[GCOV] Emit the writeout function as nested loops of global data.

Summary:
Prior to this change, LLVM would in some cases emit *massive* writeout
functions with many 10s of 1000s of function calls in straight-line
code. This is a very wasteful way to represent what are fundamentally
loops and creates a number of scalability issues. Among other things,
register allocating these calls is extremely expensive. While D46127 makes this
less severe, we'll still run into scaling issues with this eventually. If not
in the compile time, just from the code size.

Now the pass builds up global data structures modeling the inputs to
these functions, and simply loops over the data structures calling the
relevant functions with those values. This ensures that the code size is
a fixed and only data size grows with larger amounts of coverage data.

A trivial change to IRBuilder is included to make it easier to build
the constants that make up the global data.

Reviewers: wmi, echristo

Subscribers: sanjoy, mcrosier, llvm-commits, hiraditya

Differential Revision: https://reviews.llvm.org/D46357

llvm-svn: 331407

1 files changed, 66 insertions, 3 deletions

diff --git a/llvm/test/Transforms/GCOVProfiling/function-numbering.ll b/llvm/test/Transforms/GCOVProfiling/function-numbering.ll
index 4e8beb79c02..2acf6b49a79 100644
--- a/llvm/test/Transforms/GCOVProfiling/function-numbering.ll
+++ b/llvm/test/Transforms/GCOVProfiling/function-numbering.ll
@@ -19,9 +19,72 @@ target triple = "x86_64-apple-macosx10.10.0"
 ; GCDA: @[[FOO:[0-9]+]] = private unnamed_addr constant [4 x i8] c"foo\00"
 ; GCDA-NOT: @{{[0-9]+}} = private unnamed_addr constant .* c"bar\00"
 ; GCDA: @[[BAZ:[0-9]+]] = private unnamed_addr constant [4 x i8] c"baz\00"
-; GCDA: define internal void @__llvm_gcov_writeout()
-; GCDA: call void @llvm_gcda_emit_function(i32 0, i8* getelementptr inbounds ([4 x i8], [4 x i8]* @[[FOO]]
-; GCDA: call void @llvm_gcda_emit_function(i32 1, i8* getelementptr inbounds ([4 x i8], [4 x i8]* @[[BAZ]]
+; GCDA: @__llvm_internal_gcov_emit_function_args.0 = internal unnamed_addr constant
+; GCDA-SAME: { i32 0, i8* getelementptr inbounds ([4 x i8], [4 x i8]* @[[FOO]]
+; GCDA-SAME: { i32 1, i8* getelementptr inbounds ([4 x i8], [4 x i8]* @[[BAZ]]
+;
+; GCDA-LABEL: define internal void @__llvm_gcov_writeout() {{.*}} {
+; GCDA-NEXT:  entry:
+; GCDA-NEXT:    br label %[[FILE_LOOP_HEADER:.*]]
+;
+; GCDA:       [[FILE_LOOP_HEADER]]:
+; GCDA-NEXT:    %[[IV:.*]] = phi i32 [ 0, %entry ], [ %[[NEXT_IV:.*]], %[[FILE_LOOP_LATCH:.*]] ]
+; GCDA-NEXT:    %[[FILE_INFO:.*]] = getelementptr inbounds {{.*}}, {{.*}}* @__llvm_internal_gcov_emit_file_info, i32 0, i32 %[[IV]]
+; GCDA-NEXT:    %[[START_FILE_ARGS:.*]] = getelementptr inbounds {{.*}}, {{.*}}* %[[FILE_INFO]], i32 0, i32 0
+; GCDA-NEXT:    %[[START_FILE_ARG_0_PTR:.*]] = getelementptr inbounds {{.*}}, {{.*}}* %[[START_FILE_ARGS]], i32 0, i32 0
+; GCDA-NEXT:    %[[START_FILE_ARG_0:.*]] = load i8*, i8** %[[START_FILE_ARG_0_PTR]]
+; GCDA-NEXT:    %[[START_FILE_ARG_1_PTR:.*]] = getelementptr inbounds {{.*}}, {{.*}}* %[[START_FILE_ARGS]], i32 0, i32 1
+; GCDA-NEXT:    %[[START_FILE_ARG_1:.*]] = load i8*, i8** %[[START_FILE_ARG_1_PTR]]
+; GCDA-NEXT:    %[[START_FILE_ARG_2_PTR:.*]] = getelementptr inbounds {{.*}}, {{.*}}* %[[START_FILE_ARGS]], i32 0, i32 2
+; GCDA-NEXT:    %[[START_FILE_ARG_2:.*]] = load i32, i32* %[[START_FILE_ARG_2_PTR]]
+; GCDA-NEXT:    call void @llvm_gcda_start_file(i8* %[[START_FILE_ARG_0]], i8* %[[START_FILE_ARG_1]], i32 %[[START_FILE_ARG_2]])
+; GCDA-NEXT:    %[[NUM_COUNTERS_PTR:.*]] = getelementptr inbounds {{.*}}, {{.*}}* %[[FILE_INFO]], i32 0, i32 1
+; GCDA-NEXT:    %[[NUM_COUNTERS:.*]] = load i32, i32* %[[NUM_COUNTERS_PTR]]
+; GCDA-NEXT:    %[[EMIT_FUN_ARGS_ARRAY_PTR:.*]] = getelementptr inbounds {{.*}}, {{.*}}* %[[FILE_INFO]], i32 0, i32 2
+; GCDA-NEXT:    %[[EMIT_FUN_ARGS_ARRAY:.*]] = load {{.*}}*, {{.*}}** %[[EMIT_FUN_ARGS_ARRAY_PTR]]
+; GCDA-NEXT:    %[[EMIT_ARCS_ARGS_ARRAY_PTR:.*]] = getelementptr inbounds {{.*}}, {{.*}}* %[[FILE_INFO]], i32 0, i32 3
+; GCDA-NEXT:    %[[EMIT_ARCS_ARGS_ARRAY:.*]] = load {{.*}}*, {{.*}}** %[[EMIT_ARCS_ARGS_ARRAY_PTR]]
+; GCDA-NEXT:    %[[ENTER_COUNTER_LOOP_COND:.*]] = icmp slt i32 0, %[[NUM_COUNTERS]]
+; GCDA-NEXT:    br i1 %[[ENTER_COUNTER_LOOP_COND]], label %[[COUNTER_LOOP:.*]], label %[[FILE_LOOP_LATCH]]
+;
+; GCDA:       [[COUNTER_LOOP]]:
+; GCDA-NEXT:    %[[JV:.*]] = phi i32 [ 0, %[[FILE_LOOP_HEADER]] ], [ %[[NEXT_JV:.*]], %[[COUNTER_LOOP]] ]
+; GCDA-NEXT:    %[[EMIT_FUN_ARGS:.*]] = getelementptr inbounds {{.*}}, {{.*}}* %[[EMIT_FUN_ARGS_ARRAY]], i32 %[[JV]]
+; GCDA-NEXT:    %[[EMIT_FUN_ARG_0_PTR:.*]] = getelementptr inbounds {{.*}}, {{.*}}* %[[EMIT_FUN_ARGS]], i32 0, i32 0
+; GCDA-NEXT:    %[[EMIT_FUN_ARG_0:.*]] = load i32, i32* %[[EMIT_FUN_ARG_0_PTR]]
+; GCDA-NEXT:    %[[EMIT_FUN_ARG_1_PTR:.*]] = getelementptr inbounds {{.*}}, {{.*}}* %[[EMIT_FUN_ARGS]], i32 0, i32 1
+; GCDA-NEXT:    %[[EMIT_FUN_ARG_1:.*]] = load i8*, i8** %[[EMIT_FUN_ARG_1_PTR]]
+; GCDA-NEXT:    %[[EMIT_FUN_ARG_2_PTR:.*]] = getelementptr inbounds {{.*}}, {{.*}}* %[[EMIT_FUN_ARGS]], i32 0, i32 2
+; GCDA-NEXT:    %[[EMIT_FUN_ARG_2:.*]] = load i32, i32* %[[EMIT_FUN_ARG_2_PTR]]
+; GCDA-NEXT:    %[[EMIT_FUN_ARG_3_PTR:.*]] = getelementptr inbounds {{.*}}, {{.*}}* %[[EMIT_FUN_ARGS]], i32 0, i32 3
+; GCDA-NEXT:    %[[EMIT_FUN_ARG_3:.*]] = load i8, i8* %[[EMIT_FUN_ARG_3_PTR]]
+; GCDA-NEXT:    %[[EMIT_FUN_ARG_4_PTR:.*]] = getelementptr inbounds {{.*}}, {{.*}}* %[[EMIT_FUN_ARGS]], i32 0, i32 4
+; GCDA-NEXT:    %[[EMIT_FUN_ARG_4:.*]] = load i32, i32* %[[EMIT_FUN_ARG_4_PTR]]
+; GCDA-NEXT:    call void @llvm_gcda_emit_function(i32 %[[EMIT_FUN_ARG_0]],
+; GCDA-SAME:                                       i8* %[[EMIT_FUN_ARG_1]],
+; GCDA-SAME:                                       i32 %[[EMIT_FUN_ARG_2]],
+; GCDA-SAME:                                       i8 %[[EMIT_FUN_ARG_3]],
+; GCDA-SAME:                                       i32 %[[EMIT_FUN_ARG_4]])
+; GCDA-NEXT:    %[[EMIT_ARCS_ARGS:.*]] = getelementptr inbounds {{.*}}, {{.*}}* %[[EMIT_ARCS_ARGS_ARRAY]], i32 %[[JV]]
+; GCDA-NEXT:    %[[EMIT_ARCS_ARG_0_PTR:.*]] = getelementptr inbounds {{.*}}, {{.*}}* %[[EMIT_ARCS_ARGS]], i32 0, i32 0
+; GCDA-NEXT:    %[[EMIT_ARCS_ARG_0:.*]] = load i32, i32* %[[EMIT_ARCS_ARG_0_PTR]]
+; GCDA-NEXT:    %[[EMIT_ARCS_ARG_1_PTR:.*]] = getelementptr inbounds {{.*}}, {{.*}}* %[[EMIT_ARCS_ARGS]], i32 0, i32 1
+; GCDA-NEXT:    %[[EMIT_ARCS_ARG_1:.*]] = load i64*, i64** %[[EMIT_ARCS_ARG_1_PTR]]
+; GCDA-NEXT:    call void @llvm_gcda_emit_arcs(i32 %[[EMIT_ARCS_ARG_0]],
+; GCDA-SAME:                                   i64* %[[EMIT_ARCS_ARG_1]])
+; GCDA-NEXT:    %[[NEXT_JV]] = add i32 %[[JV]], 1
+; GCDA-NEXT:    %[[COUNTER_LOOP_COND:.*]] = icmp slt i32 %[[NEXT_JV]], %[[NUM_COUNTERS]]
+; GCDA-NEXT:    br i1 %[[COUNTER_LOOP_COND]], label %[[COUNTER_LOOP]], label %[[FILE_LOOP_LATCH]]
+;
+; GCDA:       [[FILE_LOOP_LATCH]]:
+; GCDA-NEXT:    call void @llvm_gcda_summary_info()
+; GCDA-NEXT:    call void @llvm_gcda_end_file()
+; GCDA-NEXT:    %[[NEXT_IV]] = add i32 %[[IV]], 1
+; GCDA-NEXT:    %[[FILE_LOOP_COND:.*]] = icmp slt i32 %[[NEXT_IV]], 1
+; GCDA-NEXT:    br i1 %[[FILE_LOOP_COND]], label %[[FILE_LOOP_HEADER]], label %[[EXIT:.*]]
+;
+; GCDA:       [[EXIT]]:
+; GCDA-NEXT:    ret void
 
 ; GCNO: == foo (0) @
 ; GCNO-NOT: == bar ({{[0-9]+}}) @