llvm-project/llvm/test/CodeGen/X86/2009-04-29-LinearScanBug.ll

Ignoring revisions in .git-blame-ignore-revs. Click here to bypass and see the normal blame view.

216 lines
14 KiB
LLVM
Raw Normal View History

; RUN: llc < %s -mtriple=i386-apple-darwin10
; rdar://6837009
2011-06-19 20:14:34 +08:00
%0 = type { %struct.pf_state*, %struct.pf_state*, %struct.pf_state*, i32 }
%1 = type { %2 }
%2 = type { %struct.pf_addr, %struct.pf_addr }
%3 = type { %struct.in6_addr }
%4 = type { [4 x i32] }
%5 = type { %struct.pfi_dynaddr*, [4 x i8] }
%6 = type { %struct.pfi_dynaddr*, %struct.pfi_dynaddr** }
%7 = type { %struct.pfr_ktable*, %struct.pfr_ktable*, %struct.pfr_ktable*, i32 }
%8 = type { %struct.pfr_ktable* }
%9 = type { i8* }
%10 = type { %11 }
%11 = type { i8*, i8*, %struct.radix_node* }
%12 = type { [2 x %struct.pf_rulequeue], %13, %13 }
%13 = type { %struct.pf_rulequeue*, %struct.pf_rule**, i32, i32, i32 }
%14 = type { %struct.pf_anchor*, %struct.pf_anchor*, %struct.pf_anchor*, i32 }
%15 = type { %struct.pfi_kif*, %struct.pfi_kif*, %struct.pfi_kif*, i32 }
%16 = type { %struct.ifnet*, %struct.ifnet** }
%17 = type { %18 }
%18 = type { %struct.pkthdr, %19 }
%19 = type { %struct.m_ext, [176 x i8] }
%20 = type { %struct.ifmultiaddr*, %struct.ifmultiaddr** }
%21 = type { i32, %22 }
%22 = type { i8*, [4 x i8] }
%23 = type { %struct.tcphdr* }
%24 = type { %struct.pf_ike_state }
%25 = type { %struct.pf_state_key*, %struct.pf_state_key*, %struct.pf_state_key*, i32 }
%26 = type { %struct.pf_src_node*, %struct.pf_src_node*, %struct.pf_src_node*, i32 }
%struct.anon = type { %struct.pf_state*, %struct.pf_state** }
%struct.au_mask_t = type { i32, i32 }
%struct.bpf_if = type opaque
%struct.dlil_threading_info = type opaque
%struct.ether_header = type { [6 x i8], [6 x i8], i16 }
%struct.ext_refsq = type { %struct.ext_refsq*, %struct.ext_refsq* }
%struct.hook_desc = type { %struct.hook_desc_head, void (i8*)*, i8* }
%struct.hook_desc_head = type { %struct.hook_desc*, %struct.hook_desc** }
%struct.if_data_internal = type { i8, i8, i8, i8, i8, i8, i8, i8, i32, i32, i32, i64, i64, i64, i64, i64, i64, i64, i64, i64, i64, i64, i32, i32, %struct.au_mask_t, i32, i32, i32 }
%struct.ifaddr = type { %struct.sockaddr*, %struct.sockaddr*, %struct.sockaddr*, %struct.ifnet*, %struct.ifaddrhead, void (i32, %struct.rtentry*, %struct.sockaddr*)*, i32, i32, i32, void (%struct.ifaddr*)*, void (%struct.ifaddr*, i32)*, i32 }
%struct.ifaddrhead = type { %struct.ifaddr*, %struct.ifaddr** }
%struct.ifmultiaddr = type { %20, %struct.sockaddr*, %struct.ifmultiaddr*, %struct.ifnet*, i32, i8*, i32, void (i8*)* }
%struct.ifmultihead = type { %struct.ifmultiaddr* }
%struct.ifnet = type { i8*, i8*, %16, %struct.ifaddrhead, i32, i32 (%struct.ifnet*, %struct.sockaddr*)*, i32, %struct.bpf_if*, i16, i16, i16, i16, i32, i8*, i32, %struct.if_data_internal, i32, i32 (%struct.ifnet*, %struct.mbuf*)*, i32 (%struct.ifnet*, i32, i8*)*, i32 (%struct.ifnet*, i32, i32 (%struct.ifnet*, %struct.mbuf*)*)*, void (%struct.ifnet*)*, i32 (%struct.ifnet*, %struct.mbuf*, i8*, i32*)*, void (%struct.ifnet*, %struct.kev_msg*)*, i32 (%struct.ifnet*, %struct.mbuf**, %struct.sockaddr*, i8*, i8*)*, i32, %struct.ifnet_filter_head, i32, i8*, i32, %struct.ifmultihead, i32, i32 (%struct.ifnet*, i32, %struct.ifnet_demux_desc*, i32)*, i32 (%struct.ifnet*, i32)*, %struct.proto_hash_entry*, i8*, %struct.dlil_threading_info*, i8*, %struct.ifqueue, [1 x i32], i32, %struct.ifprefixhead, %struct.lck_rw_t*, %21, i32, %struct.thread*, %struct.pfi_kif*, %struct.lck_mtx_t*, %struct.route }
%struct.ifnet_demux_desc = type { i32, i8*, i32 }
%struct.ifnet_filter = type opaque
%struct.ifnet_filter_head = type { %struct.ifnet_filter*, %struct.ifnet_filter** }
%struct.ifprefix = type { %struct.sockaddr*, %struct.ifnet*, %struct.ifprefixhead, i8, i8 }
%struct.ifprefixhead = type { %struct.ifprefix*, %struct.ifprefix** }
%struct.ifqueue = type { i8*, i8*, i32, i32, i32 }
%struct.in6_addr = type { %4 }
%struct.in_addr = type { i32 }
%struct.kev_d_vectors = type { i32, i8* }
%struct.kev_msg = type { i32, i32, i32, i32, [5 x %struct.kev_d_vectors] }
%struct.lck_mtx_t = type { [3 x i32] }
%struct.lck_rw_t = type <{ [3 x i32] }>
%struct.m_ext = type { i8*, void (i8*, i32, i8*)*, i32, i8*, %struct.ext_refsq, %struct.au_mask_t* }
%struct.m_hdr = type { %struct.mbuf*, %struct.mbuf*, i32, i8*, i16, i16 }
%struct.m_tag = type { %struct.packet_tags, i16, i16, i32 }
%struct.mbuf = type { %struct.m_hdr, %17 }
%struct.packet_tags = type { %struct.m_tag* }
%struct.pf_addr = type { %3 }
%struct.pf_addr_wrap = type <{ %1, %5, i8, i8, [6 x i8] }>
%struct.pf_anchor = type { %14, %14, %struct.pf_anchor*, %struct.pf_anchor_node, [64 x i8], [1024 x i8], %struct.pf_ruleset, i32, i32 }
%struct.pf_anchor_node = type { %struct.pf_anchor* }
%struct.pf_app_state = type { void (%struct.pf_state*, i32, i32, %struct.pf_pdesc*, %struct.pfi_kif*)*, i32 (%struct.pf_app_state*, %struct.pf_app_state*)*, i32 (%struct.pf_app_state*, %struct.pf_app_state*)*, %24 }
%struct.pf_ike_state = type { i64 }
%struct.pf_mtag = type { i8*, i32, i32, i16, i8, i8 }
%struct.pf_palist = type { %struct.pf_pooladdr*, %struct.pf_pooladdr** }
%struct.pf_pdesc = type { %struct.pf_threshold, i64, %23, %struct.pf_addr, %struct.pf_addr, %struct.pf_rule*, %struct.pf_addr*, %struct.pf_addr*, %struct.ether_header*, %struct.mbuf*, i32, %struct.pf_mtag*, i16*, i32, i16, i8, i8, i8, i8 }
%struct.pf_pool = type { %struct.pf_palist, [2 x i32], %struct.pf_pooladdr*, [4 x i8], %struct.in6_addr, %struct.pf_addr, i32, [2 x i16], i8, i8, [1 x i32] }
%struct.pf_pooladdr = type <{ %struct.pf_addr_wrap, %struct.pf_palist, [2 x i32], [16 x i8], %struct.pfi_kif*, [1 x i32] }>
%struct.pf_rule = type <{ %struct.pf_rule_addr, %struct.pf_rule_addr, [8 x %struct.pf_rule_ptr], [64 x i8], [16 x i8], [64 x i8], [64 x i8], [64 x i8], [64 x i8], [32 x i8], %struct.pf_rulequeue, [2 x i32], %struct.pf_pool, i64, [2 x i64], [2 x i64], %struct.pfi_kif*, [4 x i8], %struct.pf_anchor*, [4 x i8], %struct.pfr_ktable*, [4 x i8], i32, i32, [26 x i32], i32, i32, i32, i32, i32, i32, %struct.au_mask_t, i32, i32, i32, i32, i32, i32, i32, i16, i16, i16, i16, i16, [2 x i8], %struct.pf_rule_gid, %struct.pf_rule_gid, i32, i8, i8, i8, i8, i8, i8, i8, i8, i8, i8, i8, i8, i8, i8, i8, i8, i8, i8, i8, i8, i8, i8, i8, i8, i8, i8, [2 x i8] }>
%struct.pf_rule_addr = type <{ %struct.pf_addr_wrap, %struct.pf_rule_xport, i8, [7 x i8] }>
%struct.pf_rule_gid = type { [2 x i32], i8, [3 x i8] }
%struct.pf_rule_ptr = type { %struct.pf_rule*, [4 x i8] }
%struct.pf_rule_xport = type { i32, [4 x i8] }
%struct.pf_rulequeue = type { %struct.pf_rule*, %struct.pf_rule** }
%struct.pf_ruleset = type { [5 x %12], %struct.pf_anchor*, i32, i32, i32 }
%struct.pf_src_node = type <{ %26, %struct.pf_addr, %struct.pf_addr, %struct.pf_rule_ptr, %struct.pfi_kif*, [2 x i64], [2 x i64], i32, i32, %struct.pf_threshold, i64, i64, i8, i8, [2 x i8] }>
%struct.pf_state = type <{ i64, i32, i32, %struct.anon, %struct.anon, %0, %struct.pf_state_peer, %struct.pf_state_peer, %struct.pf_rule_ptr, %struct.pf_rule_ptr, %struct.pf_rule_ptr, %struct.pf_addr, %struct.hook_desc_head, %struct.pf_state_key*, %struct.pfi_kif*, %struct.pfi_kif*, %struct.pf_src_node*, %struct.pf_src_node*, [2 x i64], [2 x i64], i64, i64, i64, i16, i8, i8, i8, i8, [6 x i8] }>
%struct.pf_state_host = type { %struct.pf_addr, %struct.in_addr }
%struct.pf_state_key = type { %struct.pf_state_host, %struct.pf_state_host, %struct.pf_state_host, i8, i8, i8, i8, %struct.pf_app_state*, %25, %25, %struct.anon, i16 }
%struct.pf_state_peer = type { i32, i32, i32, i16, i8, i8, i16, i8, %struct.pf_state_scrub*, [3 x i8] }
%struct.pf_state_scrub = type { %struct.au_mask_t, i32, i32, i32, i16, i8, i8, i32 }
%struct.pf_threshold = type { i32, i32, i32, i32 }
%struct.pfi_dynaddr = type { %6, %struct.pf_addr, %struct.pf_addr, %struct.pf_addr, %struct.pf_addr, %struct.pfr_ktable*, %struct.pfi_kif*, i8*, i32, i32, i32, i8, i8 }
%struct.pfi_kif = type { [16 x i8], %15, [2 x [2 x [2 x i64]]], [2 x [2 x [2 x i64]]], i64, i32, i8*, %struct.ifnet*, i32, i32, %6 }
%struct.pfr_ktable = type { %struct.pfr_tstats, %7, %8, %struct.radix_node_head*, %struct.radix_node_head*, %struct.pfr_ktable*, %struct.pfr_ktable*, %struct.pf_ruleset*, i64, i32 }
%struct.pfr_table = type { [1024 x i8], [32 x i8], i32, i8 }
%struct.pfr_tstats = type { %struct.pfr_table, [2 x [3 x i64]], [2 x [3 x i64]], i64, i64, i64, i32, [2 x i32] }
%struct.pkthdr = type { i32, %struct.ifnet*, i8*, i32, i32, i32, i16, i16, %struct.packet_tags }
%struct.proto_hash_entry = type opaque
%struct.radix_mask = type { i16, i8, i8, %struct.radix_mask*, %9, i32 }
%struct.radix_node = type { %struct.radix_mask*, %struct.radix_node*, i16, i8, i8, %10 }
%struct.radix_node_head = type { %struct.radix_node*, i32, i32, %struct.radix_node* (i8*, i8*, %struct.radix_node_head*, %struct.radix_node*)*, %struct.radix_node* (i8*, i8*, %struct.radix_node_head*, %struct.radix_node*)*, %struct.radix_node* (i8*, i8*, %struct.radix_node_head*)*, %struct.radix_node* (i8*, i8*, %struct.radix_node_head*)*, %struct.radix_node* (i8*, %struct.radix_node_head*)*, %struct.radix_node* (i8*, %struct.radix_node_head*, i32 (%struct.radix_node*, i8*)*, i8*)*, %struct.radix_node* (i8*, i8*, %struct.radix_node_head*)*, %struct.radix_node* (i8*, i8*, %struct.radix_node_head*, i32 (%struct.radix_node*, i8*)*, i8*)*, %struct.radix_node* (i8*, %struct.radix_node_head*)*, i32 (%struct.radix_node_head*, i32 (%struct.radix_node*, i8*)*, i8*)*, i32 (%struct.radix_node_head*, i8*, i8*, i32 (%struct.radix_node*, i8*)*, i8*)*, void (%struct.radix_node*, %struct.radix_node_head*)*, [3 x %struct.radix_node], i32 }
%struct.route = type { %struct.rtentry*, i32, %struct.sockaddr }
%struct.rt_metrics = type { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, [4 x i32] }
%struct.rtentry = type { [2 x %struct.radix_node], %struct.sockaddr*, i32, i32, %struct.ifnet*, %struct.ifaddr*, %struct.sockaddr*, i8*, void (i8*)*, %struct.rt_metrics, %struct.rtentry*, %struct.rtentry*, i32, %struct.lck_mtx_t }
%struct.sockaddr = type { i8, i8, [14 x i8] }
%struct.tcphdr = type { i16, i16, i32, i32, i8, i8, i16, i16, i16 }
%struct.thread = type opaque
@llvm.used = appending global [1 x i8*] [i8* bitcast (i32 (%struct.pf_state_key*, %struct.pf_state_key*)* @pf_state_compare_ext_gwy to i8*)], section "llvm.metadata" ; <[1 x i8*]*> [#uses=0]
define fastcc i32 @pf_state_compare_ext_gwy(%struct.pf_state_key* nocapture %a, %struct.pf_state_key* nocapture %b) nounwind optsize ssp {
entry:
%0 = zext i8 0 to i32 ; <i32> [#uses=2]
%1 = load i8, i8* null, align 1 ; <i8> [#uses=2]
%2 = zext i8 %1 to i32 ; <i32> [#uses=1]
%3 = sub i32 %0, %2 ; <i32> [#uses=1]
%4 = icmp eq i8 0, %1 ; <i1> [#uses=1]
br i1 %4, label %bb1, label %bb79
bb1: ; preds = %entry
%5 = load i8, i8* null, align 4 ; <i8> [#uses=2]
%6 = zext i8 %5 to i32 ; <i32> [#uses=2]
[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
2015-02-28 03:29:02 +08:00
%7 = getelementptr %struct.pf_state_key, %struct.pf_state_key* %b, i32 0, i32 3 ; <i8*> [#uses=1]
%8 = load i8, i8* %7, align 4 ; <i8> [#uses=2]
%9 = zext i8 %8 to i32 ; <i32> [#uses=1]
%10 = sub i32 %6, %9 ; <i32> [#uses=1]
%11 = icmp eq i8 %5, %8 ; <i1> [#uses=1]
br i1 %11, label %bb3, label %bb79
bb3: ; preds = %bb1
switch i32 %0, label %bb23 [
i32 1, label %bb4
i32 6, label %bb6
i32 17, label %bb10
i32 47, label %bb17
i32 50, label %bb21
i32 58, label %bb4
]
bb4: ; preds = %bb3, %bb3
%12 = load i16, i16* null, align 4 ; <i16> [#uses=1]
%13 = zext i16 %12 to i32 ; <i32> [#uses=1]
%14 = sub i32 0, %13 ; <i32> [#uses=1]
br i1 false, label %bb23, label %bb79
bb6: ; preds = %bb3
%15 = load i16, i16* null, align 4 ; <i16> [#uses=1]
%16 = zext i16 %15 to i32 ; <i32> [#uses=1]
%17 = sub i32 0, %16 ; <i32> [#uses=1]
ret i32 %17
bb10: ; preds = %bb3
%18 = load i8, i8* null, align 1 ; <i8> [#uses=2]
%19 = zext i8 %18 to i32 ; <i32> [#uses=1]
%20 = sub i32 0, %19 ; <i32> [#uses=1]
%21 = icmp eq i8 0, %18 ; <i1> [#uses=1]
br i1 %21, label %bb12, label %bb79
bb12: ; preds = %bb10
%22 = load i16, i16* null, align 4 ; <i16> [#uses=1]
%23 = zext i16 %22 to i32 ; <i32> [#uses=1]
%24 = sub i32 0, %23 ; <i32> [#uses=1]
ret i32 %24
bb17: ; preds = %bb3
%25 = load i8, i8* null, align 1 ; <i8> [#uses=2]
%26 = icmp eq i8 %25, 1 ; <i1> [#uses=1]
br i1 %26, label %bb18, label %bb23
bb18: ; preds = %bb17
%27 = icmp eq i8 %25, 0 ; <i1> [#uses=1]
br i1 %27, label %bb19, label %bb23
bb19: ; preds = %bb18
%28 = load i16, i16* null, align 4 ; <i16> [#uses=1]
%29 = zext i16 %28 to i32 ; <i32> [#uses=1]
%30 = sub i32 0, %29 ; <i32> [#uses=1]
br i1 false, label %bb23, label %bb79
bb21: ; preds = %bb3
[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
2015-02-28 03:29:02 +08:00
%31 = getelementptr %struct.pf_state_key, %struct.pf_state_key* %a, i32 0, i32 1, i32 1, i32 0 ; <i32*> [#uses=1]
%32 = load i32, i32* %31, align 4 ; <i32> [#uses=2]
[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
2015-02-28 03:29:02 +08:00
%33 = getelementptr %struct.pf_state_key, %struct.pf_state_key* %b, i32 0, i32 1, i32 1, i32 0 ; <i32*> [#uses=1]
%34 = load i32, i32* %33, align 4 ; <i32> [#uses=2]
%35 = sub i32 %32, %34 ; <i32> [#uses=1]
%36 = icmp eq i32 %32, %34 ; <i1> [#uses=1]
br i1 %36, label %bb23, label %bb79
bb23: ; preds = %bb21, %bb19, %bb18, %bb17, %bb4, %bb3
%cond = icmp eq i32 %6, 2 ; <i1> [#uses=1]
br i1 %cond, label %bb24, label %bb70
bb24: ; preds = %bb23
ret i32 1
bb70: ; preds = %bb23
%37 = load i32 (%struct.pf_app_state*, %struct.pf_app_state*)*, i32 (%struct.pf_app_state*, %struct.pf_app_state*)** null, align 4 ; <i32 (%struct.pf_app_state*, %struct.pf_app_state*)*> [#uses=3]
br i1 false, label %bb78, label %bb73
bb73: ; preds = %bb70
%38 = load i32 (%struct.pf_app_state*, %struct.pf_app_state*)*, i32 (%struct.pf_app_state*, %struct.pf_app_state*)** null, align 4 ; <i32 (%struct.pf_app_state*, %struct.pf_app_state*)*> [#uses=2]
%39 = icmp eq i32 (%struct.pf_app_state*, %struct.pf_app_state*)* %38, null ; <i1> [#uses=1]
br i1 %39, label %bb78, label %bb74
bb74: ; preds = %bb73
%40 = ptrtoint i32 (%struct.pf_app_state*, %struct.pf_app_state*)* %37 to i32 ; <i32> [#uses=1]
%41 = sub i32 0, %40 ; <i32> [#uses=1]
%42 = icmp eq i32 (%struct.pf_app_state*, %struct.pf_app_state*)* %38, %37 ; <i1> [#uses=1]
br i1 %42, label %bb76, label %bb79
bb76: ; preds = %bb74
%43 = tail call i32 %37(%struct.pf_app_state* null, %struct.pf_app_state* null) nounwind ; <i32> [#uses=1]
ret i32 %43
bb78: ; preds = %bb73, %bb70
ret i32 0
bb79: ; preds = %bb74, %bb21, %bb19, %bb10, %bb4, %bb1, %entry
%.0 = phi i32 [ %3, %entry ], [ %10, %bb1 ], [ %14, %bb4 ], [ %20, %bb10 ], [ %30, %bb19 ], [ %35, %bb21 ], [ %41, %bb74 ] ; <i32> [#uses=1]
ret i32 %.0
}