-CMP-DAG: selnez $[[T0:[0-9]+]], $[[R1]], $[[CC]]
; 64-CMP-DAG: seleqz $[[T1:[0-9]+]], $[[R0]], $[[CC]]
; 64-CMP-DAG: or $[[T2:[0-9]+]], $[[T0]], $[[T1]]
[mips][mips64r6] c.cond.fmt, mov[fntz], and mov[fntz].[ds] are not available on MIPS32r6/MIPS64r6 Summary: c.cond.fmt has been replaced by cmp.cond.fmt. Where c.cond.fmt wrote to dedicated condition registers, cmp.cond.fmt writes 1 or 0 to normal FGR's (like the GPR comparisons). mov[fntz] have been replaced by seleqz and selnez. These instructions conditionally zero a register based on a bool in a GPR. The results can then be or'd together to act as a select without, for example, requiring a third register read port. mov[fntz].[ds] have been replaced with sel.[ds] MIPS64r6 currently generates unnecessary sign-extensions for most selects. This is because the result of a SETCC is currently an i32. Bits 32-63 are undefined in i32 and the behaviour of seleqz/selnez would otherwise depend on undefined bits. Later, we will fix this by making the result of SETCC an i64 on MIPS64 targets. Depends on D3958 Reviewers: jkolek, vmedic, zoran.jovanovic Reviewed By: vmedic, zoran.jovanovic Differential Revision: http://reviews.llvm.org/D4003 llvm-svn: 210777
2014-06-12 21:39:06 +08:00
; 64-CMP-DAG: ld $2, 0($[[T2]])
[mips] Promote i32 arguments to i64 for the N32/N64 ABI and fix <64-bit structs... Summary: ... and after all that refactoring, it's possible to distinguish softfloat floating point values from integers so this patch no longer breaks softfloat to do it. Remove direct handling of i32's in the N32/N64 ABI by promoting them to i64. This more closely reflects the ABI documentation and also fixes problems with stack arguments on big-endian targets. We now rely on signext/zeroext annotations (already generated by clang) and the Assert[SZ]ext nodes to avoid the introduction of unnecessary sign/zero extends. It was not possible to convert three tests to use signext/zeroext. These tests are bswap.ll, ctlz-v.ll, ctlz-v.ll. It's not possible to put signext on a vector type so we just accept the sign extends here for now. These tests don't pass the vectors the same way clang does (clang puts multiple elements in the same argument, these map 1 element to 1 argument) so we don't need to worry too much about it. With this patch, all known N32/N64 bugs should be fixed and we now pass the first 10,000 tests generated by ABITest.py. Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D6117 llvm-svn: 221534
2014-11-08 00:54:21 +08:00
define i32* @cmov1(i32 signext %s) nounwind readonly {
Match a pattern generated by a dag combiner opt where: (select (load (load tga0)) (load tga1)) => (load (select (load tga0) tga1)) Thanks to Akira for pointing that. llvm-svn: 121163
2010-12-08 03:00:20 +08:00
entry:
%tobool = icmp ne i32 %s, 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
%tmp1 = load i32*, i32** @i3, 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
%cond = select i1 %tobool, i32* getelementptr inbounds ([3 x i32], [3 x i32]* @i1, i32 0, i32 0), i32* %tmp1
Match a pattern generated by a dag combiner opt where: (select (load (load tga0)) (load tga1)) => (load (select (load tga0) tga1)) Thanks to Akira for pointing that. llvm-svn: 121163
2010-12-08 03:00:20 +08:00
ret i32* %cond
}
Define a wrapper node for target constant nodes (tglobaladdr, etc.). Need this to prevent emitting illegal conditional move instructions. llvm-svn: 132240
2011-05-28 09:07:07 +08:00
@c = global i32 1, align 4
@d = global i32 0, align 4
[mips][mips64r6] c.cond.fmt, mov[fntz], and mov[fntz].[ds] are not available on MIPS32r6/MIPS64r6 Summary: c.cond.fmt has been replaced by cmp.cond.fmt. Where c.cond.fmt wrote to dedicated condition registers, cmp.cond.fmt writes 1 or 0 to normal FGR's (like the GPR comparisons). mov[fntz] have been replaced by seleqz and selnez. These instructions conditionally zero a register based on a bool in a GPR. The results can then be or'd together to act as a select without, for example, requiring a third register read port. mov[fntz].[ds] have been replaced with sel.[ds] MIPS64r6 currently generates unnecessary sign-extensions for most selects. This is because the result of a SETCC is currently an i32. Bits 32-63 are undefined in i32 and the behaviour of seleqz/selnez would otherwise depend on undefined bits. Later, we will fix this by making the result of SETCC an i64 on MIPS64 targets. Depends on D3958 Reviewers: jkolek, vmedic, zoran.jovanovic Reviewed By: vmedic, zoran.jovanovic Differential Revision: http://reviews.llvm.org/D4003 llvm-svn: 210777
2014-06-12 21:39:06 +08:00
; ALL-LABEL: cmov2:
; 32-CMOV-DAG: addiu $[[R1:[0-9]+]], ${{[0-9]+}}, %got(d)
; 32-CMOV-DAG: addiu $[[R0:[0-9]+]], ${{[0-9]+}}, %got(c)
; 32-CMOV-DAG: movn $[[R1]], $[[R0]], $4
; 32-CMOV-DAG: lw $2, 0($[[R0]])
; 32-CMP-DAG: addiu $[[R1:[0-9]+]], ${{[0-9]+}}, %got(d)
; 32-CMP-DAG: addiu $[[R0:[0-9]+]], ${{[0-9]+}}, %got(c)
[mips][mips64r6] Correct select patterns that have the condition or true/false values backwards Summary: This bug caused SingleSource/Regression/C/uint64_to_float and SingleSource/UnitTests/2002-05-02-CastTest3 to fail (among others). Differential Revision: http://reviews.llvm.org/D4388 llvm-svn: 212608
2014-07-09 18:47:26 +08:00
; 32-CMP-DAG: selnez $[[T0:[0-9]+]], $[[R0]], $4
; 32-CMP-DAG: seleqz $[[T1:[0-9]+]], $[[R1]], $4
; 32-CMP-DAG: or $[[T2:[0-9]+]], $[[T0]], $[[T1]]
[mips][mips64r6] c.cond.fmt, mov[fntz], and mov[fntz].[ds] are not available on MIPS32r6/MIPS64r6 Summary: c.cond.fmt has been replaced by cmp.cond.fmt. Where c.cond.fmt wrote to dedicated condition registers, cmp.cond.fmt writes 1 or 0 to normal FGR's (like the GPR comparisons). mov[fntz] have been replaced by seleqz and selnez. These instructions conditionally zero a register based on a bool in a GPR. The results can then be or'd together to act as a select without, for example, requiring a third register read port. mov[fntz].[ds] have been replaced with sel.[ds] MIPS64r6 currently generates unnecessary sign-extensions for most selects. This is because the result of a SETCC is currently an i32. Bits 32-63 are undefined in i32 and the behaviour of seleqz/selnez would otherwise depend on undefined bits. Later, we will fix this by making the result of SETCC an i64 on MIPS64 targets. Depends on D3958 Reviewers: jkolek, vmedic, zoran.jovanovic Reviewed By: vmedic, zoran.jovanovic Differential Revision: http://reviews.llvm.org/D4003 llvm-svn: 210777
2014-06-12 21:39:06 +08:00
; 32-CMP-DAG: lw $2, 0($[[T2]])
; 64-CMOV: daddiu $[[R1:[0-9]+]], ${{[0-9]+}}, %got_disp(d)
; 64-CMOV: daddiu $[[R0:[0-9]+]], ${{[0-9]+}}, %got_disp(c)
; 64-CMOV: movn $[[R1]], $[[R0]], $4
; 64-CMP-DAG: daddiu $[[R1:[0-9]+]], ${{[0-9]+}}, %got_disp(d)
; 64-CMP-DAG: daddiu $[[R0:[0-9]+]], ${{[0-9]+}}, %got_disp(c)
; FIXME: This sll works around an implementation detail in the code generator
; (setcc's result is i32 so bits 32-63 are undefined). It's not really
; needed.
; 64-CMP-DAG: sll $[[CC:[0-9]+]], $4, 0
[mips][mips64r6] Correct select patterns that have the condition or true/false values backwards Summary: This bug caused SingleSource/Regression/C/uint64_to_float and SingleSource/UnitTests/2002-05-02-CastTest3 to fail (among others). Differential Revision: http://reviews.llvm.org/D4388 llvm-svn: 212608
2014-07-09 18:47:26 +08:00
; 64-CMP-DAG: selnez $[[T0:[0-9]+]], $[[R0]], $[[CC]]
; 64-CMP-DAG: seleqz $[[T1:[0-9]+]], $[[R1]], $[[CC]]
; 64-CMP-DAG: or $[[T2:[0-9]+]], $[[T0]], $[[T1]]
[mips][mips64r6] c.cond.fmt, mov[fntz], and mov[fntz].[ds] are not available on MIPS32r6/MIPS64r6 Summary: c.cond.fmt has been replaced by cmp.cond.fmt. Where c.cond.fmt wrote to dedicated condition registers, cmp.cond.fmt writes 1 or 0 to normal FGR's (like the GPR comparisons). mov[fntz] have been replaced by seleqz and selnez. These instructions conditionally zero a register based on a bool in a GPR. The results can then be or'd together to act as a select without, for example, requiring a third register read port. mov[fntz].[ds] have been replaced with sel.[ds] MIPS64r6 currently generates unnecessary sign-extensions for most selects. This is because the result of a SETCC is currently an i32. Bits 32-63 are undefined in i32 and the behaviour of seleqz/selnez would otherwise depend on undefined bits. Later, we will fix this by making the result of SETCC an i64 on MIPS64 targets. Depends on D3958 Reviewers: jkolek, vmedic, zoran.jovanovic Reviewed By: vmedic, zoran.jovanovic Differential Revision: http://reviews.llvm.org/D4003 llvm-svn: 210777
2014-06-12 21:39:06 +08:00
; 64-CMP-DAG: lw $2, 0($[[T2]])
[mips] Promote i32 arguments to i64 for the N32/N64 ABI and fix <64-bit structs... Summary: ... and after all that refactoring, it's possible to distinguish softfloat floating point values from integers so this patch no longer breaks softfloat to do it. Remove direct handling of i32's in the N32/N64 ABI by promoting them to i64. This more closely reflects the ABI documentation and also fixes problems with stack arguments on big-endian targets. We now rely on signext/zeroext annotations (already generated by clang) and the Assert[SZ]ext nodes to avoid the introduction of unnecessary sign/zero extends. It was not possible to convert three tests to use signext/zeroext. These tests are bswap.ll, ctlz-v.ll, ctlz-v.ll. It's not possible to put signext on a vector type so we just accept the sign extends here for now. These tests don't pass the vectors the same way clang does (clang puts multiple elements in the same argument, these map 1 element to 1 argument) so we don't need to worry too much about it. With this patch, all known N32/N64 bugs should be fixed and we now pass the first 10,000 tests generated by ABITest.py. Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D6117 llvm-svn: 221534
2014-11-08 00:54:21 +08:00
define i32 @cmov2(i32 signext %s) nounwind readonly {
Define a wrapper node for target constant nodes (tglobaladdr, etc.). Need this to prevent emitting illegal conditional move instructions. llvm-svn: 132240
2011-05-28 09:07:07 +08:00
entry:
%tobool = icmp ne i32 %s, 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
%tmp1 = load i32, i32* @c, align 4
%tmp2 = load i32, i32* @d, align 4
Define a wrapper node for target constant nodes (tglobaladdr, etc.). Need this to prevent emitting illegal conditional move instructions. llvm-svn: 132240
2011-05-28 09:07:07 +08:00
%cond = select i1 %tobool, i32 %tmp1, i32 %tmp2
ret i32 %cond
}
[mips][mips64r6] c.cond.fmt, mov[fntz], and mov[fntz].[ds] are not available on MIPS32r6/MIPS64r6 Summary: c.cond.fmt has been replaced by cmp.cond.fmt. Where c.cond.fmt wrote to dedicated condition registers, cmp.cond.fmt writes 1 or 0 to normal FGR's (like the GPR comparisons). mov[fntz] have been replaced by seleqz and selnez. These instructions conditionally zero a register based on a bool in a GPR. The results can then be or'd together to act as a select without, for example, requiring a third register read port. mov[fntz].[ds] have been replaced with sel.[ds] MIPS64r6 currently generates unnecessary sign-extensions for most selects. This is because the result of a SETCC is currently an i32. Bits 32-63 are undefined in i32 and the behaviour of seleqz/selnez would otherwise depend on undefined bits. Later, we will fix this by making the result of SETCC an i64 on MIPS64 targets. Depends on D3958 Reviewers: jkolek, vmedic, zoran.jovanovic Reviewed By: vmedic, zoran.jovanovic Differential Revision: http://reviews.llvm.org/D4003 llvm-svn: 210777
2014-06-12 21:39:06 +08:00
; ALL-LABEL: cmov3:
; We won't check the result register since we can't know if the move is first
; or last. We do know it will be either one of two registers so we can at least
; check that.
; 32-CMOV: xori $[[R0:[0-9]+]], $4, 234
; 32-CMOV: movz ${{[26]}}, $5, $[[R0]]
; 32-CMP-DAG: xori $[[CC:[0-9]+]], $4, 234
[mips][mips64r6] Correct select patterns that have the condition or true/false values backwards Summary: This bug caused SingleSource/Regression/C/uint64_to_float and SingleSource/UnitTests/2002-05-02-CastTest3 to fail (among others). Differential Revision: http://reviews.llvm.org/D4388 llvm-svn: 212608
2014-07-09 18:47:26 +08:00
; 32-CMP-DAG: seleqz $[[T0:[0-9]+]], $5, $[[CC]]
; 32-CMP-DAG: selnez $[[T1:[0-9]+]], $6, $[[CC]]
[mips][mips64r6] c.cond.fmt, mov[fntz], and mov[fntz].[ds] are not available on MIPS32r6/MIPS64r6 Summary: c.cond.fmt has been replaced by cmp.cond.fmt. Where c.cond.fmt wrote to dedicated condition registers, cmp.cond.fmt writes 1 or 0 to normal FGR's (like the GPR comparisons). mov[fntz] have been replaced by seleqz and selnez. These instructions conditionally zero a register based on a bool in a GPR. The results can then be or'd together to act as a select without, for example, requiring a third register read port. mov[fntz].[ds] have been replaced with sel.[ds] MIPS64r6 currently generates unnecessary sign-extensions for most selects. This is because the result of a SETCC is currently an i32. Bits 32-63 are undefined in i32 and the behaviour of seleqz/selnez would otherwise depend on undefined bits. Later, we will fix this by making the result of SETCC an i64 on MIPS64 targets. Depends on D3958 Reviewers: jkolek, vmedic, zoran.jovanovic Reviewed By: vmedic, zoran.jovanovic Differential Revision: http://reviews.llvm.org/D4003 llvm-svn: 210777
2014-06-12 21:39:06 +08:00
; 32-CMP-DAG: or $2, $[[T0]], $[[T1]]
; 64-CMOV: xori $[[R0:[0-9]+]], $4, 234
; 64-CMOV: movz ${{[26]}}, $5, $[[R0]]
; 64-CMP-DAG: xori $[[CC:[0-9]+]], $4, 234
[mips][mips64r6] Correct select patterns that have the condition or true/false values backwards Summary: This bug caused SingleSource/Regression/C/uint64_to_float and SingleSource/UnitTests/2002-05-02-CastTest3 to fail (among others). Differential Revision: http://reviews.llvm.org/D4388 llvm-svn: 212608
2014-07-09 18:47:26 +08:00
; 64-CMP-DAG: seleqz $[[T0:[0-9]+]], $5, $[[CC]]
; 64-CMP-DAG: selnez $[[T1:[0-9]+]], $6, $[[CC]]
[mips][mips64r6] c.cond.fmt, mov[fntz], and mov[fntz].[ds] are not available on MIPS32r6/MIPS64r6 Summary: c.cond.fmt has been replaced by cmp.cond.fmt. Where c.cond.fmt wrote to dedicated condition registers, cmp.cond.fmt writes 1 or 0 to normal FGR's (like the GPR comparisons). mov[fntz] have been replaced by seleqz and selnez. These instructions conditionally zero a register based on a bool in a GPR. The results can then be or'd together to act as a select without, for example, requiring a third register read port. mov[fntz].[ds] have been replaced with sel.[ds] MIPS64r6 currently generates unnecessary sign-extensions for most selects. This is because the result of a SETCC is currently an i32. Bits 32-63 are undefined in i32 and the behaviour of seleqz/selnez would otherwise depend on undefined bits. Later, we will fix this by making the result of SETCC an i64 on MIPS64 targets. Depends on D3958 Reviewers: jkolek, vmedic, zoran.jovanovic Reviewed By: vmedic, zoran.jovanovic Differential Revision: http://reviews.llvm.org/D4003 llvm-svn: 210777
2014-06-12 21:39:06 +08:00
; 64-CMP-DAG: or $2, $[[T0]], $[[T1]]
[mips] Promote i32 arguments to i64 for the N32/N64 ABI and fix <64-bit structs... Summary: ... and after all that refactoring, it's possible to distinguish softfloat floating point values from integers so this patch no longer breaks softfloat to do it. Remove direct handling of i32's in the N32/N64 ABI by promoting them to i64. This more closely reflects the ABI documentation and also fixes problems with stack arguments on big-endian targets. We now rely on signext/zeroext annotations (already generated by clang) and the Assert[SZ]ext nodes to avoid the introduction of unnecessary sign/zero extends. It was not possible to convert three tests to use signext/zeroext. These tests are bswap.ll, ctlz-v.ll, ctlz-v.ll. It's not possible to put signext on a vector type so we just accept the sign extends here for now. These tests don't pass the vectors the same way clang does (clang puts multiple elements in the same argument, these map 1 element to 1 argument) so we don't need to worry too much about it. With this patch, all known N32/N64 bugs should be fixed and we now pass the first 10,000 tests generated by ABITest.py. Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D6117 llvm-svn: 221534
2014-11-08 00:54:21 +08:00
define i32 @cmov3(i32 signext %a, i32 signext %b, i32 signext %c) nounwind readnone {
Add another peephole pattern for conditional moves. llvm-svn: 156460
2012-05-09 10:29:29 +08:00
entry:
%cmp = icmp eq i32 %a, 234
%cond = select i1 %cmp, i32 %b, i32 %c
ret i32 %cond
}
[mips] Fix two patterns that select i32's (for MIPS32r6) / i64's (for MIPS64r6) from setne comparison with an i32. The patterns that are fixed: * (select (i32 (setne i32, immZExt16)), i32, i32) (for MIPS32r6) * (select (i32 (setne i32, immZExt16)), i64, i64) (for MIPS64r6) llvm-svn: 213653
2014-07-22 21:36:02 +08:00
; ALL-LABEL: cmov3_ne:
; We won't check the result register since we can't know if the move is first
; or last. We do know it will be either one of two registers so we can at least
; check that.
; FIXME: Use xori instead of addiu+xor.
; 32-CMOV: addiu $[[R0:[0-9]+]], $zero, 234
; 32-CMOV: xor $[[R1:[0-9]+]], $4, $[[R0]]
; 32-CMOV: movn ${{[26]}}, $5, $[[R1]]
; 32-CMP-DAG: xori $[[CC:[0-9]+]], $4, 234
; 32-CMP-DAG: selnez $[[T0:[0-9]+]], $5, $[[CC]]
; 32-CMP-DAG: seleqz $[[T1:[0-9]+]], $6, $[[CC]]
; 32-CMP-DAG: or $2, $[[T0]], $[[T1]]
; FIXME: Use xori instead of addiu+xor.
; 64-CMOV: addiu $[[R0:[0-9]+]], $zero, 234
; 64-CMOV: xor $[[R1:[0-9]+]], $4, $[[R0]]
; 64-CMOV: movn ${{[26]}}, $5, $[[R1]]
; 64-CMP-DAG: xori $[[CC:[0-9]+]], $4, 234
; 64-CMP-DAG: selnez $[[T0:[0-9]+]], $5, $[[CC]]
; 64-CMP-DAG: seleqz $[[T1:[0-9]+]], $6, $[[CC]]
; 64-CMP-DAG: or $2, $[[T0]], $[[T1]]
[mips] Promote i32 arguments to i64 for the N32/N64 ABI and fix <64-bit structs... Summary: ... and after all that refactoring, it's possible to distinguish softfloat floating point values from integers so this patch no longer breaks softfloat to do it. Remove direct handling of i32's in the N32/N64 ABI by promoting them to i64. This more closely reflects the ABI documentation and also fixes problems with stack arguments on big-endian targets. We now rely on signext/zeroext annotations (already generated by clang) and the Assert[SZ]ext nodes to avoid the introduction of unnecessary sign/zero extends. It was not possible to convert three tests to use signext/zeroext. These tests are bswap.ll, ctlz-v.ll, ctlz-v.ll. It's not possible to put signext on a vector type so we just accept the sign extends here for now. These tests don't pass the vectors the same way clang does (clang puts multiple elements in the same argument, these map 1 element to 1 argument) so we don't need to worry too much about it. With this patch, all known N32/N64 bugs should be fixed and we now pass the first 10,000 tests generated by ABITest.py. Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D6117 llvm-svn: 221534
2014-11-08 00:54:21 +08:00
define i32 @cmov3_ne(i32 signext %a, i32 signext %b, i32 signext %c) nounwind readnone {
[mips] Fix two patterns that select i32's (for MIPS32r6) / i64's (for MIPS64r6) from setne comparison with an i32. The patterns that are fixed: * (select (i32 (setne i32, immZExt16)), i32, i32) (for MIPS32r6) * (select (i32 (setne i32, immZExt16)), i64, i64) (for MIPS64r6) llvm-svn: 213653
2014-07-22 21:36:02 +08:00
entry:
%cmp = icmp ne i32 %a, 234
%cond = select i1 %cmp, i32 %b, i32 %c
ret i32 %cond
}
[mips][mips64r6] c.cond.fmt, mov[fntz], and mov[fntz].[ds] are not available on MIPS32r6/MIPS64r6 Summary: c.cond.fmt has been replaced by cmp.cond.fmt. Where c.cond.fmt wrote to dedicated condition registers, cmp.cond.fmt writes 1 or 0 to normal FGR's (like the GPR comparisons). mov[fntz] have been replaced by seleqz and selnez. These instructions conditionally zero a register based on a bool in a GPR. The results can then be or'd together to act as a select without, for example, requiring a third register read port. mov[fntz].[ds] have been replaced with sel.[ds] MIPS64r6 currently generates unnecessary sign-extensions for most selects. This is because the result of a SETCC is currently an i32. Bits 32-63 are undefined in i32 and the behaviour of seleqz/selnez would otherwise depend on undefined bits. Later, we will fix this by making the result of SETCC an i64 on MIPS64 targets. Depends on D3958 Reviewers: jkolek, vmedic, zoran.jovanovic Reviewed By: vmedic, zoran.jovanovic Differential Revision: http://reviews.llvm.org/D4003 llvm-svn: 210777
2014-06-12 21:39:06 +08:00
; ALL-LABEL: cmov4:
; We won't check the result register since we can't know if the move is first
; or last. We do know it will be one of two registers so we can at least check
; that.
; 32-CMOV-DAG: xori $[[R0:[0-9]+]], $4, 234
; 32-CMOV-DAG: lw $[[R1:2]], 16($sp)
; 32-CMOV-DAG: lw $[[R2:3]], 20($sp)
; 32-CMOV-DAG: movz $[[R1]], $6, $[[R0]]
; 32-CMOV-DAG: movz $[[R2]], $7, $[[R0]]
; 32-CMP-DAG: xori $[[R0:[0-9]+]], $4, 234
; 32-CMP-DAG: lw $[[R1:[0-9]+]], 16($sp)
; 32-CMP-DAG: lw $[[R2:[0-9]+]], 20($sp)
[mips][mips64r6] Correct select patterns that have the condition or true/false values backwards Summary: This bug caused SingleSource/Regression/C/uint64_to_float and SingleSource/UnitTests/2002-05-02-CastTest3 to fail (among others). Differential Revision: http://reviews.llvm.org/D4388 llvm-svn: 212608
2014-07-09 18:47:26 +08:00
; 32-CMP-DAG: seleqz $[[T0:[0-9]+]], $6, $[[R0]]
; 32-CMP-DAG: seleqz $[[T1:[0-9]+]], $7, $[[R0]]
; 32-CMP-DAG: selnez $[[T2:[0-9]+]], $[[R1]], $[[R0]]
; 32-CMP-DAG: selnez $[[T3:[0-9]+]], $[[R2]], $[[R0]]
[mips][mips64r6] c.cond.fmt, mov[fntz], and mov[fntz].[ds] are not available on MIPS32r6/MIPS64r6 Summary: c.cond.fmt has been replaced by cmp.cond.fmt. Where c.cond.fmt wrote to dedicated condition registers, cmp.cond.fmt writes 1 or 0 to normal FGR's (like the GPR comparisons). mov[fntz] have been replaced by seleqz and selnez. These instructions conditionally zero a register based on a bool in a GPR. The results can then be or'd together to act as a select without, for example, requiring a third register read port. mov[fntz].[ds] have been replaced with sel.[ds] MIPS64r6 currently generates unnecessary sign-extensions for most selects. This is because the result of a SETCC is currently an i32. Bits 32-63 are undefined in i32 and the behaviour of seleqz/selnez would otherwise depend on undefined bits. Later, we will fix this by making the result of SETCC an i64 on MIPS64 targets. Depends on D3958 Reviewers: jkolek, vmedic, zoran.jovanovic Reviewed By: vmedic, zoran.jovanovic Differential Revision: http://reviews.llvm.org/D4003 llvm-svn: 210777
2014-06-12 21:39:06 +08:00
; 32-CMP-DAG: or $2, $[[T0]], $[[T2]]
; 32-CMP-DAG: or $3, $[[T1]], $[[T3]]
; 64-CMOV: xori $[[R0:[0-9]+]], $4, 234
; 64-CMOV: movz ${{[26]}}, $5, $[[R0]]
; 64-CMP-DAG: xori $[[R0:[0-9]+]], $4, 234
[mips][mips64r6] Correct select patterns that have the condition or true/false values backwards Summary: This bug caused SingleSource/Regression/C/uint64_to_float and SingleSource/UnitTests/2002-05-02-CastTest3 to fail (among others). Differential Revision: http://reviews.llvm.org/D4388 llvm-svn: 212608
2014-07-09 18:47:26 +08:00
; 64-CMP-DAG: seleqz $[[T0:[0-9]+]], $5, $[[R0]]
; 64-CMP-DAG: selnez $[[T1:[0-9]+]], $6, $[[R0]]
[mips][mips64r6] c.cond.fmt, mov[fntz], and mov[fntz].[ds] are not available on MIPS32r6/MIPS64r6 Summary: c.cond.fmt has been replaced by cmp.cond.fmt. Where c.cond.fmt wrote to dedicated condition registers, cmp.cond.fmt writes 1 or 0 to normal FGR's (like the GPR comparisons). mov[fntz] have been replaced by seleqz and selnez. These instructions conditionally zero a register based on a bool in a GPR. The results can then be or'd together to act as a select without, for example, requiring a third register read port. mov[fntz].[ds] have been replaced with sel.[ds] MIPS64r6 currently generates unnecessary sign-extensions for most selects. This is because the result of a SETCC is currently an i32. Bits 32-63 are undefined in i32 and the behaviour of seleqz/selnez would otherwise depend on undefined bits. Later, we will fix this by making the result of SETCC an i64 on MIPS64 targets. Depends on D3958 Reviewers: jkolek, vmedic, zoran.jovanovic Reviewed By: vmedic, zoran.jovanovic Differential Revision: http://reviews.llvm.org/D4003 llvm-svn: 210777
2014-06-12 21:39:06 +08:00
; 64-CMP-DAG: or $2, $[[T0]], $[[T1]]
[mips] Promote i32 arguments to i64 for the N32/N64 ABI and fix <64-bit structs... Summary: ... and after all that refactoring, it's possible to distinguish softfloat floating point values from integers so this patch no longer breaks softfloat to do it. Remove direct handling of i32's in the N32/N64 ABI by promoting them to i64. This more closely reflects the ABI documentation and also fixes problems with stack arguments on big-endian targets. We now rely on signext/zeroext annotations (already generated by clang) and the Assert[SZ]ext nodes to avoid the introduction of unnecessary sign/zero extends. It was not possible to convert three tests to use signext/zeroext. These tests are bswap.ll, ctlz-v.ll, ctlz-v.ll. It's not possible to put signext on a vector type so we just accept the sign extends here for now. These tests don't pass the vectors the same way clang does (clang puts multiple elements in the same argument, these map 1 element to 1 argument) so we don't need to worry too much about it. With this patch, all known N32/N64 bugs should be fixed and we now pass the first 10,000 tests generated by ABITest.py. Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D6117 llvm-svn: 221534
2014-11-08 00:54:21 +08:00
define i64 @cmov4(i32 signext %a, i64 %b, i64 %c) nounwind readnone {
Add another peephole pattern for conditional moves. llvm-svn: 156460
2012-05-09 10:29:29 +08:00
entry:
%cmp = icmp eq i32 %a, 234
%cond = select i1 %cmp, i64 %b, i64 %c
ret i64 %cond
}
[mips] Fix two patterns that select i32's (for MIPS32r6) / i64's (for MIPS64r6) from setne comparison with an i32. The patterns that are fixed: * (select (i32 (setne i32, immZExt16)), i32, i32) (for MIPS32r6) * (select (i32 (setne i32, immZExt16)), i64, i64) (for MIPS64r6) llvm-svn: 213653
2014-07-22 21:36:02 +08:00
; ALL-LABEL: cmov4_ne:
; We won't check the result register since we can't know if the move is first
; or last. We do know it will be one of two registers so we can at least check
; that.
; FIXME: Use xori instead of addiu+xor.
; 32-CMOV-DAG: addiu $[[R0:[0-9]+]], $zero, 234
; 32-CMOV-DAG: xor $[[R1:[0-9]+]], $4, $[[R0]]
; 32-CMOV-DAG: lw $[[R2:2]], 16($sp)
; 32-CMOV-DAG: lw $[[R3:3]], 20($sp)
; 32-CMOV-DAG: movn $[[R2]], $6, $[[R1]]
; 32-CMOV-DAG: movn $[[R3]], $7, $[[R1]]
; 32-CMP-DAG: xori $[[R0:[0-9]+]], $4, 234
; 32-CMP-DAG: lw $[[R1:[0-9]+]], 16($sp)
; 32-CMP-DAG: lw $[[R2:[0-9]+]], 20($sp)
; 32-CMP-DAG: selnez $[[T0:[0-9]+]], $6, $[[R0]]
; 32-CMP-DAG: selnez $[[T1:[0-9]+]], $7, $[[R0]]
; 32-CMP-DAG: seleqz $[[T2:[0-9]+]], $[[R1]], $[[R0]]
; 32-CMP-DAG: seleqz $[[T3:[0-9]+]], $[[R2]], $[[R0]]
; 32-CMP-DAG: or $2, $[[T0]], $[[T2]]
; 32-CMP-DAG: or $3, $[[T1]], $[[T3]]
; FIXME: Use xori instead of addiu+xor.
; 64-CMOV: addiu $[[R0:[0-9]+]], $zero, 234
; 64-CMOV: xor $[[R1:[0-9]+]], $4, $[[R0]]
; 64-CMOV: movn ${{[26]}}, $5, $[[R1]]
; 64-CMP-DAG: xori $[[R0:[0-9]+]], $4, 234
; 64-CMP-DAG: selnez $[[T0:[0-9]+]], $5, $[[R0]]
; 64-CMP-DAG: seleqz $[[T1:[0-9]+]], $6, $[[R0]]
; 64-CMP-DAG: or $2, $[[T0]], $[[T1]]
[mips] Promote i32 arguments to i64 for the N32/N64 ABI and fix <64-bit structs... Summary: ... and after all that refactoring, it's possible to distinguish softfloat floating point values from integers so this patch no longer breaks softfloat to do it. Remove direct handling of i32's in the N32/N64 ABI by promoting them to i64. This more closely reflects the ABI documentation and also fixes problems with stack arguments on big-endian targets. We now rely on signext/zeroext annotations (already generated by clang) and the Assert[SZ]ext nodes to avoid the introduction of unnecessary sign/zero extends. It was not possible to convert three tests to use signext/zeroext. These tests are bswap.ll, ctlz-v.ll, ctlz-v.ll. It's not possible to put signext on a vector type so we just accept the sign extends here for now. These tests don't pass the vectors the same way clang does (clang puts multiple elements in the same argument, these map 1 element to 1 argument) so we don't need to worry too much about it. With this patch, all known N32/N64 bugs should be fixed and we now pass the first 10,000 tests generated by ABITest.py. Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D6117 llvm-svn: 221534
2014-11-08 00:54:21 +08:00
define i64 @cmov4_ne(i32 signext %a, i64 %b, i64 %c) nounwind readnone {
[mips] Fix two patterns that select i32's (for MIPS32r6) / i64's (for MIPS64r6) from setne comparison with an i32. The patterns that are fixed: * (select (i32 (setne i32, immZExt16)), i32, i32) (for MIPS32r6) * (select (i32 (setne i32, immZExt16)), i64, i64) (for MIPS64r6) llvm-svn: 213653
2014-07-22 21:36:02 +08:00
entry:
%cmp = icmp ne i32 %a, 234
%cond = select i1 %cmp, i64 %b, i64 %c
ret i64 %cond
}
[mips] Fix inefficient code generation. This patch eliminates the need to emit a constant move instruction when this pattern is matched: (select (setgt a, Constant), T, F) The pattern above effectively turns into this: (conditional-move (setlt a, Constant + 1), F, T) llvm-svn: 176384
2013-03-02 05:52:08 +08:00
; slti and conditional move.
;
; Check that, pattern
; (select (setgt a, N), t, f)
; turns into
; (movz t, (setlt a, N + 1), f)
; if N + 1 fits in 16-bit.
[mips][mips64r6] c.cond.fmt, mov[fntz], and mov[fntz].[ds] are not available on MIPS32r6/MIPS64r6 Summary: c.cond.fmt has been replaced by cmp.cond.fmt. Where c.cond.fmt wrote to dedicated condition registers, cmp.cond.fmt writes 1 or 0 to normal FGR's (like the GPR comparisons). mov[fntz] have been replaced by seleqz and selnez. These instructions conditionally zero a register based on a bool in a GPR. The results can then be or'd together to act as a select without, for example, requiring a third register read port. mov[fntz].[ds] have been replaced with sel.[ds] MIPS64r6 currently generates unnecessary sign-extensions for most selects. This is because the result of a SETCC is currently an i32. Bits 32-63 are undefined in i32 and the behaviour of seleqz/selnez would otherwise depend on undefined bits. Later, we will fix this by making the result of SETCC an i64 on MIPS64 targets. Depends on D3958 Reviewers: jkolek, vmedic, zoran.jovanovic Reviewed By: vmedic, zoran.jovanovic Differential Revision: http://reviews.llvm.org/D4003 llvm-svn: 210777
2014-06-12 21:39:06 +08:00
; ALL-LABEL: slti0:
; 32-CMOV-DAG: addiu $[[I3:[0-9]+]], $zero, 3
; 32-CMOV-DAG: addiu $[[I5:2]], $zero, 5
; 32-CMOV-DAG: slti $[[R0:[0-9]+]], $4, 32767
; 32-CMOV-DAG: movz $[[I5]], $[[I3]], $[[R0]]
; 32-CMP-DAG: addiu $[[I3:[0-9]+]], $zero, 3
; 32-CMP-DAG: addiu $[[I5:[0-9]+]], $zero, 5
; 32-CMP-DAG: slti $[[R0:[0-9]+]], $4, 32767
; FIXME: We can do better than this by using selccz to choose between +0 and +2
[mips][mips64r6] Correct select patterns that have the condition or true/false values backwards Summary: This bug caused SingleSource/Regression/C/uint64_to_float and SingleSource/UnitTests/2002-05-02-CastTest3 to fail (among others). Differential Revision: http://reviews.llvm.org/D4388 llvm-svn: 212608
2014-07-09 18:47:26 +08:00
; 32-CMP-DAG: seleqz $[[T0:[0-9]+]], $[[I3]], $[[R0]]
; 32-CMP-DAG: selnez $[[T1:[0-9]+]], $[[I5]], $[[R0]]
[mips][mips64r6] c.cond.fmt, mov[fntz], and mov[fntz].[ds] are not available on MIPS32r6/MIPS64r6 Summary: c.cond.fmt has been replaced by cmp.cond.fmt. Where c.cond.fmt wrote to dedicated condition registers, cmp.cond.fmt writes 1 or 0 to normal FGR's (like the GPR comparisons). mov[fntz] have been replaced by seleqz and selnez. These instructions conditionally zero a register based on a bool in a GPR. The results can then be or'd together to act as a select without, for example, requiring a third register read port. mov[fntz].[ds] have been replaced with sel.[ds] MIPS64r6 currently generates unnecessary sign-extensions for most selects. This is because the result of a SETCC is currently an i32. Bits 32-63 are undefined in i32 and the behaviour of seleqz/selnez would otherwise depend on undefined bits. Later, we will fix this by making the result of SETCC an i64 on MIPS64 targets. Depends on D3958 Reviewers: jkolek, vmedic, zoran.jovanovic Reviewed By: vmedic, zoran.jovanovic Differential Revision: http://reviews.llvm.org/D4003 llvm-svn: 210777
2014-06-12 21:39:06 +08:00
; 32-CMP-DAG: or $2, $[[T0]], $[[T1]]
; 64-CMOV-DAG: addiu $[[I3:[0-9]+]], $zero, 3
; 64-CMOV-DAG: addiu $[[I5:2]], $zero, 5
; 64-CMOV-DAG: slti $[[R0:[0-9]+]], $4, 32767
; 64-CMOV-DAG: movz $[[I5]], $[[I3]], $[[R0]]
; 64-CMP-DAG: addiu $[[I3:[0-9]+]], $zero, 3
; 64-CMP-DAG: addiu $[[I5:[0-9]+]], $zero, 5
; 64-CMP-DAG: slti $[[R0:[0-9]+]], $4, 32767
; FIXME: We can do better than this by using selccz to choose between +0 and +2
[mips][mips64r6] Correct select patterns that have the condition or true/false values backwards Summary: This bug caused SingleSource/Regression/C/uint64_to_float and SingleSource/UnitTests/2002-05-02-CastTest3 to fail (among others). Differential Revision: http://reviews.llvm.org/D4388 llvm-svn: 212608
2014-07-09 18:47:26 +08:00
; 64-CMP-DAG: seleqz $[[T0:[0-9]+]], $[[I3]], $[[R0]]
; 64-CMP-DAG: selnez $[[T1:[0-9]+]], $[[I5]], $[[R0]]
[mips][mips64r6] c.cond.fmt, mov[fntz], and mov[fntz].[ds] are not available on MIPS32r6/MIPS64r6 Summary: c.cond.fmt has been replaced by cmp.cond.fmt. Where c.cond.fmt wrote to dedicated condition registers, cmp.cond.fmt writes 1 or 0 to normal FGR's (like the GPR comparisons). mov[fntz] have been replaced by seleqz and selnez. These instructions conditionally zero a register based on a bool in a GPR. The results can then be or'd together to act as a select without, for example, requiring a third register read port. mov[fntz].[ds] have been replaced with sel.[ds] MIPS64r6 currently generates unnecessary sign-extensions for most selects. This is because the result of a SETCC is currently an i32. Bits 32-63 are undefined in i32 and the behaviour of seleqz/selnez would otherwise depend on undefined bits. Later, we will fix this by making the result of SETCC an i64 on MIPS64 targets. Depends on D3958 Reviewers: jkolek, vmedic, zoran.jovanovic Reviewed By: vmedic, zoran.jovanovic Differential Revision: http://reviews.llvm.org/D4003 llvm-svn: 210777
2014-06-12 21:39:06 +08:00
; 64-CMP-DAG: or $2, $[[T0]], $[[T1]]
[mips] Fix inefficient code generation. This patch eliminates the need to emit a constant move instruction when this pattern is matched: (select (setgt a, Constant), T, F) The pattern above effectively turns into this: (conditional-move (setlt a, Constant + 1), F, T) llvm-svn: 176384
2013-03-02 05:52:08 +08:00
[mips] Promote i32 arguments to i64 for the N32/N64 ABI and fix <64-bit structs... Summary: ... and after all that refactoring, it's possible to distinguish softfloat floating point values from integers so this patch no longer breaks softfloat to do it. Remove direct handling of i32's in the N32/N64 ABI by promoting them to i64. This more closely reflects the ABI documentation and also fixes problems with stack arguments on big-endian targets. We now rely on signext/zeroext annotations (already generated by clang) and the Assert[SZ]ext nodes to avoid the introduction of unnecessary sign/zero extends. It was not possible to convert three tests to use signext/zeroext. These tests are bswap.ll, ctlz-v.ll, ctlz-v.ll. It's not possible to put signext on a vector type so we just accept the sign extends here for now. These tests don't pass the vectors the same way clang does (clang puts multiple elements in the same argument, these map 1 element to 1 argument) so we don't need to worry too much about it. With this patch, all known N32/N64 bugs should be fixed and we now pass the first 10,000 tests generated by ABITest.py. Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D6117 llvm-svn: 221534
2014-11-08 00:54:21 +08:00
define i32 @slti0(i32 signext %a) {
[mips] Fix inefficient code generation. This patch eliminates the need to emit a constant move instruction when this pattern is matched: (select (setgt a, Constant), T, F) The pattern above effectively turns into this: (conditional-move (setlt a, Constant + 1), F, T) llvm-svn: 176384
2013-03-02 05:52:08 +08:00
entry:
%cmp = icmp sgt i32 %a, 32766
[mips] Small code generation improvement for conditional operator (select) in case the operands are constants and its difference is |1|. It should be possible in those cases to rematerialize the result using MIPS's slt and similar instructions. The small update to some of the tests in cmov.ll, sel1c.ll and sel2c.ll was needed otherwise the optimization implemented in this patch would have been triggered (difference between the operands was 1) and that would have changed the semantic of the tests. llvm-svn: 196498
2013-12-05 20:07:05 +08:00
%cond = select i1 %cmp, i32 3, i32 5
[mips] Fix inefficient code generation. This patch eliminates the need to emit a constant move instruction when this pattern is matched: (select (setgt a, Constant), T, F) The pattern above effectively turns into this: (conditional-move (setlt a, Constant + 1), F, T) llvm-svn: 176384
2013-03-02 05:52:08 +08:00
ret i32 %cond
}
[mips][mips64r6] c.cond.fmt, mov[fntz], and mov[fntz].[ds] are not available on MIPS32r6/MIPS64r6 Summary: c.cond.fmt has been replaced by cmp.cond.fmt. Where c.cond.fmt wrote to dedicated condition registers, cmp.cond.fmt writes 1 or 0 to normal FGR's (like the GPR comparisons). mov[fntz] have been replaced by seleqz and selnez. These instructions conditionally zero a register based on a bool in a GPR. The results can then be or'd together to act as a select without, for example, requiring a third register read port. mov[fntz].[ds] have been replaced with sel.[ds] MIPS64r6 currently generates unnecessary sign-extensions for most selects. This is because the result of a SETCC is currently an i32. Bits 32-63 are undefined in i32 and the behaviour of seleqz/selnez would otherwise depend on undefined bits. Later, we will fix this by making the result of SETCC an i64 on MIPS64 targets. Depends on D3958 Reviewers: jkolek, vmedic, zoran.jovanovic Reviewed By: vmedic, zoran.jovanovic Differential Revision: http://reviews.llvm.org/D4003 llvm-svn: 210777
2014-06-12 21:39:06 +08:00
; ALL-LABEL: slti1:
; 32-CMOV-DAG: addiu $[[I7:[0-9]+]], $zero, 7
; 32-CMOV-DAG: addiu $[[I5:2]], $zero, 5
; 32-CMOV-DAG: addiu $[[R1:[0-9]+]], $zero, 32767
; 32-CMOV-DAG: slt $[[R0:[0-9]+]], $[[R1]], $4
; 32-CMOV-DAG: movn $[[I5]], $[[I7]], $[[R0]]
; 32-CMP-DAG: addiu $[[I7:[0-9]+]], $zero, 7
; 32-CMP-DAG: addiu $[[I5:[0-9]+]], $zero, 5
; 32-CMP-DAG: addiu $[[I32767:[0-9]+]], $zero, 32767
; 32-CMP-DAG: slt $[[R0:[0-9]+]], $[[I32767]], $4
; FIXME: We can do better than this by using selccz to choose between -0 and -2
; 32-CMP-DAG: selnez $[[T0:[0-9]+]], $[[I7]], $[[R0]]
; 32-CMP-DAG: seleqz $[[T1:[0-9]+]], $[[I5]], $[[R0]]
; 32-CMP-DAG: or $2, $[[T0]], $[[T1]]
; 64-CMOV-DAG: addiu $[[I7:[0-9]+]], $zero, 7
; 64-CMOV-DAG: addiu $[[I5:2]], $zero, 5
; 64-CMOV-DAG: addiu $[[R1:[0-9]+]], $zero, 32767
; 64-CMOV-DAG: slt $[[R0:[0-9]+]], $[[R1]], $4
; 64-CMOV-DAG: movn $[[I5]], $[[I7]], $[[R0]]
; 64-CMP-DAG: addiu $[[I7:[0-9]+]], $zero, 7
; 64-CMP-DAG: addiu $[[I5:2]], $zero, 5
; 64-CMP-DAG: addiu $[[R1:[0-9]+]], $zero, 32767
; 64-CMP-DAG: slt $[[R0:[0-9]+]], $[[R1]], $4
; FIXME: We can do better than this by using selccz to choose between -0 and -2
; 64-CMP-DAG: selnez $[[T0:[0-9]+]], $[[I7]], $[[R0]]
; 64-CMP-DAG: seleqz $[[T1:[0-9]+]], $[[I5]], $[[R0]]
; 64-CMP-DAG: or $2, $[[T0]], $[[T1]]
[mips] Fix inefficient code generation. This patch eliminates the need to emit a constant move instruction when this pattern is matched: (select (setgt a, Constant), T, F) The pattern above effectively turns into this: (conditional-move (setlt a, Constant + 1), F, T) llvm-svn: 176384
2013-03-02 05:52:08 +08:00
[mips] Promote i32 arguments to i64 for the N32/N64 ABI and fix <64-bit structs... Summary: ... and after all that refactoring, it's possible to distinguish softfloat floating point values from integers so this patch no longer breaks softfloat to do it. Remove direct handling of i32's in the N32/N64 ABI by promoting them to i64. This more closely reflects the ABI documentation and also fixes problems with stack arguments on big-endian targets. We now rely on signext/zeroext annotations (already generated by clang) and the Assert[SZ]ext nodes to avoid the introduction of unnecessary sign/zero extends. It was not possible to convert three tests to use signext/zeroext. These tests are bswap.ll, ctlz-v.ll, ctlz-v.ll. It's not possible to put signext on a vector type so we just accept the sign extends here for now. These tests don't pass the vectors the same way clang does (clang puts multiple elements in the same argument, these map 1 element to 1 argument) so we don't need to worry too much about it. With this patch, all known N32/N64 bugs should be fixed and we now pass the first 10,000 tests generated by ABITest.py. Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D6117 llvm-svn: 221534
2014-11-08 00:54:21 +08:00
define i32 @slti1(i32 signext %a) {
[mips] Fix inefficient code generation. This patch eliminates the need to emit a constant move instruction when this pattern is matched: (select (setgt a, Constant), T, F) The pattern above effectively turns into this: (conditional-move (setlt a, Constant + 1), F, T) llvm-svn: 176384
2013-03-02 05:52:08 +08:00
entry:
%cmp = icmp sgt i32 %a, 32767
[mips][mips64r6] c.cond.fmt, mov[fntz], and mov[fntz].[ds] are not available on MIPS32r6/MIPS64r6 Summary: c.cond.fmt has been replaced by cmp.cond.fmt. Where c.cond.fmt wrote to dedicated condition registers, cmp.cond.fmt writes 1 or 0 to normal FGR's (like the GPR comparisons). mov[fntz] have been replaced by seleqz and selnez. These instructions conditionally zero a register based on a bool in a GPR. The results can then be or'd together to act as a select without, for example, requiring a third register read port. mov[fntz].[ds] have been replaced with sel.[ds] MIPS64r6 currently generates unnecessary sign-extensions for most selects. This is because the result of a SETCC is currently an i32. Bits 32-63 are undefined in i32 and the behaviour of seleqz/selnez would otherwise depend on undefined bits. Later, we will fix this by making the result of SETCC an i64 on MIPS64 targets. Depends on D3958 Reviewers: jkolek, vmedic, zoran.jovanovic Reviewed By: vmedic, zoran.jovanovic Differential Revision: http://reviews.llvm.org/D4003 llvm-svn: 210777
2014-06-12 21:39:06 +08:00
%cond = select i1 %cmp, i32 7, i32 5
[mips] Fix inefficient code generation. This patch eliminates the need to emit a constant move instruction when this pattern is matched: (select (setgt a, Constant), T, F) The pattern above effectively turns into this: (conditional-move (setlt a, Constant + 1), F, T) llvm-svn: 176384
2013-03-02 05:52:08 +08:00
ret i32 %cond
}
[mips][mips64r6] c.cond.fmt, mov[fntz], and mov[fntz].[ds] are not available on MIPS32r6/MIPS64r6 Summary: c.cond.fmt has been replaced by cmp.cond.fmt. Where c.cond.fmt wrote to dedicated condition registers, cmp.cond.fmt writes 1 or 0 to normal FGR's (like the GPR comparisons). mov[fntz] have been replaced by seleqz and selnez. These instructions conditionally zero a register based on a bool in a GPR. The results can then be or'd together to act as a select without, for example, requiring a third register read port. mov[fntz].[ds] have been replaced with sel.[ds] MIPS64r6 currently generates unnecessary sign-extensions for most selects. This is because the result of a SETCC is currently an i32. Bits 32-63 are undefined in i32 and the behaviour of seleqz/selnez would otherwise depend on undefined bits. Later, we will fix this by making the result of SETCC an i64 on MIPS64 targets. Depends on D3958 Reviewers: jkolek, vmedic, zoran.jovanovic Reviewed By: vmedic, zoran.jovanovic Differential Revision: http://reviews.llvm.org/D4003 llvm-svn: 210777
2014-06-12 21:39:06 +08:00
; ALL-LABEL: slti2:
; 32-CMOV-DAG: addiu $[[I3:[0-9]+]], $zero, 3
; 32-CMOV-DAG: addiu $[[I5:2]], $zero, 5
; 32-CMOV-DAG: slti $[[R0:[0-9]+]], $4, -32768
; 32-CMOV-DAG: movz $[[I5]], $[[I3]], $[[R0]]
; 32-CMP-DAG: addiu $[[I3:[0-9]+]], $zero, 3
; 32-CMP-DAG: addiu $[[I5:[0-9]+]], $zero, 5
; 32-CMP-DAG: slti $[[R0:[0-9]+]], $4, -32768
; FIXME: We can do better than this by using selccz to choose between +0 and +2
[mips][mips64r6] Correct select patterns that have the condition or true/false values backwards Summary: This bug caused SingleSource/Regression/C/uint64_to_float and SingleSource/UnitTests/2002-05-02-CastTest3 to fail (among others). Differential Revision: http://reviews.llvm.org/D4388 llvm-svn: 212608
2014-07-09 18:47:26 +08:00
; 32-CMP-DAG: seleqz $[[T0:[0-9]+]], $[[I3]], $[[R0]]
; 32-CMP-DAG: selnez $[[T1:[0-9]+]], $[[I5]], $[[R0]]
[mips][mips64r6] c.cond.fmt, mov[fntz], and mov[fntz].[ds] are not available on MIPS32r6/MIPS64r6 Summary: c.cond.fmt has been replaced by cmp.cond.fmt. Where c.cond.fmt wrote to dedicated condition registers, cmp.cond.fmt writes 1 or 0 to normal FGR's (like the GPR comparisons). mov[fntz] have been replaced by seleqz and selnez. These instructions conditionally zero a register based on a bool in a GPR. The results can then be or'd together to act as a select without, for example, requiring a third register read port. mov[fntz].[ds] have been replaced with sel.[ds] MIPS64r6 currently generates unnecessary sign-extensions for most selects. This is because the result of a SETCC is currently an i32. Bits 32-63 are undefined in i32 and the behaviour of seleqz/selnez would otherwise depend on undefined bits. Later, we will fix this by making the result of SETCC an i64 on MIPS64 targets. Depends on D3958 Reviewers: jkolek, vmedic, zoran.jovanovic Reviewed By: vmedic, zoran.jovanovic Differential Revision: http://reviews.llvm.org/D4003 llvm-svn: 210777
2014-06-12 21:39:06 +08:00
; 32-CMP-DAG: or $2, $[[T0]], $[[T1]]
; 64-CMOV-DAG: addiu $[[I3:[0-9]+]], $zero, 3
; 64-CMOV-DAG: addiu $[[I5:2]], $zero, 5
; 64-CMOV-DAG: slti $[[R0:[0-9]+]], $4, -32768
; 64-CMOV-DAG: movz $[[I5]], $[[I3]], $[[R0]]
; 64-CMP-DAG: addiu $[[I3:[0-9]+]], $zero, 3
; 64-CMP-DAG: addiu $[[I5:[0-9]+]], $zero, 5
; 64-CMP-DAG: slti $[[R0:[0-9]+]], $4, -32768
; FIXME: We can do better than this by using selccz to choose between +0 and +2
[mips][mips64r6] Correct select patterns that have the condition or true/false values backwards Summary: This bug caused SingleSource/Regression/C/uint64_to_float and SingleSource/UnitTests/2002-05-02-CastTest3 to fail (among others). Differential Revision: http://reviews.llvm.org/D4388 llvm-svn: 212608
2014-07-09 18:47:26 +08:00
; 64-CMP-DAG: seleqz $[[T0:[0-9]+]], $[[I3]], $[[R0]]
; 64-CMP-DAG: selnez $[[T1:[0-9]+]], $[[I5]], $[[R0]]
[mips][mips64r6] c.cond.fmt, mov[fntz], and mov[fntz].[ds] are not available on MIPS32r6/MIPS64r6 Summary: c.cond.fmt has been replaced by cmp.cond.fmt. Where c.cond.fmt wrote to dedicated condition registers, cmp.cond.fmt writes 1 or 0 to normal FGR's (like the GPR comparisons). mov[fntz] have been replaced by seleqz and selnez. These instructions conditionally zero a register based on a bool in a GPR. The results can then be or'd together to act as a select without, for example, requiring a third register read port. mov[fntz].[ds] have been replaced with sel.[ds] MIPS64r6 currently generates unnecessary sign-extensions for most selects. This is because the result of a SETCC is currently an i32. Bits 32-63 are undefined in i32 and the behaviour of seleqz/selnez would otherwise depend on undefined bits. Later, we will fix this by making the result of SETCC an i64 on MIPS64 targets. Depends on D3958 Reviewers: jkolek, vmedic, zoran.jovanovic Reviewed By: vmedic, zoran.jovanovic Differential Revision: http://reviews.llvm.org/D4003 llvm-svn: 210777
2014-06-12 21:39:06 +08:00
; 64-CMP-DAG: or $2, $[[T0]], $[[T1]]
[mips] Fix inefficient code generation. This patch eliminates the need to emit a constant move instruction when this pattern is matched: (select (setgt a, Constant), T, F) The pattern above effectively turns into this: (conditional-move (setlt a, Constant + 1), F, T) llvm-svn: 176384
2013-03-02 05:52:08 +08:00
[mips] Promote i32 arguments to i64 for the N32/N64 ABI and fix <64-bit structs... Summary: ... and after all that refactoring, it's possible to distinguish softfloat floating point values from integers so this patch no longer breaks softfloat to do it. Remove direct handling of i32's in the N32/N64 ABI by promoting them to i64. This more closely reflects the ABI documentation and also fixes problems with stack arguments on big-endian targets. We now rely on signext/zeroext annotations (already generated by clang) and the Assert[SZ]ext nodes to avoid the introduction of unnecessary sign/zero extends. It was not possible to convert three tests to use signext/zeroext. These tests are bswap.ll, ctlz-v.ll, ctlz-v.ll. It's not possible to put signext on a vector type so we just accept the sign extends here for now. These tests don't pass the vectors the same way clang does (clang puts multiple elements in the same argument, these map 1 element to 1 argument) so we don't need to worry too much about it. With this patch, all known N32/N64 bugs should be fixed and we now pass the first 10,000 tests generated by ABITest.py. Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D6117 llvm-svn: 221534
2014-11-08 00:54:21 +08:00
define i32 @slti2(i32 signext %a) {
[mips] Fix inefficient code generation. This patch eliminates the need to emit a constant move instruction when this pattern is matched: (select (setgt a, Constant), T, F) The pattern above effectively turns into this: (conditional-move (setlt a, Constant + 1), F, T) llvm-svn: 176384
2013-03-02 05:52:08 +08:00
entry:
%cmp = icmp sgt i32 %a, -32769
[mips] Small code generation improvement for conditional operator (select) in case the operands are constants and its difference is |1|. It should be possible in those cases to rematerialize the result using MIPS's slt and similar instructions. The small update to some of the tests in cmov.ll, sel1c.ll and sel2c.ll was needed otherwise the optimization implemented in this patch would have been triggered (difference between the operands was 1) and that would have changed the semantic of the tests. llvm-svn: 196498
2013-12-05 20:07:05 +08:00
%cond = select i1 %cmp, i32 3, i32 5
[mips] Fix inefficient code generation. This patch eliminates the need to emit a constant move instruction when this pattern is matched: (select (setgt a, Constant), T, F) The pattern above effectively turns into this: (conditional-move (setlt a, Constant + 1), F, T) llvm-svn: 176384
2013-03-02 05:52:08 +08:00
ret i32 %cond
}
[mips][mips64r6] c.cond.fmt, mov[fntz], and mov[fntz].[ds] are not available on MIPS32r6/MIPS64r6 Summary: c.cond.fmt has been replaced by cmp.cond.fmt. Where c.cond.fmt wrote to dedicated condition registers, cmp.cond.fmt writes 1 or 0 to normal FGR's (like the GPR comparisons). mov[fntz] have been replaced by seleqz and selnez. These instructions conditionally zero a register based on a bool in a GPR. The results can then be or'd together to act as a select without, for example, requiring a third register read port. mov[fntz].[ds] have been replaced with sel.[ds] MIPS64r6 currently generates unnecessary sign-extensions for most selects. This is because the result of a SETCC is currently an i32. Bits 32-63 are undefined in i32 and the behaviour of seleqz/selnez would otherwise depend on undefined bits. Later, we will fix this by making the result of SETCC an i64 on MIPS64 targets. Depends on D3958 Reviewers: jkolek, vmedic, zoran.jovanovic Reviewed By: vmedic, zoran.jovanovic Differential Revision: http://reviews.llvm.org/D4003 llvm-svn: 210777
2014-06-12 21:39:06 +08:00
; ALL-LABEL: slti3:
; 32-CMOV-DAG: addiu $[[I3:[0-9]+]], $zero, 3
; 32-CMOV-DAG: addiu $[[I5:2]], $zero, 5
; 32-CMOV-DAG: lui $[[R1:[0-9]+]], 65535
; 32-CMOV-DAG: ori $[[R1]], $[[R1]], 32766
; 32-CMOV-DAG: slt $[[R0:[0-9]+]], $[[R1]], $4
; 32-CMOV-DAG: movn $[[I5]], $[[I3]], $[[R0]]
; 32-CMP-DAG: addiu $[[I3:[0-9]+]], $zero, 3
; 32-CMP-DAG: addiu $[[I5:[0-9]+]], $zero, 5
; 32-CMP-DAG: lui $[[IMM:[0-9]+]], 65535
; 32-CMP-DAG: ori $[[IMM]], $[[IMM]], 32766
; 32-CMP-DAG: slt $[[R0:[0-9]+]], $[[I32767]], $4
; FIXME: We can do better than this by using selccz to choose between -0 and -2
; 32-CMP-DAG: selnez $[[T0:[0-9]+]], $[[I3]], $[[R0]]
; 32-CMP-DAG: seleqz $[[T1:[0-9]+]], $[[I5]], $[[R0]]
; 32-CMP-DAG: or $2, $[[T0]], $[[T1]]
; 64-CMOV-DAG: addiu $[[I3:[0-9]+]], $zero, 3
; 64-CMOV-DAG: addiu $[[I5:2]], $zero, 5
; 64-CMOV-DAG: lui $[[R1:[0-9]+]], 65535
; 64-CMOV-DAG: ori $[[R1]], $[[R1]], 32766
; 64-CMOV-DAG: slt $[[R0:[0-9]+]], $[[R1]], $4
; 64-CMOV-DAG: movn $[[I5]], $[[I3]], $[[R0]]
; 64-CMP-DAG: addiu $[[I3:[0-9]+]], $zero, 3
; 64-CMP-DAG: addiu $[[I5:2]], $zero, 5
; 64-CMP-DAG: lui $[[IMM:[0-9]+]], 65535
; 64-CMP-DAG: ori $[[IMM]], $[[IMM]], 32766
; 64-CMP-DAG: slt $[[R0:[0-9]+]], $[[IMM]], $4
; FIXME: We can do better than this by using selccz to choose between -0 and -2
; 64-CMP-DAG: selnez $[[T0:[0-9]+]], $[[I3]], $[[R0]]
; 64-CMP-DAG: seleqz $[[T1:[0-9]+]], $[[I5]], $[[R0]]
; 64-CMP-DAG: or $2, $[[T0]], $[[T1]]
[mips] Fix inefficient code generation. This patch eliminates the need to emit a constant move instruction when this pattern is matched: (select (setgt a, Constant), T, F) The pattern above effectively turns into this: (conditional-move (setlt a, Constant + 1), F, T) llvm-svn: 176384
2013-03-02 05:52:08 +08:00
[mips] Promote i32 arguments to i64 for the N32/N64 ABI and fix <64-bit structs... Summary: ... and after all that refactoring, it's possible to distinguish softfloat floating point values from integers so this patch no longer breaks softfloat to do it. Remove direct handling of i32's in the N32/N64 ABI by promoting them to i64. This more closely reflects the ABI documentation and also fixes problems with stack arguments on big-endian targets. We now rely on signext/zeroext annotations (already generated by clang) and the Assert[SZ]ext nodes to avoid the introduction of unnecessary sign/zero extends. It was not possible to convert three tests to use signext/zeroext. These tests are bswap.ll, ctlz-v.ll, ctlz-v.ll. It's not possible to put signext on a vector type so we just accept the sign extends here for now. These tests don't pass the vectors the same way clang does (clang puts multiple elements in the same argument, these map 1 element to 1 argument) so we don't need to worry too much about it. With this patch, all known N32/N64 bugs should be fixed and we now pass the first 10,000 tests generated by ABITest.py. Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D6117 llvm-svn: 221534
2014-11-08 00:54:21 +08:00
define i32 @slti3(i32 signext %a) {
[mips] Fix inefficient code generation. This patch eliminates the need to emit a constant move instruction when this pattern is matched: (select (setgt a, Constant), T, F) The pattern above effectively turns into this: (conditional-move (setlt a, Constant + 1), F, T) llvm-svn: 176384
2013-03-02 05:52:08 +08:00
entry:
%cmp = icmp sgt i32 %a, -32770
[mips] Small code generation improvement for conditional operator (select) in case the operands are constants and its difference is |1|. It should be possible in those cases to rematerialize the result using MIPS's slt and similar instructions. The small update to some of the tests in cmov.ll, sel1c.ll and sel2c.ll was needed otherwise the optimization implemented in this patch would have been triggered (difference between the operands was 1) and that would have changed the semantic of the tests. llvm-svn: 196498
2013-12-05 20:07:05 +08:00
%cond = select i1 %cmp, i32 3, i32 5
[mips] Fix inefficient code generation. This patch eliminates the need to emit a constant move instruction when this pattern is matched: (select (setgt a, Constant), T, F) The pattern above effectively turns into this: (conditional-move (setlt a, Constant + 1), F, T) llvm-svn: 176384
2013-03-02 05:52:08 +08:00
ret i32 %cond
}
; 64-bit patterns.
[mips][mips64r6] c.cond.fmt, mov[fntz], and mov[fntz].[ds] are not available on MIPS32r6/MIPS64r6 Summary: c.cond.fmt has been replaced by cmp.cond.fmt. Where c.cond.fmt wrote to dedicated condition registers, cmp.cond.fmt writes 1 or 0 to normal FGR's (like the GPR comparisons). mov[fntz] have been replaced by seleqz and selnez. These instructions conditionally zero a register based on a bool in a GPR. The results can then be or'd together to act as a select without, for example, requiring a third register read port. mov[fntz].[ds] have been replaced with sel.[ds] MIPS64r6 currently generates unnecessary sign-extensions for most selects. This is because the result of a SETCC is currently an i32. Bits 32-63 are undefined in i32 and the behaviour of seleqz/selnez would otherwise depend on undefined bits. Later, we will fix this by making the result of SETCC an i64 on MIPS64 targets. Depends on D3958 Reviewers: jkolek, vmedic, zoran.jovanovic Reviewed By: vmedic, zoran.jovanovic Differential Revision: http://reviews.llvm.org/D4003 llvm-svn: 210777
2014-06-12 21:39:06 +08:00
; ALL-LABEL: slti64_0:
; 32-CMOV-DAG: slt $[[CC:[0-9]+]], $zero, $4
; 32-CMOV-DAG: addiu $[[I32766:[0-9]+]], $zero, 32766
; 32-CMOV-DAG: sltu $[[R1:[0-9]+]], $[[I32766]], $5
; 32-CMOV-DAG: movz $[[CC:[0-9]+]], $[[R1]], $4
; 32-CMOV-DAG: addiu $[[I5:[0-9]+]], $zero, 5
; 32-CMOV-DAG: addiu $[[I4:3]], $zero, 4
; 32-CMOV-DAG: movn $[[I4]], $[[I5]], $[[CC]]
; 32-CMOV-DAG: addiu $2, $zero, 0
; 32-CMP-DAG: slt $[[CC0:[0-9]+]], $zero, $4
; 32-CMP-DAG: addiu $[[I32766:[0-9]+]], $zero, 32766
; 32-CMP-DAG: sltu $[[CC1:[0-9]+]], $[[I32766]], $5
[mips][mips64r6] Correct select patterns that have the condition or true/false values backwards Summary: This bug caused SingleSource/Regression/C/uint64_to_float and SingleSource/UnitTests/2002-05-02-CastTest3 to fail (among others). Differential Revision: http://reviews.llvm.org/D4388 llvm-svn: 212608
2014-07-09 18:47:26 +08:00
; 32-CMP-DAG: selnez $[[CC2:[0-9]+]], $[[CC0]], $4
; 32-CMP-DAG: seleqz $[[CC3:[0-9]+]], $[[CC1]], $4
[mips][mips64r6] c.cond.fmt, mov[fntz], and mov[fntz].[ds] are not available on MIPS32r6/MIPS64r6 Summary: c.cond.fmt has been replaced by cmp.cond.fmt. Where c.cond.fmt wrote to dedicated condition registers, cmp.cond.fmt writes 1 or 0 to normal FGR's (like the GPR comparisons). mov[fntz] have been replaced by seleqz and selnez. These instructions conditionally zero a register based on a bool in a GPR. The results can then be or'd together to act as a select without, for example, requiring a third register read port. mov[fntz].[ds] have been replaced with sel.[ds] MIPS64r6 currently generates unnecessary sign-extensions for most selects. This is because the result of a SETCC is currently an i32. Bits 32-63 are undefined in i32 and the behaviour of seleqz/selnez would otherwise depend on undefined bits. Later, we will fix this by making the result of SETCC an i64 on MIPS64 targets. Depends on D3958 Reviewers: jkolek, vmedic, zoran.jovanovic Reviewed By: vmedic, zoran.jovanovic Differential Revision: http://reviews.llvm.org/D4003 llvm-svn: 210777
2014-06-12 21:39:06 +08:00
; 32-CMP: or $[[CC:[0-9]+]], $[[CC3]], $[[CC2]]
; 32-CMP-DAG: addiu $[[I5:[0-9]+]], $zero, 5
; 32-CMP-DAG: addiu $[[I4:[0-9]+]], $zero, 4
; 32-CMP-DAG: seleqz $[[T0:[0-9]+]], $[[I4]], $[[CC]]
; 32-CMP-DAG: selnez $[[T1:[0-9]+]], $[[I5]], $[[CC]]
; 32-CMP-DAG: or $3, $[[T1]], $[[T0]]
; 32-CMP-DAG: addiu $2, $zero, 0
; 64-CMOV-DAG: addiu $[[I5:[0-9]+]], $zero, 5
; 64-CMOV-DAG: addiu $[[I4:2]], $zero, 4
; 64-CMOV-DAG: slti $[[R0:[0-9]+]], $4, 32767
; 64-CMOV-DAG: movz $[[I4]], $[[I5]], $[[R0]]
; 64-CMP-DAG: addiu $[[I5:[0-9]+]], $zero, 5
; 64-CMP-DAG: addiu $[[I4:[0-9]+]], $zero, 4
; 64-CMP-DAG: slti $[[R0:[0-9]+]], $4, 32767
; FIXME: We can do better than this by adding/subtracting the result of slti
; to/from one of the constants.
[mips][mips64r6] Correct select patterns that have the condition or true/false values backwards Summary: This bug caused SingleSource/Regression/C/uint64_to_float and SingleSource/UnitTests/2002-05-02-CastTest3 to fail (among others). Differential Revision: http://reviews.llvm.org/D4388 llvm-svn: 212608
2014-07-09 18:47:26 +08:00
; 64-CMP-DAG: seleqz $[[T0:[0-9]+]], $[[I5]], $[[R0]]
; 64-CMP-DAG: selnez $[[T1:[0-9]+]], $[[I4]], $[[R0]]
[mips][mips64r6] c.cond.fmt, mov[fntz], and mov[fntz].[ds] are not available on MIPS32r6/MIPS64r6 Summary: c.cond.fmt has been replaced by cmp.cond.fmt. Where c.cond.fmt wrote to dedicated condition registers, cmp.cond.fmt writes 1 or 0 to normal FGR's (like the GPR comparisons). mov[fntz] have been replaced by seleqz and selnez. These instructions conditionally zero a register based on a bool in a GPR. The results can then be or'd together to act as a select without, for example, requiring a third register read port. mov[fntz].[ds] have been replaced with sel.[ds] MIPS64r6 currently generates unnecessary sign-extensions for most selects. This is because the result of a SETCC is currently an i32. Bits 32-63 are undefined in i32 and the behaviour of seleqz/selnez would otherwise depend on undefined bits. Later, we will fix this by making the result of SETCC an i64 on MIPS64 targets. Depends on D3958 Reviewers: jkolek, vmedic, zoran.jovanovic Reviewed By: vmedic, zoran.jovanovic Differential Revision: http://reviews.llvm.org/D4003 llvm-svn: 210777
2014-06-12 21:39:06 +08:00
; 64-CMP-DAG: or $2, $[[T0]], $[[T1]]
[mips] Fix inefficient code generation. This patch eliminates the need to emit a constant move instruction when this pattern is matched: (select (setgt a, Constant), T, F) The pattern above effectively turns into this: (conditional-move (setlt a, Constant + 1), F, T) llvm-svn: 176384
2013-03-02 05:52:08 +08:00
define i64 @slti64_0(i64 %a) {
entry:
%cmp = icmp sgt i64 %a, 32766
[mips][mips64r6] c.cond.fmt, mov[fntz], and mov[fntz].[ds] are not available on MIPS32r6/MIPS64r6 Summary: c.cond.fmt has been replaced by cmp.cond.fmt. Where c.cond.fmt wrote to dedicated condition registers, cmp.cond.fmt writes 1 or 0 to normal FGR's (like the GPR comparisons). mov[fntz] have been replaced by seleqz and selnez. These instructions conditionally zero a register based on a bool in a GPR. The results can then be or'd together to act as a select without, for example, requiring a third register read port. mov[fntz].[ds] have been replaced with sel.[ds] MIPS64r6 currently generates unnecessary sign-extensions for most selects. This is because the result of a SETCC is currently an i32. Bits 32-63 are undefined in i32 and the behaviour of seleqz/selnez would otherwise depend on undefined bits. Later, we will fix this by making the result of SETCC an i64 on MIPS64 targets. Depends on D3958 Reviewers: jkolek, vmedic, zoran.jovanovic Reviewed By: vmedic, zoran.jovanovic Differential Revision: http://reviews.llvm.org/D4003 llvm-svn: 210777
2014-06-12 21:39:06 +08:00
%conv = select i1 %cmp, i64 5, i64 4
[mips] Fix inefficient code generation. This patch eliminates the need to emit a constant move instruction when this pattern is matched: (select (setgt a, Constant), T, F) The pattern above effectively turns into this: (conditional-move (setlt a, Constant + 1), F, T) llvm-svn: 176384
2013-03-02 05:52:08 +08:00
ret i64 %conv
}
[mips][mips64r6] c.cond.fmt, mov[fntz], and mov[fntz].[ds] are not available on MIPS32r6/MIPS64r6 Summary: c.cond.fmt has been replaced by cmp.cond.fmt. Where c.cond.fmt wrote to dedicated condition registers, cmp.cond.fmt writes 1 or 0 to normal FGR's (like the GPR comparisons). mov[fntz] have been replaced by seleqz and selnez. These instructions conditionally zero a register based on a bool in a GPR. The results can then be or'd together to act as a select without, for example, requiring a third register read port. mov[fntz].[ds] have been replaced with sel.[ds] MIPS64r6 currently generates unnecessary sign-extensions for most selects. This is because the result of a SETCC is currently an i32. Bits 32-63 are undefined in i32 and the behaviour of seleqz/selnez would otherwise depend on undefined bits. Later, we will fix this by making the result of SETCC an i64 on MIPS64 targets. Depends on D3958 Reviewers: jkolek, vmedic, zoran.jovanovic Reviewed By: vmedic, zoran.jovanovic Differential Revision: http://reviews.llvm.org/D4003 llvm-svn: 210777
2014-06-12 21:39:06 +08:00
; ALL-LABEL: slti64_1:
; 32-CMOV-DAG: slt $[[CC:[0-9]+]], $zero, $4
; 32-CMOV-DAG: addiu $[[I32766:[0-9]+]], $zero, 32767
; 32-CMOV-DAG: sltu $[[R1:[0-9]+]], $[[I32766]], $5
; 32-CMOV-DAG: movz $[[CC:[0-9]+]], $[[R1]], $4
; 32-CMOV-DAG: addiu $[[I5:[0-9]+]], $zero, 5
; 32-CMOV-DAG: addiu $[[I4:3]], $zero, 4
; 32-CMOV-DAG: movn $[[I4]], $[[I5]], $[[CC]]
; 32-CMOV-DAG: addiu $2, $zero, 0
; 32-CMP-DAG: slt $[[CC0:[0-9]+]], $zero, $4
; 32-CMP-DAG: addiu $[[I32766:[0-9]+]], $zero, 32767
; 32-CMP-DAG: sltu $[[CC1:[0-9]+]], $[[I32766]], $5
[mips][mips64r6] Correct select patterns that have the condition or true/false values backwards Summary: This bug caused SingleSource/Regression/C/uint64_to_float and SingleSource/UnitTests/2002-05-02-CastTest3 to fail (among others). Differential Revision: http://reviews.llvm.org/D4388 llvm-svn: 212608
2014-07-09 18:47:26 +08:00
; 32-CMP-DAG: selnez $[[CC2:[0-9]+]], $[[CC0]], $4
; 32-CMP-DAG: seleqz $[[CC3:[0-9]+]], $[[CC1]], $4
[mips][mips64r6] c.cond.fmt, mov[fntz], and mov[fntz].[ds] are not available on MIPS32r6/MIPS64r6 Summary: c.cond.fmt has been replaced by cmp.cond.fmt. Where c.cond.fmt wrote to dedicated condition registers, cmp.cond.fmt writes 1 or 0 to normal FGR's (like the GPR comparisons). mov[fntz] have been replaced by seleqz and selnez. These instructions conditionally zero a register based on a bool in a GPR. The results can then be or'd together to act as a select without, for example, requiring a third register read port. mov[fntz].[ds] have been replaced with sel.[ds] MIPS64r6 currently generates unnecessary sign-extensions for most selects. This is because the result of a SETCC is currently an i32. Bits 32-63 are undefined in i32 and the behaviour of seleqz/selnez would otherwise depend on undefined bits. Later, we will fix this by making the result of SETCC an i64 on MIPS64 targets. Depends on D3958 Reviewers: jkolek, vmedic, zoran.jovanovic Reviewed By: vmedic, zoran.jovanovic Differential Revision: http://reviews.llvm.org/D4003 llvm-svn: 210777
2014-06-12 21:39:06 +08:00
; 32-CMP: or $[[CC:[0-9]+]], $[[CC3]], $[[CC2]]
; 32-CMP-DAG: addiu $[[I5:[0-9]+]], $zero, 5
; 32-CMP-DAG: addiu $[[I4:[0-9]+]], $zero, 4
; 32-CMP-DAG: seleqz $[[T0:[0-9]+]], $[[I4]], $[[CC]]
; 32-CMP-DAG: selnez $[[T1:[0-9]+]], $[[I5]], $[[CC]]
; 32-CMP-DAG: or $3, $[[T1]], $[[T0]]
; 32-CMP-DAG: addiu $2, $zero, 0
; 64-CMOV-DAG: daddiu $[[I5:[0-9]+]], $zero, 5
; 64-CMOV-DAG: daddiu $[[I4:2]], $zero, 4
; 64-CMOV-DAG: daddiu $[[R1:[0-9]+]], $zero, 32767
; 64-CMOV-DAG: slt $[[R0:[0-9]+]], $[[R1]], $4
; 64-CMOV-DAG: movn $[[I4]], $[[I5]], $[[R0]]
; 64-CMP-DAG: daddiu $[[I5:[0-9]+]], $zero, 5
; 64-CMP-DAG: daddiu $[[I4:2]], $zero, 4
; 64-CMP-DAG: daddiu $[[R1:[0-9]+]], $zero, 32767
; 64-CMP-DAG: slt $[[R0:[0-9]+]], $[[R1]], $4
; FIXME: We can do better than this by using selccz to choose between -0 and -2
; 64-CMP-DAG: selnez $[[T0:[0-9]+]], $[[I5]], $[[R0]]
; 64-CMP-DAG: seleqz $[[T1:[0-9]+]], $[[I4]], $[[R0]]
; 64-CMP-DAG: or $2, $[[T0]], $[[T1]]
[mips] Fix inefficient code generation. This patch eliminates the need to emit a constant move instruction when this pattern is matched: (select (setgt a, Constant), T, F) The pattern above effectively turns into this: (conditional-move (setlt a, Constant + 1), F, T) llvm-svn: 176384
2013-03-02 05:52:08 +08:00
define i64 @slti64_1(i64 %a) {
entry:
%cmp = icmp sgt i64 %a, 32767
[mips][mips64r6] c.cond.fmt, mov[fntz], and mov[fntz].[ds] are not available on MIPS32r6/MIPS64r6 Summary: c.cond.fmt has been replaced by cmp.cond.fmt. Where c.cond.fmt wrote to dedicated condition registers, cmp.cond.fmt writes 1 or 0 to normal FGR's (like the GPR comparisons). mov[fntz] have been replaced by seleqz and selnez. These instructions conditionally zero a register based on a bool in a GPR. The results can then be or'd together to act as a select without, for example, requiring a third register read port. mov[fntz].[ds] have been replaced with sel.[ds] MIPS64r6 currently generates unnecessary sign-extensions for most selects. This is because the result of a SETCC is currently an i32. Bits 32-63 are undefined in i32 and the behaviour of seleqz/selnez would otherwise depend on undefined bits. Later, we will fix this by making the result of SETCC an i64 on MIPS64 targets. Depends on D3958 Reviewers: jkolek, vmedic, zoran.jovanovic Reviewed By: vmedic, zoran.jovanovic Differential Revision: http://reviews.llvm.org/D4003 llvm-svn: 210777
2014-06-12 21:39:06 +08:00
%conv = select i1 %cmp, i64 5, i64 4
[mips] Fix inefficient code generation. This patch eliminates the need to emit a constant move instruction when this pattern is matched: (select (setgt a, Constant), T, F) The pattern above effectively turns into this: (conditional-move (setlt a, Constant + 1), F, T) llvm-svn: 176384
2013-03-02 05:52:08 +08:00
ret i64 %conv
}
[mips][mips64r6] c.cond.fmt, mov[fntz], and mov[fntz].[ds] are not available on MIPS32r6/MIPS64r6 Summary: c.cond.fmt has been replaced by cmp.cond.fmt. Where c.cond.fmt wrote to dedicated condition registers, cmp.cond.fmt writes 1 or 0 to normal FGR's (like the GPR comparisons). mov[fntz] have been replaced by seleqz and selnez. These instructions conditionally zero a register based on a bool in a GPR. The results can then be or'd together to act as a select without, for example, requiring a third register read port. mov[fntz].[ds] have been replaced with sel.[ds] MIPS64r6 currently generates unnecessary sign-extensions for most selects. This is because the result of a SETCC is currently an i32. Bits 32-63 are undefined in i32 and the behaviour of seleqz/selnez would otherwise depend on undefined bits. Later, we will fix this by making the result of SETCC an i64 on MIPS64 targets. Depends on D3958 Reviewers: jkolek, vmedic, zoran.jovanovic Reviewed By: vmedic, zoran.jovanovic Differential Revision: http://reviews.llvm.org/D4003 llvm-svn: 210777
2014-06-12 21:39:06 +08:00
; ALL-LABEL: slti64_2:
; FIXME: The 32-bit versions of this test are too complicated to reasonably
; match at the moment. They do show some missing optimizations though
; such as:
; (movz $a, $b, (neg $c)) -> (movn $a, $b, $c)
; 64-CMOV-DAG: addiu $[[I3:[0-9]+]], $zero, 3
; 64-CMOV-DAG: addiu $[[I4:2]], $zero, 4
; 64-CMOV-DAG: slti $[[R0:[0-9]+]], $4, -32768
; 64-CMOV-DAG: movz $[[I4]], $[[I3]], $[[R0]]
; 64-CMP-DAG: addiu $[[I3:[0-9]+]], $zero, 3
; 64-CMP-DAG: addiu $[[I4:[0-9]+]], $zero, 4
; 64-CMP-DAG: slti $[[R0:[0-9]+]], $4, -32768
; FIXME: We can do better than this by adding/subtracting the result of slti
; to/from one of the constants.
[mips][mips64r6] Correct select patterns that have the condition or true/false values backwards Summary: This bug caused SingleSource/Regression/C/uint64_to_float and SingleSource/UnitTests/2002-05-02-CastTest3 to fail (among others). Differential Revision: http://reviews.llvm.org/D4388 llvm-svn: 212608
2014-07-09 18:47:26 +08:00
; 64-CMP-DAG: seleqz $[[T0:[0-9]+]], $[[I3]], $[[R0]]
; 64-CMP-DAG: selnez $[[T1:[0-9]+]], $[[I4]], $[[R0]]
[mips][mips64r6] c.cond.fmt, mov[fntz], and mov[fntz].[ds] are not available on MIPS32r6/MIPS64r6 Summary: c.cond.fmt has been replaced by cmp.cond.fmt. Where c.cond.fmt wrote to dedicated condition registers, cmp.cond.fmt writes 1 or 0 to normal FGR's (like the GPR comparisons). mov[fntz] have been replaced by seleqz and selnez. These instructions conditionally zero a register based on a bool in a GPR. The results can then be or'd together to act as a select without, for example, requiring a third register read port. mov[fntz].[ds] have been replaced with sel.[ds] MIPS64r6 currently generates unnecessary sign-extensions for most selects. This is because the result of a SETCC is currently an i32. Bits 32-63 are undefined in i32 and the behaviour of seleqz/selnez would otherwise depend on undefined bits. Later, we will fix this by making the result of SETCC an i64 on MIPS64 targets. Depends on D3958 Reviewers: jkolek, vmedic, zoran.jovanovic Reviewed By: vmedic, zoran.jovanovic Differential Revision: http://reviews.llvm.org/D4003 llvm-svn: 210777
2014-06-12 21:39:06 +08:00
; 64-CMP-DAG: or $2, $[[T0]], $[[T1]]
[mips] Fix inefficient code generation. This patch eliminates the need to emit a constant move instruction when this pattern is matched: (select (setgt a, Constant), T, F) The pattern above effectively turns into this: (conditional-move (setlt a, Constant + 1), F, T) llvm-svn: 176384
2013-03-02 05:52:08 +08:00
define i64 @slti64_2(i64 %a) {
entry:
%cmp = icmp sgt i64 %a, -32769
%conv = select i1 %cmp, i64 3, i64 4
ret i64 %conv
}
[mips][mips64r6] c.cond.fmt, mov[fntz], and mov[fntz].[ds] are not available on MIPS32r6/MIPS64r6 Summary: c.cond.fmt has been replaced by cmp.cond.fmt. Where c.cond.fmt wrote to dedicated condition registers, cmp.cond.fmt writes 1 or 0 to normal FGR's (like the GPR comparisons). mov[fntz] have been replaced by seleqz and selnez. These instructions conditionally zero a register based on a bool in a GPR. The results can then be or'd together to act as a select without, for example, requiring a third register read port. mov[fntz].[ds] have been replaced with sel.[ds] MIPS64r6 currently generates unnecessary sign-extensions for most selects. This is because the result of a SETCC is currently an i32. Bits 32-63 are undefined in i32 and the behaviour of seleqz/selnez would otherwise depend on undefined bits. Later, we will fix this by making the result of SETCC an i64 on MIPS64 targets. Depends on D3958 Reviewers: jkolek, vmedic, zoran.jovanovic Reviewed By: vmedic, zoran.jovanovic Differential Revision: http://reviews.llvm.org/D4003 llvm-svn: 210777
2014-06-12 21:39:06 +08:00
; ALL-LABEL: slti64_3:
; FIXME: The 32-bit versions of this test are too complicated to reasonably
; match at the moment. They do show some missing optimizations though
; such as:
; (movz $a, $b, (neg $c)) -> (movn $a, $b, $c)
; 64-CMOV-DAG: daddiu $[[I5:[0-9]+]], $zero, 5
; 64-CMOV-DAG: daddiu $[[I4:2]], $zero, 4
; 64-CMOV-DAG: daddiu $[[R1:[0-9]+]], ${{[0-9]+}}, 32766
; 64-CMOV-DAG: slt $[[R0:[0-9]+]], $[[R1]], $4
; 64-CMOV-DAG: movn $[[I4]], $[[I5]], $[[R0]]
; 64-CMP-DAG: daddiu $[[I5:[0-9]+]], $zero, 5
; 64-CMP-DAG: daddiu $[[I4:2]], $zero, 4
; 64-CMP-DAG: daddiu $[[R1:[0-9]+]], ${{[0-9]+}}, 32766
; 64-CMP-DAG: slt $[[R0:[0-9]+]], $[[R1]], $4
; FIXME: We can do better than this by using selccz to choose between -0 and -2
; 64-CMP-DAG: selnez $[[T0:[0-9]+]], $[[I5]], $[[R0]]
; 64-CMP-DAG: seleqz $[[T1:[0-9]+]], $[[I4]], $[[R0]]
; 64-CMP-DAG: or $2, $[[T0]], $[[T1]]
[mips] Fix inefficient code generation. This patch eliminates the need to emit a constant move instruction when this pattern is matched: (select (setgt a, Constant), T, F) The pattern above effectively turns into this: (conditional-move (setlt a, Constant + 1), F, T) llvm-svn: 176384
2013-03-02 05:52:08 +08:00
define i64 @slti64_3(i64 %a) {
entry:
%cmp = icmp sgt i64 %a, -32770
[mips][mips64r6] c.cond.fmt, mov[fntz], and mov[fntz].[ds] are not available on MIPS32r6/MIPS64r6 Summary: c.cond.fmt has been replaced by cmp.cond.fmt. Where c.cond.fmt wrote to dedicated condition registers, cmp.cond.fmt writes 1 or 0 to normal FGR's (like the GPR comparisons). mov[fntz] have been replaced by seleqz and selnez. These instructions conditionally zero a register based on a bool in a GPR. The results can then be or'd together to act as a select without, for example, requiring a third register read port. mov[fntz].[ds] have been replaced with sel.[ds] MIPS64r6 currently generates unnecessary sign-extensions for most selects. This is because the result of a SETCC is currently an i32. Bits 32-63 are undefined in i32 and the behaviour of seleqz/selnez would otherwise depend on undefined bits. Later, we will fix this by making the result of SETCC an i64 on MIPS64 targets. Depends on D3958 Reviewers: jkolek, vmedic, zoran.jovanovic Reviewed By: vmedic, zoran.jovanovic Differential Revision: http://reviews.llvm.org/D4003 llvm-svn: 210777
2014-06-12 21:39:06 +08:00
%conv = select i1 %cmp, i64 5, i64 4
[mips] Fix inefficient code generation. This patch eliminates the need to emit a constant move instruction when this pattern is matched: (select (setgt a, Constant), T, F) The pattern above effectively turns into this: (conditional-move (setlt a, Constant + 1), F, T) llvm-svn: 176384
2013-03-02 05:52:08 +08:00
ret i64 %conv
}
; sltiu instructions.
[mips][mips64r6] c.cond.fmt, mov[fntz], and mov[fntz].[ds] are not available on MIPS32r6/MIPS64r6 Summary: c.cond.fmt has been replaced by cmp.cond.fmt. Where c.cond.fmt wrote to dedicated condition registers, cmp.cond.fmt writes 1 or 0 to normal FGR's (like the GPR comparisons). mov[fntz] have been replaced by seleqz and selnez. These instructions conditionally zero a register based on a bool in a GPR. The results can then be or'd together to act as a select without, for example, requiring a third register read port. mov[fntz].[ds] have been replaced with sel.[ds] MIPS64r6 currently generates unnecessary sign-extensions for most selects. This is because the result of a SETCC is currently an i32. Bits 32-63 are undefined in i32 and the behaviour of seleqz/selnez would otherwise depend on undefined bits. Later, we will fix this by making the result of SETCC an i64 on MIPS64 targets. Depends on D3958 Reviewers: jkolek, vmedic, zoran.jovanovic Reviewed By: vmedic, zoran.jovanovic Differential Revision: http://reviews.llvm.org/D4003 llvm-svn: 210777
2014-06-12 21:39:06 +08:00
; ALL-LABEL: sltiu0:
; 32-CMOV-DAG: addiu $[[I3:[0-9]+]], $zero, 3
; 32-CMOV-DAG: addiu $[[I5:2]], $zero, 5
; 32-CMOV-DAG: sltiu $[[R0:[0-9]+]], $4, 32767
; 32-CMOV-DAG: movz $[[I5]], $[[I3]], $[[R0]]
; 32-CMP-DAG: addiu $[[I3:[0-9]+]], $zero, 3
; 32-CMP-DAG: addiu $[[I5:[0-9]+]], $zero, 5
; 32-CMP-DAG: sltiu $[[R0:[0-9]+]], $4, 32767
; FIXME: We can do better than this by using selccz to choose between +0 and +2
[mips][mips64r6] Correct select patterns that have the condition or true/false values backwards Summary: This bug caused SingleSource/Regression/C/uint64_to_float and SingleSource/UnitTests/2002-05-02-CastTest3 to fail (among others). Differential Revision: http://reviews.llvm.org/D4388 llvm-svn: 212608
2014-07-09 18:47:26 +08:00
; 32-CMP-DAG: seleqz $[[T0:[0-9]+]], $[[I3]], $[[R0]]
; 32-CMP-DAG: selnez $[[T1:[0-9]+]], $[[I5]], $[[R0]]
[mips][mips64r6] c.cond.fmt, mov[fntz], and mov[fntz].[ds] are not available on MIPS32r6/MIPS64r6 Summary: c.cond.fmt has been replaced by cmp.cond.fmt. Where c.cond.fmt wrote to dedicated condition registers, cmp.cond.fmt writes 1 or 0 to normal FGR's (like the GPR comparisons). mov[fntz] have been replaced by seleqz and selnez. These instructions conditionally zero a register based on a bool in a GPR. The results can then be or'd together to act as a select without, for example, requiring a third register read port. mov[fntz].[ds] have been replaced with sel.[ds] MIPS64r6 currently generates unnecessary sign-extensions for most selects. This is because the result of a SETCC is currently an i32. Bits 32-63 are undefined in i32 and the behaviour of seleqz/selnez would otherwise depend on undefined bits. Later, we will fix this by making the result of SETCC an i64 on MIPS64 targets. Depends on D3958 Reviewers: jkolek, vmedic, zoran.jovanovic Reviewed By: vmedic, zoran.jovanovic Differential Revision: http://reviews.llvm.org/D4003 llvm-svn: 210777
2014-06-12 21:39:06 +08:00
; 32-CMP-DAG: or $2, $[[T0]], $[[T1]]
; 64-CMOV-DAG: addiu $[[I3:[0-9]+]], $zero, 3
; 64-CMOV-DAG: addiu $[[I5:2]], $zero, 5
; 64-CMOV-DAG: sltiu $[[R0:[0-9]+]], $4, 32767
; 64-CMOV-DAG: movz $[[I5]], $[[I3]], $[[R0]]
; 64-CMP-DAG: addiu $[[I3:[0-9]+]], $zero, 3
; 64-CMP-DAG: addiu $[[I5:[0-9]+]], $zero, 5
; 64-CMP-DAG: sltiu $[[R0:[0-9]+]], $4, 32767
; FIXME: We can do better than this by using selccz to choose between +0 and +2
[mips][mips64r6] Correct select patterns that have the condition or true/false values backwards Summary: This bug caused SingleSource/Regression/C/uint64_to_float and SingleSource/UnitTests/2002-05-02-CastTest3 to fail (among others). Differential Revision: http://reviews.llvm.org/D4388 llvm-svn: 212608
2014-07-09 18:47:26 +08:00
; 64-CMP-DAG: seleqz $[[T0:[0-9]+]], $[[I3]], $[[R0]]
; 64-CMP-DAG: selnez $[[T1:[0-9]+]], $[[I5]], $[[R0]]
[mips][mips64r6] c.cond.fmt, mov[fntz], and mov[fntz].[ds] are not available on MIPS32r6/MIPS64r6 Summary: c.cond.fmt has been replaced by cmp.cond.fmt. Where c.cond.fmt wrote to dedicated condition registers, cmp.cond.fmt writes 1 or 0 to normal FGR's (like the GPR comparisons). mov[fntz] have been replaced by seleqz and selnez. These instructions conditionally zero a register based on a bool in a GPR. The results can then be or'd together to act as a select without, for example, requiring a third register read port. mov[fntz].[ds] have been replaced with sel.[ds] MIPS64r6 currently generates unnecessary sign-extensions for most selects. This is because the result of a SETCC is currently an i32. Bits 32-63 are undefined in i32 and the behaviour of seleqz/selnez would otherwise depend on undefined bits. Later, we will fix this by making the result of SETCC an i64 on MIPS64 targets. Depends on D3958 Reviewers: jkolek, vmedic, zoran.jovanovic Reviewed By: vmedic, zoran.jovanovic Differential Revision: http://reviews.llvm.org/D4003 llvm-svn: 210777
2014-06-12 21:39:06 +08:00
; 64-CMP-DAG: or $2, $[[T0]], $[[T1]]
[mips] Fix inefficient code generation. This patch eliminates the need to emit a constant move instruction when this pattern is matched: (select (setgt a, Constant), T, F) The pattern above effectively turns into this: (conditional-move (setlt a, Constant + 1), F, T) llvm-svn: 176384
2013-03-02 05:52:08 +08:00
[mips] Promote i32 arguments to i64 for the N32/N64 ABI and fix <64-bit structs... Summary: ... and after all that refactoring, it's possible to distinguish softfloat floating point values from integers so this patch no longer breaks softfloat to do it. Remove direct handling of i32's in the N32/N64 ABI by promoting them to i64. This more closely reflects the ABI documentation and also fixes problems with stack arguments on big-endian targets. We now rely on signext/zeroext annotations (already generated by clang) and the Assert[SZ]ext nodes to avoid the introduction of unnecessary sign/zero extends. It was not possible to convert three tests to use signext/zeroext. These tests are bswap.ll, ctlz-v.ll, ctlz-v.ll. It's not possible to put signext on a vector type so we just accept the sign extends here for now. These tests don't pass the vectors the same way clang does (clang puts multiple elements in the same argument, these map 1 element to 1 argument) so we don't need to worry too much about it. With this patch, all known N32/N64 bugs should be fixed and we now pass the first 10,000 tests generated by ABITest.py. Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D6117 llvm-svn: 221534
2014-11-08 00:54:21 +08:00
define i32 @sltiu0(i32 signext %a) {
[mips] Fix inefficient code generation. This patch eliminates the need to emit a constant move instruction when this pattern is matched: (select (setgt a, Constant), T, F) The pattern above effectively turns into this: (conditional-move (setlt a, Constant + 1), F, T) llvm-svn: 176384
2013-03-02 05:52:08 +08:00
entry:
%cmp = icmp ugt i32 %a, 32766
[mips] Small code generation improvement for conditional operator (select) in case the operands are constants and its difference is |1|. It should be possible in those cases to rematerialize the result using MIPS's slt and similar instructions. The small update to some of the tests in cmov.ll, sel1c.ll and sel2c.ll was needed otherwise the optimization implemented in this patch would have been triggered (difference between the operands was 1) and that would have changed the semantic of the tests. llvm-svn: 196498
2013-12-05 20:07:05 +08:00
%cond = select i1 %cmp, i32 3, i32 5
[mips] Fix inefficient code generation. This patch eliminates the need to emit a constant move instruction when this pattern is matched: (select (setgt a, Constant), T, F) The pattern above effectively turns into this: (conditional-move (setlt a, Constant + 1), F, T) llvm-svn: 176384
2013-03-02 05:52:08 +08:00
ret i32 %cond
}
[mips][mips64r6] c.cond.fmt, mov[fntz], and mov[fntz].[ds] are not available on MIPS32r6/MIPS64r6 Summary: c.cond.fmt has been replaced by cmp.cond.fmt. Where c.cond.fmt wrote to dedicated condition registers, cmp.cond.fmt writes 1 or 0 to normal FGR's (like the GPR comparisons). mov[fntz] have been replaced by seleqz and selnez. These instructions conditionally zero a register based on a bool in a GPR. The results can then be or'd together to act as a select without, for example, requiring a third register read port. mov[fntz].[ds] have been replaced with sel.[ds] MIPS64r6 currently generates unnecessary sign-extensions for most selects. This is because the result of a SETCC is currently an i32. Bits 32-63 are undefined in i32 and the behaviour of seleqz/selnez would otherwise depend on undefined bits. Later, we will fix this by making the result of SETCC an i64 on MIPS64 targets. Depends on D3958 Reviewers: jkolek, vmedic, zoran.jovanovic Reviewed By: vmedic, zoran.jovanovic Differential Revision: http://reviews.llvm.org/D4003 llvm-svn: 210777
2014-06-12 21:39:06 +08:00
; ALL-LABEL: sltiu1:
; 32-CMOV-DAG: addiu $[[I7:[0-9]+]], $zero, 7
; 32-CMOV-DAG: addiu $[[I5:2]], $zero, 5
; 32-CMOV-DAG: addiu $[[R1:[0-9]+]], $zero, 32767
; 32-CMOV-DAG: sltu $[[R0:[0-9]+]], $[[R1]], $4
; 32-CMOV-DAG: movn $[[I5]], $[[I7]], $[[R0]]
; 32-CMP-DAG: addiu $[[I7:[0-9]+]], $zero, 7
; 32-CMP-DAG: addiu $[[I5:[0-9]+]], $zero, 5
; 32-CMP-DAG: addiu $[[I32767:[0-9]+]], $zero, 32767
; 32-CMP-DAG: sltu $[[R0:[0-9]+]], $[[I32767]], $4
; FIXME: We can do better than this by using selccz to choose between -0 and -2
; 32-CMP-DAG: selnez $[[T0:[0-9]+]], $[[I7]], $[[R0]]
; 32-CMP-DAG: seleqz $[[T1:[0-9]+]], $[[I5]], $[[R0]]
; 32-CMP-DAG: or $2, $[[T0]], $[[T1]]
; 64-CMOV-DAG: addiu $[[I7:[0-9]+]], $zero, 7
; 64-CMOV-DAG: addiu $[[I5:2]], $zero, 5
; 64-CMOV-DAG: addiu $[[R1:[0-9]+]], $zero, 32767
; 64-CMOV-DAG: sltu $[[R0:[0-9]+]], $[[R1]], $4
; 64-CMOV-DAG: movn $[[I5]], $[[I7]], $[[R0]]
; 64-CMP-DAG: addiu $[[I7:[0-9]+]], $zero, 7
; 64-CMP-DAG: addiu $[[I5:2]], $zero, 5
; 64-CMP-DAG: addiu $[[R1:[0-9]+]], $zero, 32767
; 64-CMP-DAG: sltu $[[R0:[0-9]+]], $[[R1]], $4
; FIXME: We can do better than this by using selccz to choose between -0 and -2
; 64-CMP-DAG: selnez $[[T0:[0-9]+]], $[[I7]], $[[R0]]
; 64-CMP-DAG: seleqz $[[T1:[0-9]+]], $[[I5]], $[[R0]]
; 64-CMP-DAG: or $2, $[[T0]], $[[T1]]
[mips] Fix inefficient code generation. This patch eliminates the need to emit a constant move instruction when this pattern is matched: (select (setgt a, Constant), T, F) The pattern above effectively turns into this: (conditional-move (setlt a, Constant + 1), F, T) llvm-svn: 176384
2013-03-02 05:52:08 +08:00
[mips] Promote i32 arguments to i64 for the N32/N64 ABI and fix <64-bit structs... Summary: ... and after all that refactoring, it's possible to distinguish softfloat floating point values from integers so this patch no longer breaks softfloat to do it. Remove direct handling of i32's in the N32/N64 ABI by promoting them to i64. This more closely reflects the ABI documentation and also fixes problems with stack arguments on big-endian targets. We now rely on signext/zeroext annotations (already generated by clang) and the Assert[SZ]ext nodes to avoid the introduction of unnecessary sign/zero extends. It was not possible to convert three tests to use signext/zeroext. These tests are bswap.ll, ctlz-v.ll, ctlz-v.ll. It's not possible to put signext on a vector type so we just accept the sign extends here for now. These tests don't pass the vectors the same way clang does (clang puts multiple elements in the same argument, these map 1 element to 1 argument) so we don't need to worry too much about it. With this patch, all known N32/N64 bugs should be fixed and we now pass the first 10,000 tests generated by ABITest.py. Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D6117 llvm-svn: 221534
2014-11-08 00:54:21 +08:00
define i32 @sltiu1(i32 signext %a) {
[mips] Fix inefficient code generation. This patch eliminates the need to emit a constant move instruction when this pattern is matched: (select (setgt a, Constant), T, F) The pattern above effectively turns into this: (conditional-move (setlt a, Constant + 1), F, T) llvm-svn: 176384
2013-03-02 05:52:08 +08:00
entry:
%cmp = icmp ugt i32 %a, 32767
[mips][mips64r6] c.cond.fmt, mov[fntz], and mov[fntz].[ds] are not available on MIPS32r6/MIPS64r6 Summary: c.cond.fmt has been replaced by cmp.cond.fmt. Where c.cond.fmt wrote to dedicated condition registers, cmp.cond.fmt writes 1 or 0 to normal FGR's (like the GPR comparisons). mov[fntz] have been replaced by seleqz and selnez. These instructions conditionally zero a register based on a bool in a GPR. The results can then be or'd together to act as a select without, for example, requiring a third register read port. mov[fntz].[ds] have been replaced with sel.[ds] MIPS64r6 currently generates unnecessary sign-extensions for most selects. This is because the result of a SETCC is currently an i32. Bits 32-63 are undefined in i32 and the behaviour of seleqz/selnez would otherwise depend on undefined bits. Later, we will fix this by making the result of SETCC an i64 on MIPS64 targets. Depends on D3958 Reviewers: jkolek, vmedic, zoran.jovanovic Reviewed By: vmedic, zoran.jovanovic Differential Revision: http://reviews.llvm.org/D4003 llvm-svn: 210777
2014-06-12 21:39:06 +08:00
%cond = select i1 %cmp, i32 7, i32 5
[mips] Fix inefficient code generation. This patch eliminates the need to emit a constant move instruction when this pattern is matched: (select (setgt a, Constant), T, F) The pattern above effectively turns into this: (conditional-move (setlt a, Constant + 1), F, T) llvm-svn: 176384
2013-03-02 05:52:08 +08:00
ret i32 %cond
}
[mips][mips64r6] c.cond.fmt, mov[fntz], and mov[fntz].[ds] are not available on MIPS32r6/MIPS64r6 Summary: c.cond.fmt has been replaced by cmp.cond.fmt. Where c.cond.fmt wrote to dedicated condition registers, cmp.cond.fmt writes 1 or 0 to normal FGR's (like the GPR comparisons). mov[fntz] have been replaced by seleqz and selnez. These instructions conditionally zero a register based on a bool in a GPR. The results can then be or'd together to act as a select without, for example, requiring a third register read port. mov[fntz].[ds] have been replaced with sel.[ds] MIPS64r6 currently generates unnecessary sign-extensions for most selects. This is because the result of a SETCC is currently an i32. Bits 32-63 are undefined in i32 and the behaviour of seleqz/selnez would otherwise depend on undefined bits. Later, we will fix this by making the result of SETCC an i64 on MIPS64 targets. Depends on D3958 Reviewers: jkolek, vmedic, zoran.jovanovic Reviewed By: vmedic, zoran.jovanovic Differential Revision: http://reviews.llvm.org/D4003 llvm-svn: 210777
2014-06-12 21:39:06 +08:00
; ALL-LABEL: sltiu2:
; 32-CMOV-DAG: addiu $[[I3:[0-9]+]], $zero, 3
; 32-CMOV-DAG: addiu $[[I5:2]], $zero, 5
; 32-CMOV-DAG: sltiu $[[R0:[0-9]+]], $4, -32768
; 32-CMOV-DAG: movz $[[I5]], $[[I3]], $[[R0]]
; 32-CMP-DAG: addiu $[[I3:[0-9]+]], $zero, 3
; 32-CMP-DAG: addiu $[[I5:[0-9]+]], $zero, 5
; 32-CMP-DAG: sltiu $[[R0:[0-9]+]], $4, -32768
; FIXME: We can do better than this by using selccz to choose between +0 and +2
[mips][mips64r6] Correct select patterns that have the condition or true/false values backwards Summary: This bug caused SingleSource/Regression/C/uint64_to_float and SingleSource/UnitTests/2002-05-02-CastTest3 to fail (among others). Differential Revision: http://reviews.llvm.org/D4388 llvm-svn: 212608
2014-07-09 18:47:26 +08:00
; 32-CMP-DAG: seleqz $[[T0:[0-9]+]], $[[I3]], $[[R0]]
; 32-CMP-DAG: selnez $[[T1:[0-9]+]], $[[I5]], $[[R0]]
[mips][mips64r6] c.cond.fmt, mov[fntz], and mov[fntz].[ds] are not available on MIPS32r6/MIPS64r6 Summary: c.cond.fmt has been replaced by cmp.cond.fmt. Where c.cond.fmt wrote to dedicated condition registers, cmp.cond.fmt writes 1 or 0 to normal FGR's (like the GPR comparisons). mov[fntz] have been replaced by seleqz and selnez. These instructions conditionally zero a register based on a bool in a GPR. The results can then be or'd together to act as a select without, for example, requiring a third register read port. mov[fntz].[ds] have been replaced with sel.[ds] MIPS64r6 currently generates unnecessary sign-extensions for most selects. This is because the result of a SETCC is currently an i32. Bits 32-63 are undefined in i32 and the behaviour of seleqz/selnez would otherwise depend on undefined bits. Later, we will fix this by making the result of SETCC an i64 on MIPS64 targets. Depends on D3958 Reviewers: jkolek, vmedic, zoran.jovanovic Reviewed By: vmedic, zoran.jovanovic Differential Revision: http://reviews.llvm.org/D4003 llvm-svn: 210777
2014-06-12 21:39:06 +08:00
; 32-CMP-DAG: or $2, $[[T0]], $[[T1]]
; 64-CMOV-DAG: addiu $[[I3:[0-9]+]], $zero, 3
; 64-CMOV-DAG: addiu $[[I5:2]], $zero, 5
; 64-CMOV-DAG: sltiu $[[R0:[0-9]+]], $4, -32768
; 64-CMOV-DAG: movz $[[I5]], $[[I3]], $[[R0]]
; 64-CMP-DAG: addiu $[[I3:[0-9]+]], $zero, 3
; 64-CMP-DAG: addiu $[[I5:[0-9]+]], $zero, 5
; 64-CMP-DAG: sltiu $[[R0:[0-9]+]], $4, -32768
; FIXME: We can do better than this by using selccz to choose between +0 and +2
[mips][mips64r6] Correct select patterns that have the condition or true/false values backwards Summary: This bug caused SingleSource/Regression/C/uint64_to_float and SingleSource/UnitTests/2002-05-02-CastTest3 to fail (among others). Differential Revision: http://reviews.llvm.org/D4388 llvm-svn: 212608
2014-07-09 18:47:26 +08:00
; 64-CMP-DAG: seleqz $[[T0:[0-9]+]], $[[I3]], $[[R0]]
; 64-CMP-DAG: selnez $[[T1:[0-9]+]], $[[I5]], $[[R0]]
[mips][mips64r6] c.cond.fmt, mov[fntz], and mov[fntz].[ds] are not available on MIPS32r6/MIPS64r6 Summary: c.cond.fmt has been replaced by cmp.cond.fmt. Where c.cond.fmt wrote to dedicated condition registers, cmp.cond.fmt writes 1 or 0 to normal FGR's (like the GPR comparisons). mov[fntz] have been replaced by seleqz and selnez. These instructions conditionally zero a register based on a bool in a GPR. The results can then be or'd together to act as a select without, for example, requiring a third register read port. mov[fntz].[ds] have been replaced with sel.[ds] MIPS64r6 currently generates unnecessary sign-extensions for most selects. This is because the result of a SETCC is currently an i32. Bits 32-63 are undefined in i32 and the behaviour of seleqz/selnez would otherwise depend on undefined bits. Later, we will fix this by making the result of SETCC an i64 on MIPS64 targets. Depends on D3958 Reviewers: jkolek, vmedic, zoran.jovanovic Reviewed By: vmedic, zoran.jovanovic Differential Revision: http://reviews.llvm.org/D4003 llvm-svn: 210777
2014-06-12 21:39:06 +08:00
; 64-CMP-DAG: or $2, $[[T0]], $[[T1]]
[mips] Fix inefficient code generation. This patch eliminates the need to emit a constant move instruction when this pattern is matched: (select (setgt a, Constant), T, F) The pattern above effectively turns into this: (conditional-move (setlt a, Constant + 1), F, T) llvm-svn: 176384
2013-03-02 05:52:08 +08:00
[mips] Promote i32 arguments to i64 for the N32/N64 ABI and fix <64-bit structs... Summary: ... and after all that refactoring, it's possible to distinguish softfloat floating point values from integers so this patch no longer breaks softfloat to do it. Remove direct handling of i32's in the N32/N64 ABI by promoting them to i64. This more closely reflects the ABI documentation and also fixes problems with stack arguments on big-endian targets. We now rely on signext/zeroext annotations (already generated by clang) and the Assert[SZ]ext nodes to avoid the introduction of unnecessary sign/zero extends. It was not possible to convert three tests to use signext/zeroext. These tests are bswap.ll, ctlz-v.ll, ctlz-v.ll. It's not possible to put signext on a vector type so we just accept the sign extends here for now. These tests don't pass the vectors the same way clang does (clang puts multiple elements in the same argument, these map 1 element to 1 argument) so we don't need to worry too much about it. With this patch, all known N32/N64 bugs should be fixed and we now pass the first 10,000 tests generated by ABITest.py. Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D6117 llvm-svn: 221534
2014-11-08 00:54:21 +08:00
define i32 @sltiu2(i32 signext %a) {
[mips] Fix inefficient code generation. This patch eliminates the need to emit a constant move instruction when this pattern is matched: (select (setgt a, Constant), T, F) The pattern above effectively turns into this: (conditional-move (setlt a, Constant + 1), F, T) llvm-svn: 176384
2013-03-02 05:52:08 +08:00
entry:
%cmp = icmp ugt i32 %a, -32769
[mips] Small code generation improvement for conditional operator (select) in case the operands are constants and its difference is |1|. It should be possible in those cases to rematerialize the result using MIPS's slt and similar instructions. The small update to some of the tests in cmov.ll, sel1c.ll and sel2c.ll was needed otherwise the optimization implemented in this patch would have been triggered (difference between the operands was 1) and that would have changed the semantic of the tests. llvm-svn: 196498
2013-12-05 20:07:05 +08:00
%cond = select i1 %cmp, i32 3, i32 5
[mips] Fix inefficient code generation. This patch eliminates the need to emit a constant move instruction when this pattern is matched: (select (setgt a, Constant), T, F) The pattern above effectively turns into this: (conditional-move (setlt a, Constant + 1), F, T) llvm-svn: 176384
2013-03-02 05:52:08 +08:00
ret i32 %cond
}
[mips][mips64r6] c.cond.fmt, mov[fntz], and mov[fntz].[ds] are not available on MIPS32r6/MIPS64r6 Summary: c.cond.fmt has been replaced by cmp.cond.fmt. Where c.cond.fmt wrote to dedicated condition registers, cmp.cond.fmt writes 1 or 0 to normal FGR's (like the GPR comparisons). mov[fntz] have been replaced by seleqz and selnez. These instructions conditionally zero a register based on a bool in a GPR. The results can then be or'd together to act as a select without, for example, requiring a third register read port. mov[fntz].[ds] have been replaced with sel.[ds] MIPS64r6 currently generates unnecessary sign-extensions for most selects. This is because the result of a SETCC is currently an i32. Bits 32-63 are undefined in i32 and the behaviour of seleqz/selnez would otherwise depend on undefined bits. Later, we will fix this by making the result of SETCC an i64 on MIPS64 targets. Depends on D3958 Reviewers: jkolek, vmedic, zoran.jovanovic Reviewed By: vmedic, zoran.jovanovic Differential Revision: http://reviews.llvm.org/D4003 llvm-svn: 210777
2014-06-12 21:39:06 +08:00
; ALL-LABEL: sltiu3:
; 32-CMOV-DAG: addiu $[[I3:[0-9]+]], $zero, 3
; 32-CMOV-DAG: addiu $[[I5:2]], $zero, 5
; 32-CMOV-DAG: lui $[[R1:[0-9]+]], 65535
; 32-CMOV-DAG: ori $[[R1]], $[[R1]], 32766
; 32-CMOV-DAG: sltu $[[R0:[0-9]+]], $[[R1]], $4
; 32-CMOV-DAG: movn $[[I5]], $[[I3]], $[[R0]]
; 32-CMP-DAG: addiu $[[I3:[0-9]+]], $zero, 3
; 32-CMP-DAG: addiu $[[I5:[0-9]+]], $zero, 5
; 32-CMP-DAG: lui $[[IMM:[0-9]+]], 65535
; 32-CMP-DAG: ori $[[IMM]], $[[IMM]], 32766
; 32-CMP-DAG: sltu $[[R0:[0-9]+]], $[[I32767]], $4
; FIXME: We can do better than this by using selccz to choose between -0 and -2
; 32-CMP-DAG: selnez $[[T0:[0-9]+]], $[[I3]], $[[R0]]
; 32-CMP-DAG: seleqz $[[T1:[0-9]+]], $[[I5]], $[[R0]]
; 32-CMP-DAG: or $2, $[[T0]], $[[T1]]
; 64-CMOV-DAG: addiu $[[I3:[0-9]+]], $zero, 3
; 64-CMOV-DAG: addiu $[[I5:2]], $zero, 5
; 64-CMOV-DAG: lui $[[R1:[0-9]+]], 65535
; 64-CMOV-DAG: ori $[[R1]], $[[R1]], 32766
; 64-CMOV-DAG: sltu $[[R0:[0-9]+]], $[[R1]], $4
; 64-CMOV-DAG: movn $[[I5]], $[[I3]], $[[R0]]
; 64-CMP-DAG: addiu $[[I3:[0-9]+]], $zero, 3
; 64-CMP-DAG: addiu $[[I5:2]], $zero, 5
; 64-CMP-DAG: lui $[[IMM:[0-9]+]], 65535
; 64-CMP-DAG: ori $[[IMM]], $[[IMM]], 32766
; 64-CMP-DAG: sltu $[[R0:[0-9]+]], $[[IMM]], $4
; FIXME: We can do better than this by using selccz to choose between -0 and -2
; 64-CMP-DAG: selnez $[[T0:[0-9]+]], $[[I3]], $[[R0]]
; 64-CMP-DAG: seleqz $[[T1:[0-9]+]], $[[I5]], $[[R0]]
; 64-CMP-DAG: or $2, $[[T0]], $[[T1]]
[mips] Fix inefficient code generation. This patch eliminates the need to emit a constant move instruction when this pattern is matched: (select (setgt a, Constant), T, F) The pattern above effectively turns into this: (conditional-move (setlt a, Constant + 1), F, T) llvm-svn: 176384
2013-03-02 05:52:08 +08:00
[mips] Promote i32 arguments to i64 for the N32/N64 ABI and fix <64-bit structs... Summary: ... and after all that refactoring, it's possible to distinguish softfloat floating point values from integers so this patch no longer breaks softfloat to do it. Remove direct handling of i32's in the N32/N64 ABI by promoting them to i64. This more closely reflects the ABI documentation and also fixes problems with stack arguments on big-endian targets. We now rely on signext/zeroext annotations (already generated by clang) and the Assert[SZ]ext nodes to avoid the introduction of unnecessary sign/zero extends. It was not possible to convert three tests to use signext/zeroext. These tests are bswap.ll, ctlz-v.ll, ctlz-v.ll. It's not possible to put signext on a vector type so we just accept the sign extends here for now. These tests don't pass the vectors the same way clang does (clang puts multiple elements in the same argument, these map 1 element to 1 argument) so we don't need to worry too much about it. With this patch, all known N32/N64 bugs should be fixed and we now pass the first 10,000 tests generated by ABITest.py. Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D6117 llvm-svn: 221534
2014-11-08 00:54:21 +08:00
define i32 @sltiu3(i32 signext %a) {
[mips] Fix inefficient code generation. This patch eliminates the need to emit a constant move instruction when this pattern is matched: (select (setgt a, Constant), T, F) The pattern above effectively turns into this: (conditional-move (setlt a, Constant + 1), F, T) llvm-svn: 176384
2013-03-02 05:52:08 +08:00
entry:
%cmp = icmp ugt i32 %a, -32770
[mips] Small code generation improvement for conditional operator (select) in case the operands are constants and its difference is |1|. It should be possible in those cases to rematerialize the result using MIPS's slt and similar instructions. The small update to some of the tests in cmov.ll, sel1c.ll and sel2c.ll was needed otherwise the optimization implemented in this patch would have been triggered (difference between the operands was 1) and that would have changed the semantic of the tests. llvm-svn: 196498
2013-12-05 20:07:05 +08:00
%cond = select i1 %cmp, i32 3, i32 5
[mips] Fix inefficient code generation. This patch eliminates the need to emit a constant move instruction when this pattern is matched: (select (setgt a, Constant), T, F) The pattern above effectively turns into this: (conditional-move (setlt a, Constant + 1), F, T) llvm-svn: 176384
2013-03-02 05:52:08 +08:00
ret i32 %cond
}
[mips] Small code generation improvement for conditional operator (select) in case the operands are constants and its difference is |1|. It should be possible in those cases to rematerialize the result using MIPS's slt and similar instructions. The small update to some of the tests in cmov.ll, sel1c.ll and sel2c.ll was needed otherwise the optimization implemented in this patch would have been triggered (difference between the operands was 1) and that would have changed the semantic of the tests. llvm-svn: 196498
2013-12-05 20:07:05 +08:00
; Check if
; (select (setxx a, N), x, x-1) or
; (select (setxx a, N), x-1, x)
; doesn't generate conditional moves
; for constant operands whose difference is |1|
[mips] Promote i32 arguments to i64 for the N32/N64 ABI and fix <64-bit structs... Summary: ... and after all that refactoring, it's possible to distinguish softfloat floating point values from integers so this patch no longer breaks softfloat to do it. Remove direct handling of i32's in the N32/N64 ABI by promoting them to i64. This more closely reflects the ABI documentation and also fixes problems with stack arguments on big-endian targets. We now rely on signext/zeroext annotations (already generated by clang) and the Assert[SZ]ext nodes to avoid the introduction of unnecessary sign/zero extends. It was not possible to convert three tests to use signext/zeroext. These tests are bswap.ll, ctlz-v.ll, ctlz-v.ll. It's not possible to put signext on a vector type so we just accept the sign extends here for now. These tests don't pass the vectors the same way clang does (clang puts multiple elements in the same argument, these map 1 element to 1 argument) so we don't need to worry too much about it. With this patch, all known N32/N64 bugs should be fixed and we now pass the first 10,000 tests generated by ABITest.py. Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D6117 llvm-svn: 221534
2014-11-08 00:54:21 +08:00
define i32 @slti4(i32 signext %a) nounwind readnone {
[mips] Small code generation improvement for conditional operator (select) in case the operands are constants and its difference is |1|. It should be possible in those cases to rematerialize the result using MIPS's slt and similar instructions. The small update to some of the tests in cmov.ll, sel1c.ll and sel2c.ll was needed otherwise the optimization implemented in this patch would have been triggered (difference between the operands was 1) and that would have changed the semantic of the tests. llvm-svn: 196498
2013-12-05 20:07:05 +08:00
%1 = icmp slt i32 %a, 7
%2 = select i1 %1, i32 4, i32 3
ret i32 %2
}
[mips][mips64r6] c.cond.fmt, mov[fntz], and mov[fntz].[ds] are not available on MIPS32r6/MIPS64r6 Summary: c.cond.fmt has been replaced by cmp.cond.fmt. Where c.cond.fmt wrote to dedicated condition registers, cmp.cond.fmt writes 1 or 0 to normal FGR's (like the GPR comparisons). mov[fntz] have been replaced by seleqz and selnez. These instructions conditionally zero a register based on a bool in a GPR. The results can then be or'd together to act as a select without, for example, requiring a third register read port. mov[fntz].[ds] have been replaced with sel.[ds] MIPS64r6 currently generates unnecessary sign-extensions for most selects. This is because the result of a SETCC is currently an i32. Bits 32-63 are undefined in i32 and the behaviour of seleqz/selnez would otherwise depend on undefined bits. Later, we will fix this by making the result of SETCC an i64 on MIPS64 targets. Depends on D3958 Reviewers: jkolek, vmedic, zoran.jovanovic Reviewed By: vmedic, zoran.jovanovic Differential Revision: http://reviews.llvm.org/D4003 llvm-svn: 210777
2014-06-12 21:39:06 +08:00
; ALL-LABEL: slti4:
; 32-CMOV-DAG: slti [[R1:\$[0-9]+]], $4, 7
; 32-CMOV-DAG: addiu $2, [[R1]], 3
; 32-CMOV-NOT: movn
; 32-CMP-DAG: slti [[R1:\$[0-9]+]], $4, 7
; 32-CMP-DAG: addiu $2, [[R1]], 3
; 32-CMP-NOT: seleqz
; 32-CMP-NOT: selnez
; 64-CMOV-DAG: slti [[R1:\$[0-9]+]], $4, 7
; 64-CMOV-DAG: addiu $2, [[R1]], 3
; 64-CMOV-NOT: movn
; 64-CMP-DAG: slti [[R1:\$[0-9]+]], $4, 7
; 64-CMP-DAG: addiu $2, [[R1]], 3
; 64-CMP-NOT: seleqz
; 64-CMP-NOT: selnez
[mips] Small code generation improvement for conditional operator (select) in case the operands are constants and its difference is |1|. It should be possible in those cases to rematerialize the result using MIPS's slt and similar instructions. The small update to some of the tests in cmov.ll, sel1c.ll and sel2c.ll was needed otherwise the optimization implemented in this patch would have been triggered (difference between the operands was 1) and that would have changed the semantic of the tests. llvm-svn: 196498
2013-12-05 20:07:05 +08:00
[mips] Promote i32 arguments to i64 for the N32/N64 ABI and fix <64-bit structs... Summary: ... and after all that refactoring, it's possible to distinguish softfloat floating point values from integers so this patch no longer breaks softfloat to do it. Remove direct handling of i32's in the N32/N64 ABI by promoting them to i64. This more closely reflects the ABI documentation and also fixes problems with stack arguments on big-endian targets. We now rely on signext/zeroext annotations (already generated by clang) and the Assert[SZ]ext nodes to avoid the introduction of unnecessary sign/zero extends. It was not possible to convert three tests to use signext/zeroext. These tests are bswap.ll, ctlz-v.ll, ctlz-v.ll. It's not possible to put signext on a vector type so we just accept the sign extends here for now. These tests don't pass the vectors the same way clang does (clang puts multiple elements in the same argument, these map 1 element to 1 argument) so we don't need to worry too much about it. With this patch, all known N32/N64 bugs should be fixed and we now pass the first 10,000 tests generated by ABITest.py. Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D6117 llvm-svn: 221534
2014-11-08 00:54:21 +08:00
define i32 @slti5(i32 signext %a) nounwind readnone {
[mips] Small code generation improvement for conditional operator (select) in case the operands are constants and its difference is |1|. It should be possible in those cases to rematerialize the result using MIPS's slt and similar instructions. The small update to some of the tests in cmov.ll, sel1c.ll and sel2c.ll was needed otherwise the optimization implemented in this patch would have been triggered (difference between the operands was 1) and that would have changed the semantic of the tests. llvm-svn: 196498
2013-12-05 20:07:05 +08:00
%1 = icmp slt i32 %a, 7
%2 = select i1 %1, i32 -3, i32 -4
ret i32 %2
}
[mips][mips64r6] c.cond.fmt, mov[fntz], and mov[fntz].[ds] are not available on MIPS32r6/MIPS64r6 Summary: c.cond.fmt has been replaced by cmp.cond.fmt. Where c.cond.fmt wrote to dedicated condition registers, cmp.cond.fmt writes 1 or 0 to normal FGR's (like the GPR comparisons). mov[fntz] have been replaced by seleqz and selnez. These instructions conditionally zero a register based on a bool in a GPR. The results can then be or'd together to act as a select without, for example, requiring a third register read port. mov[fntz].[ds] have been replaced with sel.[ds] MIPS64r6 currently generates unnecessary sign-extensions for most selects. This is because the result of a SETCC is currently an i32. Bits 32-63 are undefined in i32 and the behaviour of seleqz/selnez would otherwise depend on undefined bits. Later, we will fix this by making the result of SETCC an i64 on MIPS64 targets. Depends on D3958 Reviewers: jkolek, vmedic, zoran.jovanovic Reviewed By: vmedic, zoran.jovanovic Differential Revision: http://reviews.llvm.org/D4003 llvm-svn: 210777
2014-06-12 21:39:06 +08:00
; ALL-LABEL: slti5:
; 32-CMOV-DAG: slti [[R1:\$[0-9]+]], $4, 7
; 32-CMOV-DAG: addiu [[R3:\$[0-9]+]], [[R2:\$[a-z0-9]+]], -4
; 32-CMOV-NOT: movn
; 32-CMP-DAG: slti [[R1:\$[0-9]+]], $4, 7
; 32-CMP-DAG: addiu [[R3:\$[0-9]+]], [[R2:\$[a-z0-9]+]], -4
; 32-CMP-NOT: seleqz
; 32-CMP-NOT: selnez
; 64-CMOV-DAG: slti [[R1:\$[0-9]+]], $4, 7
; 64-CMOV-DAG: addiu [[R3:\$[0-9]+]], [[R2:\$[a-z0-9]+]], -4
; 64-CMOV-NOT: movn
; 64-CMP-DAG: slti [[R1:\$[0-9]+]], $4, 7
; 64-CMP-DAG: addiu [[R3:\$[0-9]+]], [[R2:\$[a-z0-9]+]], -4
; 64-CMP-NOT: seleqz
; 64-CMP-NOT: selnez
[mips] Small code generation improvement for conditional operator (select) in case the operands are constants and its difference is |1|. It should be possible in those cases to rematerialize the result using MIPS's slt and similar instructions. The small update to some of the tests in cmov.ll, sel1c.ll and sel2c.ll was needed otherwise the optimization implemented in this patch would have been triggered (difference between the operands was 1) and that would have changed the semantic of the tests. llvm-svn: 196498
2013-12-05 20:07:05 +08:00
[mips] Promote i32 arguments to i64 for the N32/N64 ABI and fix <64-bit structs... Summary: ... and after all that refactoring, it's possible to distinguish softfloat floating point values from integers so this patch no longer breaks softfloat to do it. Remove direct handling of i32's in the N32/N64 ABI by promoting them to i64. This more closely reflects the ABI documentation and also fixes problems with stack arguments on big-endian targets. We now rely on signext/zeroext annotations (already generated by clang) and the Assert[SZ]ext nodes to avoid the introduction of unnecessary sign/zero extends. It was not possible to convert three tests to use signext/zeroext. These tests are bswap.ll, ctlz-v.ll, ctlz-v.ll. It's not possible to put signext on a vector type so we just accept the sign extends here for now. These tests don't pass the vectors the same way clang does (clang puts multiple elements in the same argument, these map 1 element to 1 argument) so we don't need to worry too much about it. With this patch, all known N32/N64 bugs should be fixed and we now pass the first 10,000 tests generated by ABITest.py. Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D6117 llvm-svn: 221534
2014-11-08 00:54:21 +08:00
define i32 @slti6(i32 signext %a) nounwind readnone {
[mips] Small code generation improvement for conditional operator (select) in case the operands are constants and its difference is |1|. It should be possible in those cases to rematerialize the result using MIPS's slt and similar instructions. The small update to some of the tests in cmov.ll, sel1c.ll and sel2c.ll was needed otherwise the optimization implemented in this patch would have been triggered (difference between the operands was 1) and that would have changed the semantic of the tests. llvm-svn: 196498
2013-12-05 20:07:05 +08:00
%1 = icmp slt i32 %a, 7
%2 = select i1 %1, i32 3, i32 4
ret i32 %2
}
[mips][mips64r6] c.cond.fmt, mov[fntz], and mov[fntz].[ds] are not available on MIPS32r6/MIPS64r6 Summary: c.cond.fmt has been replaced by cmp.cond.fmt. Where c.cond.fmt wrote to dedicated condition registers, cmp.cond.fmt writes 1 or 0 to normal FGR's (like the GPR comparisons). mov[fntz] have been replaced by seleqz and selnez. These instructions conditionally zero a register based on a bool in a GPR. The results can then be or'd together to act as a select without, for example, requiring a third register read port. mov[fntz].[ds] have been replaced with sel.[ds] MIPS64r6 currently generates unnecessary sign-extensions for most selects. This is because the result of a SETCC is currently an i32. Bits 32-63 are undefined in i32 and the behaviour of seleqz/selnez would otherwise depend on undefined bits. Later, we will fix this by making the result of SETCC an i64 on MIPS64 targets. Depends on D3958 Reviewers: jkolek, vmedic, zoran.jovanovic Reviewed By: vmedic, zoran.jovanovic Differential Revision: http://reviews.llvm.org/D4003 llvm-svn: 210777
2014-06-12 21:39:06 +08:00
; ALL-LABEL: slti6:
[SDAG] Introduce a combined set to the DAG combiner which tracks nodes which have successfully round-tripped through the combine phase, and use this to ensure all operands to DAG nodes are visited by the combiner, even if they are only added during the combine phase. This is critical to have the combiner reach nodes that are *introduced* during combining. Previously these would sometimes be visited and sometimes not be visited based on whether they happened to end up on the worklist or not. Now we always run them through the combiner. This fixes quite a few bad codegen test cases lurking in the suite while also being more principled. Among these, the TLS codegeneration is particularly exciting for programs that have this in the critical path like TSan-instrumented binaries (although I think they engineer to use a different TLS that is faster anyways). I've tried to check for compile-time regressions here by running llc over a merged (but not LTO-ed) clang bitcode file and observed at most a 3% slowdown in llc. Given that this is essentially a worst case (none of opt or clang are running at this phase) I think this is tolerable. The actual LTO case should be even less costly, and the cost in normal compilation should be negligible. With this combining logic, it is possible to re-legalize as we combine which is necessary to implement PSHUFB formation on x86 as a post-legalize DAG combine (my ultimate goal). Differential Revision: http://reviews.llvm.org/D4638 llvm-svn: 213898
2014-07-25 06:15:28 +08:00
; ALL-DAG: addiu [[R1:\$[0-9]+]], $zero, 6
; ALL-DAG: slt [[R1]], [[R1]], $4
; ALL-DAG: addiu [[R2:\$[0-9]+]], [[R1]], 3
; ALL-NOT: movn
; ALL-NOT: seleqz
; ALL-NOT: selnez