RPCS3/llvm-mirror
1
0
Fork 0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-11-25 20:23:11 +01:00
Code Issues Projects Releases Wiki Activity
5ff8f4151c
llvm-mirror/test/CodeGen/AMDGPU/fdiv.f64.ll

163 lines
6.9 KiB
LLVM
Raw Normal View History

AMDGPU: Constant fold rcp node When doing arcp optimization with a constant denominator, this was leaving behind rcps with constant inputs. llvm-svn: 297248
2017-03-08 01:48:46 +01:00
; RUN: llc -march=amdgcn -mcpu=hawaii -verify-machineinstrs < %s | FileCheck -check-prefix=CI -check-prefix=GCN %s
; RUN: llc -march=amdgcn -mcpu=tahiti -verify-machineinstrs < %s | FileCheck -check-prefix=SI -check-prefix=GCN %s
; RUN: llc -march=amdgcn -mcpu=tonga -mattr=-flat-for-global -verify-machineinstrs < %s | FileCheck -check-prefix=CI -check-prefix=GCN %s
R600/SI: Implement correct f64 fdiv This version passes the OpenCL conformance test. llvm-svn: 229239
2015-02-14 05:30:08 +01:00
AMDGPU: Constant fold rcp node When doing arcp optimization with a constant denominator, this was leaving behind rcps with constant inputs. llvm-svn: 297248
2017-03-08 01:48:46 +01:00
; GCN-LABEL: {{^}}fdiv_f64:
; GCN-DAG: buffer_load_dwordx2 [[NUM:v\[[0-9]+:[0-9]+\]]], off, {{s\[[0-9]+:[0-9]+\]}}, 0
; GCN-DAG: buffer_load_dwordx2 [[DEN:v\[[0-9]+:[0-9]+\]]], off, {{s\[[0-9]+:[0-9]+\]}}, 0 offset:8
R600/SI: Implement correct f64 fdiv This version passes the OpenCL conformance test. llvm-svn: 229239
2015-02-14 05:30:08 +01:00
; CI-DAG: v_div_scale_f64 [[SCALE0:v\[[0-9]+:[0-9]+\]]], {{s\[[0-9]+:[0-9]+\]}}, [[DEN]], [[DEN]], [[NUM]]
; CI-DAG: v_div_scale_f64 [[SCALE1:v\[[0-9]+:[0-9]+\]]], vcc, [[NUM]], [[DEN]], [[NUM]]
; Check for div_scale bug workaround on SI
; SI-DAG: v_div_scale_f64 [[SCALE0:v\[[0-9]+:[0-9]+\]]], {{s\[[0-9]+:[0-9]+\]}}, [[DEN]], [[DEN]], [[NUM]]
; SI-DAG: v_div_scale_f64 [[SCALE1:v\[[0-9]+:[0-9]+\]]], {{s\[[0-9]+:[0-9]+\]}}, [[NUM]], [[DEN]], [[NUM]]
AMDGPU: Constant fold rcp node When doing arcp optimization with a constant denominator, this was leaving behind rcps with constant inputs. llvm-svn: 297248
2017-03-08 01:48:46 +01:00
; GCN-DAG: v_rcp_f64_e32 [[RCP_SCALE0:v\[[0-9]+:[0-9]+\]]], [[SCALE0]]
R600/SI: Implement correct f64 fdiv This version passes the OpenCL conformance test. llvm-svn: 229239
2015-02-14 05:30:08 +01:00
AMDGPU: Use unsigned compare for eq/ne For some reason there are both of these available, except for scalar 64-bit compares which only has u64. I'm not sure why there are both (I'm guessing it's for the one bit inputs we don't use), but for consistency always using the unsigned one. llvm-svn: 282832
2016-09-30 03:50:20 +02:00
; SI-DAG: v_cmp_eq_u32_e32 vcc, {{v[0-9]+}}, {{v[0-9]+}}
; SI-DAG: v_cmp_eq_u32_e64 [[CMP0:s\[[0-9]+:[0-9]+\]]], {{v[0-9]+}}, {{v[0-9]+}}
R600/SI: Implement correct f64 fdiv This version passes the OpenCL conformance test. llvm-svn: 229239
2015-02-14 05:30:08 +01:00
; SI-DAG: s_xor_b64 vcc, [[CMP0]], vcc
AMDGPU: Constant fold rcp node When doing arcp optimization with a constant denominator, this was leaving behind rcps with constant inputs. llvm-svn: 297248
2017-03-08 01:48:46 +01:00
; GCN-DAG: v_fma_f64 [[FMA0:v\[[0-9]+:[0-9]+\]]], -[[SCALE0]], [[RCP_SCALE0]], 1.0
; GCN-DAG: v_fma_f64 [[FMA1:v\[[0-9]+:[0-9]+\]]], [[RCP_SCALE0]], [[FMA0]], [[RCP_SCALE0]]
; GCN-DAG: v_fma_f64 [[FMA2:v\[[0-9]+:[0-9]+\]]], -[[SCALE0]], [[FMA1]], 1.0
; GCN-DAG: v_fma_f64 [[FMA3:v\[[0-9]+:[0-9]+\]]], [[FMA1]], [[FMA2]], [[FMA1]]
; GCN-DAG: v_mul_f64 [[MUL:v\[[0-9]+:[0-9]+\]]], [[SCALE1]], [[FMA3]]
; GCN-DAG: v_fma_f64 [[FMA4:v\[[0-9]+:[0-9]+\]]], -[[SCALE0]], [[MUL]], [[SCALE1]]
; GCN: v_div_fmas_f64 [[FMAS:v\[[0-9]+:[0-9]+\]]], [[FMA4]], [[FMA3]], [[MUL]]
; GCN: v_div_fixup_f64 [[RESULT:v\[[0-9]+:[0-9]+\]]], [[FMAS]], [[DEN]], [[NUM]]
; GCN: buffer_store_dwordx2 [[RESULT]]
; GCN: s_endpgm
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-21 22:39:51 +01:00
define amdgpu_kernel void @fdiv_f64(double addrspace(1)* %out, double addrspace(1)* %in) #0 {
[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 llvm-svn: 230786
2015-02-27 20:29:02 +01:00
%gep.1 = getelementptr double, double addrspace(1)* %in, i32 1
AMDGPU: Add volatile to test loads and stores When the memory vectorizer is enabled, these tests break. These tests don't really care about the memory instructions, and it's easier to write check lines with the unmerged loads. llvm-svn: 266071
2016-04-12 15:38:18 +02:00
%num = load volatile double, double addrspace(1)* %in
%den = load volatile double, double addrspace(1)* %gep.1
R600/SI: Implement correct f64 fdiv This version passes the OpenCL conformance test. llvm-svn: 229239
2015-02-14 05:30:08 +01:00
%result = fdiv double %num, %den
store double %result, double addrspace(1)* %out
ret void
}
AMDGPU: Use more accurate fast f64 fdiv A raw v_rcp_f64 isn't accurate enough, so start applying correction.
2021-01-20 19:55:55 +01:00
; GCN-LABEL: {{^}}v_fdiv_f64_afn:
; GCN: v_rcp_f64_e32 v[4:5], v[2:3]
; GCN: v_fma_f64 v[6:7], -v[2:3], v[4:5], 1.0
; GCN: v_fma_f64 v[4:5], v[6:7], v[4:5], v[4:5]
; GCN: v_fma_f64 v[6:7], -v[2:3], v[4:5], 1.0
; GCN: v_fma_f64 v[4:5], v[6:7], v[4:5], v[4:5]
; GCN: v_mul_f64 v[6:7], v[0:1], v[4:5]
; GCN: v_fma_f64 v[0:1], -v[2:3], v[6:7], v[0:1]
; GCN: v_fma_f64 v[0:1], v[0:1], v[4:5], v[6:7]
; GCN: s_setpc_b64
define double @v_fdiv_f64_afn(double %x, double %y) #0 {
%result = fdiv afn double %x, %y
ret double %result
}
; GCN-LABEL: {{^}}v_rcp_f64_afn:
; GCN: v_rcp_f64_e32 v[2:3], v[0:1]
; GCN: v_fma_f64 v[4:5], -v[0:1], v[2:3], 1.0
; GCN: v_fma_f64 v[2:3], v[4:5], v[2:3], v[2:3]
; GCN: v_fma_f64 v[4:5], -v[0:1], v[2:3], 1.0
; GCN: v_fma_f64 v[2:3], v[4:5], v[2:3], v[2:3]
; GCN: v_fma_f64 v[0:1], -v[0:1], v[2:3], 1.0
; GCN: v_fma_f64 v[0:1], v[0:1], v[2:3], v[2:3]
; GCN: s_setpc_b64
define double @v_rcp_f64_afn(double %x) #0 {
%result = fdiv afn double 1.0, %x
ret double %result
}
AMDGPU: Constant fold rcp node When doing arcp optimization with a constant denominator, this was leaving behind rcps with constant inputs. llvm-svn: 297248
2017-03-08 01:48:46 +01:00
; GCN-LABEL: {{^}}fdiv_f64_s_v:
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-21 22:39:51 +01:00
define amdgpu_kernel void @fdiv_f64_s_v(double addrspace(1)* %out, double addrspace(1)* %in, double %num) #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
2015-02-27 22:17:42 +01:00
%den = load double, double addrspace(1)* %in
R600/SI: Implement correct f64 fdiv This version passes the OpenCL conformance test. llvm-svn: 229239
2015-02-14 05:30:08 +01:00
%result = fdiv double %num, %den
store double %result, double addrspace(1)* %out
ret void
}
AMDGPU: Constant fold rcp node When doing arcp optimization with a constant denominator, this was leaving behind rcps with constant inputs. llvm-svn: 297248
2017-03-08 01:48:46 +01:00
; GCN-LABEL: {{^}}fdiv_f64_v_s:
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-21 22:39:51 +01:00
define amdgpu_kernel void @fdiv_f64_v_s(double addrspace(1)* %out, double addrspace(1)* %in, double %den) #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
2015-02-27 22:17:42 +01:00
%num = load double, double addrspace(1)* %in
R600/SI: Implement correct f64 fdiv This version passes the OpenCL conformance test. llvm-svn: 229239
2015-02-14 05:30:08 +01:00
%result = fdiv double %num, %den
store double %result, double addrspace(1)* %out
ret void
}
AMDGPU: Constant fold rcp node When doing arcp optimization with a constant denominator, this was leaving behind rcps with constant inputs. llvm-svn: 297248
2017-03-08 01:48:46 +01:00
; GCN-LABEL: {{^}}fdiv_f64_s_s:
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-21 22:39:51 +01:00
define amdgpu_kernel void @fdiv_f64_s_s(double addrspace(1)* %out, double %num, double %den) #0 {
R600/SI: Implement correct f64 fdiv This version passes the OpenCL conformance test. llvm-svn: 229239
2015-02-14 05:30:08 +01:00
%result = fdiv double %num, %den
store double %result, double addrspace(1)* %out
ret void
}
AMDGPU: Constant fold rcp node When doing arcp optimization with a constant denominator, this was leaving behind rcps with constant inputs. llvm-svn: 297248
2017-03-08 01:48:46 +01:00
; GCN-LABEL: {{^}}v_fdiv_v2f64:
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-21 22:39:51 +01:00
define amdgpu_kernel void @v_fdiv_v2f64(<2 x double> addrspace(1)* %out, <2 x double> addrspace(1)* %in) #0 {
[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 llvm-svn: 230786
2015-02-27 20:29:02 +01:00
%gep.1 = getelementptr <2 x double>, <2 x double> addrspace(1)* %in, i32 1
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-27 22:17:42 +01:00
%num = load <2 x double>, <2 x double> addrspace(1)* %in
%den = load <2 x double>, <2 x double> addrspace(1)* %gep.1
R600/SI: Implement correct f64 fdiv This version passes the OpenCL conformance test. llvm-svn: 229239
2015-02-14 05:30:08 +01:00
%result = fdiv <2 x double> %num, %den
store <2 x double> %result, <2 x double> addrspace(1)* %out
ret void
}
AMDGPU: Constant fold rcp node When doing arcp optimization with a constant denominator, this was leaving behind rcps with constant inputs. llvm-svn: 297248
2017-03-08 01:48:46 +01:00
; GCN-LABEL: {{^}}s_fdiv_v2f64:
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-21 22:39:51 +01:00
define amdgpu_kernel void @s_fdiv_v2f64(<2 x double> addrspace(1)* %out, <2 x double> %num, <2 x double> %den) {
R600/SI: Implement correct f64 fdiv This version passes the OpenCL conformance test. llvm-svn: 229239
2015-02-14 05:30:08 +01:00
%result = fdiv <2 x double> %num, %den
store <2 x double> %result, <2 x double> addrspace(1)* %out
ret void
}
AMDGPU: Constant fold rcp node When doing arcp optimization with a constant denominator, this was leaving behind rcps with constant inputs. llvm-svn: 297248
2017-03-08 01:48:46 +01:00
; GCN-LABEL: {{^}}v_fdiv_v4f64:
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-21 22:39:51 +01:00
define amdgpu_kernel void @v_fdiv_v4f64(<4 x double> addrspace(1)* %out, <4 x double> addrspace(1)* %in) #0 {
[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 llvm-svn: 230786
2015-02-27 20:29:02 +01:00
%gep.1 = getelementptr <4 x double>, <4 x double> addrspace(1)* %in, i32 1
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-27 22:17:42 +01:00
%num = load <4 x double>, <4 x double> addrspace(1)* %in
%den = load <4 x double>, <4 x double> addrspace(1)* %gep.1
R600/SI: Implement correct f64 fdiv This version passes the OpenCL conformance test. llvm-svn: 229239
2015-02-14 05:30:08 +01:00
%result = fdiv <4 x double> %num, %den
store <4 x double> %result, <4 x double> addrspace(1)* %out
ret void
}
AMDGPU: Constant fold rcp node When doing arcp optimization with a constant denominator, this was leaving behind rcps with constant inputs. llvm-svn: 297248
2017-03-08 01:48:46 +01:00
; GCN-LABEL: {{^}}s_fdiv_v4f64:
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-21 22:39:51 +01:00
define amdgpu_kernel void @s_fdiv_v4f64(<4 x double> addrspace(1)* %out, <4 x double> %num, <4 x double> %den) #0 {
R600/SI: Implement correct f64 fdiv This version passes the OpenCL conformance test. llvm-svn: 229239
2015-02-14 05:30:08 +01:00
%result = fdiv <4 x double> %num, %den
store <4 x double> %result, <4 x double> addrspace(1)* %out
ret void
}
AMDGPU: Constant fold rcp node When doing arcp optimization with a constant denominator, this was leaving behind rcps with constant inputs. llvm-svn: 297248
2017-03-08 01:48:46 +01:00
; GCN-LABEL: {{^}}div_fast_2_x_pat_f64:
; GCN: v_mul_f64 [[MUL:v\[[0-9]+:[0-9]+\]]], v{{\[[0-9]+:[0-9]+\]}}, 0.5
; GCN: buffer_store_dwordx2 [[MUL]]
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-21 22:39:51 +01:00
define amdgpu_kernel void @div_fast_2_x_pat_f64(double addrspace(1)* %out) #1 {
AMDGPU: Constant fold rcp node When doing arcp optimization with a constant denominator, this was leaving behind rcps with constant inputs. llvm-svn: 297248
2017-03-08 01:48:46 +01:00
%x = load double, double addrspace(1)* undef
%rcp = fdiv fast double %x, 2.0
store double %rcp, double addrspace(1)* %out, align 4
ret void
}
; GCN-LABEL: {{^}}div_fast_k_x_pat_f64:
; GCN-DAG: s_mov_b32 s[[K_LO:[0-9]+]], 0x9999999a
; GCN-DAG: s_mov_b32 s[[K_HI:[0-9]+]], 0x3fb99999
; GCN: v_mul_f64 [[MUL:v\[[0-9]+:[0-9]+\]]], v{{\[[0-9]+:[0-9]+\]}}, s{{\[}}[[K_LO]]:[[K_HI]]{{\]}}
; GCN: buffer_store_dwordx2 [[MUL]]
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-21 22:39:51 +01:00
define amdgpu_kernel void @div_fast_k_x_pat_f64(double addrspace(1)* %out) #1 {
AMDGPU: Constant fold rcp node When doing arcp optimization with a constant denominator, this was leaving behind rcps with constant inputs. llvm-svn: 297248
2017-03-08 01:48:46 +01:00
%x = load double, double addrspace(1)* undef
%rcp = fdiv fast double %x, 10.0
store double %rcp, double addrspace(1)* %out, align 4
ret void
}
; GCN-LABEL: {{^}}div_fast_neg_k_x_pat_f64:
; GCN-DAG: s_mov_b32 s[[K_LO:[0-9]+]], 0x9999999a
; GCN-DAG: s_mov_b32 s[[K_HI:[0-9]+]], 0xbfb99999
; GCN: v_mul_f64 [[MUL:v\[[0-9]+:[0-9]+\]]], v{{\[[0-9]+:[0-9]+\]}}, s{{\[}}[[K_LO]]:[[K_HI]]{{\]}}
; GCN: buffer_store_dwordx2 [[MUL]]
AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
2017-03-21 22:39:51 +01:00
define amdgpu_kernel void @div_fast_neg_k_x_pat_f64(double addrspace(1)* %out) #1 {
AMDGPU: Constant fold rcp node When doing arcp optimization with a constant denominator, this was leaving behind rcps with constant inputs. llvm-svn: 297248
2017-03-08 01:48:46 +01:00
%x = load double, double addrspace(1)* undef
%rcp = fdiv fast double %x, -10.0
store double %rcp, double addrspace(1)* %out, align 4
ret void
}
attributes #0 = { nounwind }
Revert "[NFC] remove explicit default value for strboolattr attribute in tests" This reverts commit bda6e5bee04c75b1f1332b4fd1ac4e8ef6c3c247. See https://lab.llvm.org/buildbot/#/builders/109/builds/15424 for instance
2021-05-24 19:43:40 +02:00
attributes #1 = { nounwind "unsafe-fp-math"="true" }
Reference in New Issue Copy Permalink