llvm-6502/test/CodeGen/ARM/vld4.ll

174 lines
7.4 KiB
LLVM
Raw Normal View History

; RUN: llc -mtriple=arm-eabi -mattr=+neon %s -o - | FileCheck %s
%struct.__neon_int8x8x4_t = type { <8 x i8>, <8 x i8>, <8 x i8>, <8 x i8> }
%struct.__neon_int16x4x4_t = type { <4 x i16>, <4 x i16>, <4 x i16>, <4 x i16> }
%struct.__neon_int32x2x4_t = type { <2 x i32>, <2 x i32>, <2 x i32>, <2 x i32> }
%struct.__neon_float32x2x4_t = type { <2 x float>, <2 x float>, <2 x float>, <2 x float> }
%struct.__neon_int64x1x4_t = type { <1 x i64>, <1 x i64>, <1 x i64>, <1 x i64> }
%struct.__neon_int8x16x4_t = type { <16 x i8>, <16 x i8>, <16 x i8>, <16 x i8> }
%struct.__neon_int16x8x4_t = type { <8 x i16>, <8 x i16>, <8 x i16>, <8 x i16> }
%struct.__neon_int32x4x4_t = type { <4 x i32>, <4 x i32>, <4 x i32>, <4 x i32> }
%struct.__neon_float32x4x4_t = type { <4 x float>, <4 x float>, <4 x float>, <4 x float> }
define <8 x i8> @vld4i8(i8* %A) nounwind {
;CHECK-LABEL: vld4i8:
;Check the alignment value. Max for this instruction is 256 bits:
;CHECK: vld4.8 {d16, d17, d18, d19}, [r0:64]
%tmp1 = call %struct.__neon_int8x8x4_t @llvm.arm.neon.vld4.v8i8(i8* %A, i32 8)
%tmp2 = extractvalue %struct.__neon_int8x8x4_t %tmp1, 0
%tmp3 = extractvalue %struct.__neon_int8x8x4_t %tmp1, 2
%tmp4 = add <8 x i8> %tmp2, %tmp3
ret <8 x i8> %tmp4
}
;Check for a post-increment updating load with register increment.
define <8 x i8> @vld4i8_update(i8** %ptr, i32 %inc) nounwind {
;CHECK-LABEL: vld4i8_update:
;CHECK: vld4.8 {d16, d17, d18, d19}, [r2:128], r1
%A = load i8*, i8** %ptr
%tmp1 = call %struct.__neon_int8x8x4_t @llvm.arm.neon.vld4.v8i8(i8* %A, i32 16)
%tmp2 = extractvalue %struct.__neon_int8x8x4_t %tmp1, 0
%tmp3 = extractvalue %struct.__neon_int8x8x4_t %tmp1, 2
%tmp4 = add <8 x i8> %tmp2, %tmp3
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@230786 91177308-0d34-0410-b5e6-96231b3b80d8
2015-02-27 19:29:02 +00:00
%tmp5 = getelementptr i8, i8* %A, i32 %inc
store i8* %tmp5, i8** %ptr
ret <8 x i8> %tmp4
}
define <4 x i16> @vld4i16(i16* %A) nounwind {
;CHECK-LABEL: vld4i16:
;Check the alignment value. Max for this instruction is 256 bits:
;CHECK: vld4.16 {d16, d17, d18, d19}, [r0:128]
%tmp0 = bitcast i16* %A to i8*
%tmp1 = call %struct.__neon_int16x4x4_t @llvm.arm.neon.vld4.v4i16(i8* %tmp0, i32 16)
%tmp2 = extractvalue %struct.__neon_int16x4x4_t %tmp1, 0
%tmp3 = extractvalue %struct.__neon_int16x4x4_t %tmp1, 2
%tmp4 = add <4 x i16> %tmp2, %tmp3
ret <4 x i16> %tmp4
}
define <2 x i32> @vld4i32(i32* %A) nounwind {
;CHECK-LABEL: vld4i32:
;Check the alignment value. Max for this instruction is 256 bits:
;CHECK: vld4.32 {d16, d17, d18, d19}, [r0:256]
%tmp0 = bitcast i32* %A to i8*
%tmp1 = call %struct.__neon_int32x2x4_t @llvm.arm.neon.vld4.v2i32(i8* %tmp0, i32 32)
%tmp2 = extractvalue %struct.__neon_int32x2x4_t %tmp1, 0
%tmp3 = extractvalue %struct.__neon_int32x2x4_t %tmp1, 2
%tmp4 = add <2 x i32> %tmp2, %tmp3
ret <2 x i32> %tmp4
}
define <2 x float> @vld4f(float* %A) nounwind {
;CHECK-LABEL: vld4f:
;CHECK: vld4.32
%tmp0 = bitcast float* %A to i8*
%tmp1 = call %struct.__neon_float32x2x4_t @llvm.arm.neon.vld4.v2f32(i8* %tmp0, i32 1)
%tmp2 = extractvalue %struct.__neon_float32x2x4_t %tmp1, 0
%tmp3 = extractvalue %struct.__neon_float32x2x4_t %tmp1, 2
%tmp4 = fadd <2 x float> %tmp2, %tmp3
ret <2 x float> %tmp4
}
define <1 x i64> @vld4i64(i64* %A) nounwind {
;CHECK-LABEL: vld4i64:
;Check the alignment value. Max for this instruction is 256 bits:
;CHECK: vld1.64 {d16, d17, d18, d19}, [r0:256]
%tmp0 = bitcast i64* %A to i8*
%tmp1 = call %struct.__neon_int64x1x4_t @llvm.arm.neon.vld4.v1i64(i8* %tmp0, i32 64)
%tmp2 = extractvalue %struct.__neon_int64x1x4_t %tmp1, 0
%tmp3 = extractvalue %struct.__neon_int64x1x4_t %tmp1, 2
%tmp4 = add <1 x i64> %tmp2, %tmp3
ret <1 x i64> %tmp4
}
define <1 x i64> @vld4i64_update(i64** %ptr, i64* %A) nounwind {
;CHECK-LABEL: vld4i64_update:
;CHECK: vld1.64 {d16, d17, d18, d19}, [r1:256]!
%tmp0 = bitcast i64* %A to i8*
%tmp1 = call %struct.__neon_int64x1x4_t @llvm.arm.neon.vld4.v1i64(i8* %tmp0, i32 64)
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@230786 91177308-0d34-0410-b5e6-96231b3b80d8
2015-02-27 19:29:02 +00:00
%tmp5 = getelementptr i64, i64* %A, i32 4
store i64* %tmp5, i64** %ptr
%tmp2 = extractvalue %struct.__neon_int64x1x4_t %tmp1, 0
%tmp3 = extractvalue %struct.__neon_int64x1x4_t %tmp1, 2
%tmp4 = add <1 x i64> %tmp2, %tmp3
ret <1 x i64> %tmp4
}
define <16 x i8> @vld4Qi8(i8* %A) nounwind {
;CHECK-LABEL: vld4Qi8:
;Check the alignment value. Max for this instruction is 256 bits:
;CHECK: vld4.8 {d16, d18, d20, d22}, [r0:256]!
;CHECK: vld4.8 {d17, d19, d21, d23}, [r0:256]
%tmp1 = call %struct.__neon_int8x16x4_t @llvm.arm.neon.vld4.v16i8(i8* %A, i32 64)
%tmp2 = extractvalue %struct.__neon_int8x16x4_t %tmp1, 0
%tmp3 = extractvalue %struct.__neon_int8x16x4_t %tmp1, 2
%tmp4 = add <16 x i8> %tmp2, %tmp3
ret <16 x i8> %tmp4
}
define <8 x i16> @vld4Qi16(i16* %A) nounwind {
;CHECK-LABEL: vld4Qi16:
;Check for no alignment specifier.
;CHECK: vld4.16 {d16, d18, d20, d22}, [r0]!
;CHECK: vld4.16 {d17, d19, d21, d23}, [r0]
%tmp0 = bitcast i16* %A to i8*
%tmp1 = call %struct.__neon_int16x8x4_t @llvm.arm.neon.vld4.v8i16(i8* %tmp0, i32 1)
%tmp2 = extractvalue %struct.__neon_int16x8x4_t %tmp1, 0
%tmp3 = extractvalue %struct.__neon_int16x8x4_t %tmp1, 2
%tmp4 = add <8 x i16> %tmp2, %tmp3
ret <8 x i16> %tmp4
}
;Check for a post-increment updating load.
define <8 x i16> @vld4Qi16_update(i16** %ptr) nounwind {
;CHECK-LABEL: vld4Qi16_update:
;CHECK: vld4.16 {d16, d18, d20, d22}, [r1:64]!
;CHECK: vld4.16 {d17, d19, d21, d23}, [r1:64]!
%A = load i16*, i16** %ptr
%tmp0 = bitcast i16* %A to i8*
%tmp1 = call %struct.__neon_int16x8x4_t @llvm.arm.neon.vld4.v8i16(i8* %tmp0, i32 8)
%tmp2 = extractvalue %struct.__neon_int16x8x4_t %tmp1, 0
%tmp3 = extractvalue %struct.__neon_int16x8x4_t %tmp1, 2
%tmp4 = add <8 x i16> %tmp2, %tmp3
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@230786 91177308-0d34-0410-b5e6-96231b3b80d8
2015-02-27 19:29:02 +00:00
%tmp5 = getelementptr i16, i16* %A, i32 32
store i16* %tmp5, i16** %ptr
ret <8 x i16> %tmp4
}
define <4 x i32> @vld4Qi32(i32* %A) nounwind {
;CHECK-LABEL: vld4Qi32:
;CHECK: vld4.32
;CHECK: vld4.32
%tmp0 = bitcast i32* %A to i8*
%tmp1 = call %struct.__neon_int32x4x4_t @llvm.arm.neon.vld4.v4i32(i8* %tmp0, i32 1)
%tmp2 = extractvalue %struct.__neon_int32x4x4_t %tmp1, 0
%tmp3 = extractvalue %struct.__neon_int32x4x4_t %tmp1, 2
%tmp4 = add <4 x i32> %tmp2, %tmp3
ret <4 x i32> %tmp4
}
define <4 x float> @vld4Qf(float* %A) nounwind {
;CHECK-LABEL: vld4Qf:
;CHECK: vld4.32
;CHECK: vld4.32
%tmp0 = bitcast float* %A to i8*
%tmp1 = call %struct.__neon_float32x4x4_t @llvm.arm.neon.vld4.v4f32(i8* %tmp0, i32 1)
%tmp2 = extractvalue %struct.__neon_float32x4x4_t %tmp1, 0
%tmp3 = extractvalue %struct.__neon_float32x4x4_t %tmp1, 2
%tmp4 = fadd <4 x float> %tmp2, %tmp3
ret <4 x float> %tmp4
}
declare %struct.__neon_int8x8x4_t @llvm.arm.neon.vld4.v8i8(i8*, i32) nounwind readonly
declare %struct.__neon_int16x4x4_t @llvm.arm.neon.vld4.v4i16(i8*, i32) nounwind readonly
declare %struct.__neon_int32x2x4_t @llvm.arm.neon.vld4.v2i32(i8*, i32) nounwind readonly
declare %struct.__neon_float32x2x4_t @llvm.arm.neon.vld4.v2f32(i8*, i32) nounwind readonly
declare %struct.__neon_int64x1x4_t @llvm.arm.neon.vld4.v1i64(i8*, i32) nounwind readonly
declare %struct.__neon_int8x16x4_t @llvm.arm.neon.vld4.v16i8(i8*, i32) nounwind readonly
declare %struct.__neon_int16x8x4_t @llvm.arm.neon.vld4.v8i16(i8*, i32) nounwind readonly
declare %struct.__neon_int32x4x4_t @llvm.arm.neon.vld4.v4i32(i8*, i32) nounwind readonly
declare %struct.__neon_float32x4x4_t @llvm.arm.neon.vld4.v4f32(i8*, i32) nounwind readonly