X-Git-Url: http://demsky.eecs.uci.edu/git/?a=blobdiff_plain;f=lib%2FTarget%2FREADME.txt;h=09e5433578cca167a85d32f1a0d680452ee0b237;hb=6140a8b0572c80383a67248e8b1c0cf2379b1c43;hp=f39d8b2a244bf2af89f180d6699d1e12692fec9d;hpb=497b7e97c05df6c7a1b8ddae8048ab8c9367038d;p=oota-llvm.git diff --git a/lib/Target/README.txt b/lib/Target/README.txt index f39d8b2a244..09e5433578c 100644 --- a/lib/Target/README.txt +++ b/lib/Target/README.txt @@ -2,13 +2,6 @@ Target Independent Opportunities: //===---------------------------------------------------------------------===// -We should make the various target's "IMPLICIT_DEF" instructions be a single -target-independent opcode like TargetInstrInfo::INLINEASM. This would allow -us to eliminate the TargetInstrDesc::isImplicitDef() method, and would allow -us to avoid having to define this for every target for every register class. - -//===---------------------------------------------------------------------===// - With the recent changes to make the implicit def/use set explicit in machineinstrs, we should change the target descriptions for 'call' instructions so that the .td files don't list all the call-clobbered registers as implicit @@ -30,7 +23,10 @@ Make the PPC branch selector target independant //===---------------------------------------------------------------------===// Get the C front-end to expand hypot(x,y) -> llvm.sqrt(x*x+y*y) when errno and -precision don't matter (ffastmath). Misc/mandel will like this. :) +precision don't matter (ffastmath). Misc/mandel will like this. :) This isn't +safe in general, even on darwin. See the libm implementation of hypot for +examples (which special case when x/y are exactly zero to get signed zeros etc +right). //===---------------------------------------------------------------------===// @@ -166,6 +162,9 @@ Expand these to calls of sin/cos and stores: Doing so could allow SROA of the destination pointers. See also: http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17687 +This is now easily doable with MRVs. We could even make an intrinsic for this +if anyone cared enough about sincos. + //===---------------------------------------------------------------------===// Scalar Repl cannot currently promote this testcase to 'ret long cst': @@ -271,15 +270,7 @@ alas... //===---------------------------------------------------------------------===// -This isn't recognized as bswap by instcombine: - -unsigned int swap_32(unsigned int v) { - v = ((v & 0x00ff00ffU) << 8) | ((v & 0xff00ff00U) >> 8); - v = ((v & 0x0000ffffU) << 16) | ((v & 0xffff0000U) >> 16); - return v; -} - -Nor is this (yes, it really is bswap): +This isn't recognized as bswap by instcombine (yes, it really is bswap): unsigned long reverse(unsigned v) { unsigned t; @@ -291,6 +282,39 @@ unsigned long reverse(unsigned v) { //===---------------------------------------------------------------------===// +These idioms should be recognized as popcount (see PR1488): + +unsigned countbits_slow(unsigned v) { + unsigned c; + for (c = 0; v; v >>= 1) + c += v & 1; + return c; +} +unsigned countbits_fast(unsigned v){ + unsigned c; + for (c = 0; v; c++) + v &= v - 1; // clear the least significant bit set + return c; +} + +BITBOARD = unsigned long long +int PopCnt(register BITBOARD a) { + register int c=0; + while(a) { + c++; + a &= a - 1; + } + return c; +} +unsigned int popcount(unsigned int input) { + unsigned int count = 0; + for (unsigned int i = 0; i < 4 * 8; i++) + count += (input >> i) & i; + return count; +} + +//===---------------------------------------------------------------------===// + These should turn into single 16-bit (unaligned?) loads on little/big endian processors. @@ -317,11 +341,6 @@ this construct. //===---------------------------------------------------------------------===// -Instcombine misses several of these cases (see the testcase in the patch): -http://gcc.gnu.org/ml/gcc-patches/2006-10/msg01519.html - -//===---------------------------------------------------------------------===// - viterbi speeds up *significantly* if the various "history" related copy loops are turned into memcpy calls at the source level. We need a "loops to memcpy" pass. @@ -397,32 +416,32 @@ followed by an uncond branch to an exit block. ; This testcase is due to tail-duplication not wanting to copy the return ; instruction into the terminating blocks because there was other code ; optimized out of the function after the taildup happened. -;RUN: llvm-upgrade < %s | llvm-as | opt -tailcallelim | llvm-dis | not grep call +; RUN: llvm-as < %s | opt -tailcallelim | llvm-dis | not grep call -int %t4(int %a) { +define i32 @t4(i32 %a) { entry: - %tmp.1 = and int %a, 1 - %tmp.2 = cast int %tmp.1 to bool - br bool %tmp.2, label %then.0, label %else.0 - -then.0: - %tmp.5 = add int %a, -1 - %tmp.3 = call int %t4( int %tmp.5 ) - br label %return - -else.0: - %tmp.7 = setne int %a, 0 - br bool %tmp.7, label %then.1, label %return - -then.1: - %tmp.11 = add int %a, -2 - %tmp.9 = call int %t4( int %tmp.11 ) - br label %return - -return: - %result.0 = phi int [ 0, %else.0 ], [ %tmp.3, %then.0 ], + %tmp.1 = and i32 %a, 1 ; [#uses=1] + %tmp.2 = icmp ne i32 %tmp.1, 0 ; [#uses=1] + br i1 %tmp.2, label %then.0, label %else.0 + +then.0: ; preds = %entry + %tmp.5 = add i32 %a, -1 ; [#uses=1] + %tmp.3 = call i32 @t4( i32 %tmp.5 ) ; [#uses=1] + br label %return + +else.0: ; preds = %entry + %tmp.7 = icmp ne i32 %a, 0 ; [#uses=1] + br i1 %tmp.7, label %then.1, label %return + +then.1: ; preds = %else.0 + %tmp.11 = add i32 %a, -2 ; [#uses=1] + %tmp.9 = call i32 @t4( i32 %tmp.11 ) ; [#uses=1] + br label %return + +return: ; preds = %then.1, %else.0, %then.0 + %result.0 = phi i32 [ 0, %else.0 ], [ %tmp.3, %then.0 ], [ %tmp.9, %then.1 ] - ret int %result.0 + ret i32 %result.0 } //===---------------------------------------------------------------------===// @@ -443,24 +462,35 @@ long long fib(const long long n) { //===---------------------------------------------------------------------===// +Tail recursion elimination should handle: + +int pow2m1(int n) { + if (n == 0) + return 0; + return 2 * pow2m1 (n - 1) + 1; +} + +Also, multiplies can be turned into SHL's, so they should be handled as if +they were associative. "return foo() << 1" can be tail recursion eliminated. + +//===---------------------------------------------------------------------===// + Argument promotion should promote arguments for recursive functions, like this: -; RUN: llvm-upgrade < %s | llvm-as | opt -argpromotion | llvm-dis | grep x.val +; RUN: llvm-as < %s | opt -argpromotion | llvm-dis | grep x.val -implementation ; Functions: - -internal int %foo(int* %x) { +define internal i32 @foo(i32* %x) { entry: - %tmp = load int* %x - %tmp.foo = call int %foo(int *%x) - ret int %tmp.foo + %tmp = load i32* %x ; [#uses=0] + %tmp.foo = call i32 @foo( i32* %x ) ; [#uses=1] + ret i32 %tmp.foo } -int %bar(int* %x) { +define i32 @bar(i32* %x) { entry: - %tmp3 = call int %foo( int* %x) ; [#uses=1] - ret int %tmp3 + %tmp3 = call i32 @foo( i32* %x ) ; [#uses=1] + ret i32 %tmp3 } //===---------------------------------------------------------------------===// @@ -480,6 +510,8 @@ int i; } } +BasicAA also doesn't do this for add. It needs to know that &A[i+1] != &A[i]. + //===---------------------------------------------------------------------===// We should investigate an instruction sinking pass. Consider this silly @@ -529,16 +561,22 @@ We should turn things like "load+fabs+store" and "load+fneg+store" into the corresponding integer operations. On a yonah, this loop: double a[256]; - for (b = 0; b < 10000000; b++) - for (i = 0; i < 256; i++) - a[i] = -a[i]; +void foo() { + int i, b; + for (b = 0; b < 10000000; b++) + for (i = 0; i < 256; i++) + a[i] = -a[i]; +} is twice as slow as this loop: long long a[256]; - for (b = 0; b < 10000000; b++) - for (i = 0; i < 256; i++) - a[i] ^= (1ULL << 63); +void foo() { + int i, b; + for (b = 0; b < 10000000; b++) + for (i = 0; i < 256; i++) + a[i] ^= (1ULL << 63); +} and I suspect other processors are similar. On X86 in particular this is a big win because doing this with integers allows the use of read/modify/write @@ -610,6 +648,1059 @@ pointer parameters for alias analysis. Some ideas: arguments when the function is inlined. These functions can be inferred by various analysis passes such as the -globalsmodrefaa pass. +globalsmodrefaa pass. Note that getting #2 right is actually really tricky. +Consider this code: + +struct S; S G; +void caller(S byvalarg) { G.field = 1; ... } +void callee() { caller(G); } + +The fact that the caller does not modify byval arg is not enough, we need +to know that it doesn't modify G either. This is very tricky. + +//===---------------------------------------------------------------------===// + +We should add an FRINT node to the DAG to model targets that have legal +implementations of ceil/floor/rint. + +//===---------------------------------------------------------------------===// + +This GCC bug: http://gcc.gnu.org/bugzilla/show_bug.cgi?id=34043 +contains a testcase that compiles down to: + + %struct.XMM128 = type { <4 x float> } +.. + %src = alloca %struct.XMM128 +.. + %tmp6263 = bitcast %struct.XMM128* %src to <2 x i64>* + %tmp65 = getelementptr %struct.XMM128* %src, i32 0, i32 0 + store <2 x i64> %tmp5899, <2 x i64>* %tmp6263, align 16 + %tmp66 = load <4 x float>* %tmp65, align 16 + %tmp71 = add <4 x float> %tmp66, %tmp66 + +If the mid-level optimizer turned the bitcast of pointer + store of tmp5899 +into a bitcast of the vector value and a store to the pointer, then the +store->load could be easily removed. + +//===---------------------------------------------------------------------===// + +Consider: + +int test() { + long long input[8] = {1,1,1,1,1,1,1,1}; + foo(input); +} + +We currently compile this into a memcpy from a global array since the +initializer is fairly large and not memset'able. This is good, but the memcpy +gets lowered to load/stores in the code generator. This is also ok, except +that the codegen lowering for memcpy doesn't handle the case when the source +is a constant global. This gives us atrocious code like this: + + call "L1$pb" +"L1$pb": + popl %eax + movl _C.0.1444-"L1$pb"+32(%eax), %ecx + movl %ecx, 40(%esp) + movl _C.0.1444-"L1$pb"+20(%eax), %ecx + movl %ecx, 28(%esp) + movl _C.0.1444-"L1$pb"+36(%eax), %ecx + movl %ecx, 44(%esp) + movl _C.0.1444-"L1$pb"+44(%eax), %ecx + movl %ecx, 52(%esp) + movl _C.0.1444-"L1$pb"+40(%eax), %ecx + movl %ecx, 48(%esp) + movl _C.0.1444-"L1$pb"+12(%eax), %ecx + movl %ecx, 20(%esp) + movl _C.0.1444-"L1$pb"+4(%eax), %ecx +... + +instead of: + movl $1, 16(%esp) + movl $0, 20(%esp) + movl $1, 24(%esp) + movl $0, 28(%esp) + movl $1, 32(%esp) + movl $0, 36(%esp) + ... + +//===---------------------------------------------------------------------===// + +http://llvm.org/PR717: + +The following code should compile into "ret int undef". Instead, LLVM +produces "ret int 0": + +int f() { + int x = 4; + int y; + if (x == 3) y = 0; + return y; +} + +//===---------------------------------------------------------------------===// + +The loop unroller should partially unroll loops (instead of peeling them) +when code growth isn't too bad and when an unroll count allows simplification +of some code within the loop. One trivial example is: + +#include +int main() { + int nRet = 17; + int nLoop; + for ( nLoop = 0; nLoop < 1000; nLoop++ ) { + if ( nLoop & 1 ) + nRet += 2; + else + nRet -= 1; + } + return nRet; +} + +Unrolling by 2 would eliminate the '&1' in both copies, leading to a net +reduction in code size. The resultant code would then also be suitable for +exit value computation. + +//===---------------------------------------------------------------------===// + +We miss a bunch of rotate opportunities on various targets, including ppc, x86, +etc. On X86, we miss a bunch of 'rotate by variable' cases because the rotate +matching code in dag combine doesn't look through truncates aggressively +enough. Here are some testcases reduces from GCC PR17886: + +unsigned long long f(unsigned long long x, int y) { + return (x << y) | (x >> 64-y); +} +unsigned f2(unsigned x, int y){ + return (x << y) | (x >> 32-y); +} +unsigned long long f3(unsigned long long x){ + int y = 9; + return (x << y) | (x >> 64-y); +} +unsigned f4(unsigned x){ + int y = 10; + return (x << y) | (x >> 32-y); +} +unsigned long long f5(unsigned long long x, unsigned long long y) { + return (x << 8) | ((y >> 48) & 0xffull); +} +unsigned long long f6(unsigned long long x, unsigned long long y, int z) { + switch(z) { + case 1: + return (x << 8) | ((y >> 48) & 0xffull); + case 2: + return (x << 16) | ((y >> 40) & 0xffffull); + case 3: + return (x << 24) | ((y >> 32) & 0xffffffull); + case 4: + return (x << 32) | ((y >> 24) & 0xffffffffull); + default: + return (x << 40) | ((y >> 16) & 0xffffffffffull); + } +} + +On X86-64, we only handle f2/f3/f4 right. On x86-32, a few of these +generate truly horrible code, instead of using shld and friends. On +ARM, we end up with calls to L___lshrdi3/L___ashldi3 in f, which is +badness. PPC64 misses f, f5 and f6. CellSPU aborts in isel. + +//===---------------------------------------------------------------------===// + +We do a number of simplifications in simplify libcalls to strength reduce +standard library functions, but we don't currently merge them together. For +example, it is useful to merge memcpy(a,b,strlen(b)) -> strcpy. This can only +be done safely if "b" isn't modified between the strlen and memcpy of course. + +//===---------------------------------------------------------------------===// + +We should be able to evaluate this loop: + +int test(int x_offs) { + while (x_offs > 4) + x_offs -= 4; + return x_offs; +} + +//===---------------------------------------------------------------------===// + +Reassociate should turn things like: + +int factorial(int X) { + return X*X*X*X*X*X*X*X; +} + +into llvm.powi calls, allowing the code generator to produce balanced +multiplication trees. + +//===---------------------------------------------------------------------===// + +We generate a horrible libcall for llvm.powi. For example, we compile: + +#include +double f(double a) { return std::pow(a, 4); } + +into: + +__Z1fd: + subl $12, %esp + movsd 16(%esp), %xmm0 + movsd %xmm0, (%esp) + movl $4, 8(%esp) + call L___powidf2$stub + addl $12, %esp + ret + +GCC produces: + +__Z1fd: + subl $12, %esp + movsd 16(%esp), %xmm0 + mulsd %xmm0, %xmm0 + mulsd %xmm0, %xmm0 + movsd %xmm0, (%esp) + fldl (%esp) + addl $12, %esp + ret + +//===---------------------------------------------------------------------===// + +We compile this program: (from GCC PR11680) +http://gcc.gnu.org/bugzilla/attachment.cgi?id=4487 + +Into code that runs the same speed in fast/slow modes, but both modes run 2x +slower than when compile with GCC (either 4.0 or 4.2): + +$ llvm-g++ perf.cpp -O3 -fno-exceptions +$ time ./a.out fast +1.821u 0.003s 0:01.82 100.0% 0+0k 0+0io 0pf+0w + +$ g++ perf.cpp -O3 -fno-exceptions +$ time ./a.out fast +0.821u 0.001s 0:00.82 100.0% 0+0k 0+0io 0pf+0w + +It looks like we are making the same inlining decisions, so this may be raw +codegen badness or something else (haven't investigated). + +//===---------------------------------------------------------------------===// + +We miss some instcombines for stuff like this: +void bar (void); +void foo (unsigned int a) { + /* This one is equivalent to a >= (3 << 2). */ + if ((a >> 2) >= 3) + bar (); +} + +A few other related ones are in GCC PR14753. + +//===---------------------------------------------------------------------===// + +Divisibility by constant can be simplified (according to GCC PR12849) from +being a mulhi to being a mul lo (cheaper). Testcase: + +void bar(unsigned n) { + if (n % 3 == 0) + true(); +} + +I think this basically amounts to a dag combine to simplify comparisons against +multiply hi's into a comparison against the mullo. + +//===---------------------------------------------------------------------===// + +SROA is not promoting the union on the stack in this example, we should end +up with no allocas. + +union vec2d { + double e[2]; + double v __attribute__((vector_size(16))); +}; +typedef union vec2d vec2d; + +static vec2d a={{1,2}}, b={{3,4}}; + +vec2d foo () { + return (vec2d){ .v = a.v + b.v * (vec2d){{5,5}}.v }; +} + +//===---------------------------------------------------------------------===// + +Better mod/ref analysis for scanf would allow us to eliminate the vtable and a +bunch of other stuff from this example (see PR1604): + +#include +struct test { + int val; + virtual ~test() {} +}; + +int main() { + test t; + std::scanf("%d", &t.val); + std::printf("%d\n", t.val); +} + +//===---------------------------------------------------------------------===// + +Instcombine will merge comparisons like (x >= 10) && (x < 20) by producing (x - +10) u< 10, but only when the comparisons have matching sign. + +This could be converted with a similiar technique. (PR1941) + +define i1 @test(i8 %x) { + %A = icmp uge i8 %x, 5 + %B = icmp slt i8 %x, 20 + %C = and i1 %A, %B + ret i1 %C +} + +//===---------------------------------------------------------------------===// + +These functions perform the same computation, but produce different assembly. + +define i8 @select(i8 %x) readnone nounwind { + %A = icmp ult i8 %x, 250 + %B = select i1 %A, i8 0, i8 1 + ret i8 %B +} + +define i8 @addshr(i8 %x) readnone nounwind { + %A = zext i8 %x to i9 + %B = add i9 %A, 6 ;; 256 - 250 == 6 + %C = lshr i9 %B, 8 + %D = trunc i9 %C to i8 + ret i8 %D +} + +//===---------------------------------------------------------------------===// + +From gcc bug 24696: +int +f (unsigned long a, unsigned long b, unsigned long c) +{ + return ((a & (c - 1)) != 0) || ((b & (c - 1)) != 0); +} +int +f (unsigned long a, unsigned long b, unsigned long c) +{ + return ((a & (c - 1)) != 0) | ((b & (c - 1)) != 0); +} +Both should combine to ((a|b) & (c-1)) != 0. Currently not optimized with +"clang -emit-llvm-bc | opt -std-compile-opts". + +//===---------------------------------------------------------------------===// + +From GCC Bug 20192: +#define PMD_MASK (~((1UL << 23) - 1)) +void clear_pmd_range(unsigned long start, unsigned long end) +{ + if (!(start & ~PMD_MASK) && !(end & ~PMD_MASK)) + f(); +} +The expression should optimize to something like +"!((start|end)&~PMD_MASK). Currently not optimized with "clang +-emit-llvm-bc | opt -std-compile-opts". + +//===---------------------------------------------------------------------===// + +From GCC Bug 15241: +unsigned int +foo (unsigned int a, unsigned int b) +{ + if (a <= 7 && b <= 7) + baz (); +} +Should combine to "(a|b) <= 7". Currently not optimized with "clang +-emit-llvm-bc | opt -std-compile-opts". + +//===---------------------------------------------------------------------===// + +From GCC Bug 3756: +int +pn (int n) +{ + return (n >= 0 ? 1 : -1); +} +Should combine to (n >> 31) | 1. Currently not optimized with "clang +-emit-llvm-bc | opt -std-compile-opts | llc". + +//===---------------------------------------------------------------------===// + +From GCC Bug 28685: +int test(int a, int b) +{ + int lt = a < b; + int eq = a == b; + + return (lt || eq); +} +Should combine to "a <= b". Currently not optimized with "clang +-emit-llvm-bc | opt -std-compile-opts | llc". + +//===---------------------------------------------------------------------===// + +void a(int variable) +{ + if (variable == 4 || variable == 6) + bar(); +} +This should optimize to "if ((variable | 2) == 6)". Currently not +optimized with "clang -emit-llvm-bc | opt -std-compile-opts | llc". + +//===---------------------------------------------------------------------===// + +unsigned int f(unsigned int i, unsigned int n) {++i; if (i == n) ++i; return +i;} +unsigned int f2(unsigned int i, unsigned int n) {++i; i += i == n; return i;} +These should combine to the same thing. Currently, the first function +produces better code on X86. + +//===---------------------------------------------------------------------===// + +From GCC Bug 15784: +#define abs(x) x>0?x:-x +int f(int x, int y) +{ + return (abs(x)) >= 0; +} +This should optimize to x == INT_MIN. (With -fwrapv.) Currently not +optimized with "clang -emit-llvm-bc | opt -std-compile-opts". + +//===---------------------------------------------------------------------===// + +From GCC Bug 14753: +void +rotate_cst (unsigned int a) +{ + a = (a << 10) | (a >> 22); + if (a == 123) + bar (); +} +void +minus_cst (unsigned int a) +{ + unsigned int tem; + + tem = 20 - a; + if (tem == 5) + bar (); +} +void +mask_gt (unsigned int a) +{ + /* This is equivalent to a > 15. */ + if ((a & ~7) > 8) + bar (); +} +void +rshift_gt (unsigned int a) +{ + /* This is equivalent to a > 23. */ + if ((a >> 2) > 5) + bar (); +} +All should simplify to a single comparison. All of these are +currently not optimized with "clang -emit-llvm-bc | opt +-std-compile-opts". + +//===---------------------------------------------------------------------===// + +From GCC Bug 32605: +int c(int* x) {return (char*)x+2 == (char*)x;} +Should combine to 0. Currently not optimized with "clang +-emit-llvm-bc | opt -std-compile-opts" (although llc can optimize it). + +//===---------------------------------------------------------------------===// + +int a(unsigned char* b) {return *b > 99;} +There's an unnecessary zext in the generated code with "clang +-emit-llvm-bc | opt -std-compile-opts". + +//===---------------------------------------------------------------------===// + +int a(unsigned b) {return ((b << 31) | (b << 30)) >> 31;} +Should be combined to "((b >> 1) | b) & 1". Currently not optimized +with "clang -emit-llvm-bc | opt -std-compile-opts". //===---------------------------------------------------------------------===// + +unsigned a(unsigned x, unsigned y) { return x | (y & 1) | (y & 2);} +Should combine to "x | (y & 3)". Currently not optimized with "clang +-emit-llvm-bc | opt -std-compile-opts". + +//===---------------------------------------------------------------------===// + +unsigned a(unsigned a) {return ((a | 1) & 3) | (a & -4);} +Should combine to "a | 1". Currently not optimized with "clang +-emit-llvm-bc | opt -std-compile-opts". + +//===---------------------------------------------------------------------===// + +int a(int a, int b, int c) {return (~a & c) | ((c|a) & b);} +Should fold to "(~a & c) | (a & b)". Currently not optimized with +"clang -emit-llvm-bc | opt -std-compile-opts". + +//===---------------------------------------------------------------------===// + +int a(int a,int b) {return (~(a|b))|a;} +Should fold to "a|~b". Currently not optimized with "clang +-emit-llvm-bc | opt -std-compile-opts". + +//===---------------------------------------------------------------------===// + +int a(int a, int b) {return (a&&b) || (a&&!b);} +Should fold to "a". Currently not optimized with "clang -emit-llvm-bc +| opt -std-compile-opts". + +//===---------------------------------------------------------------------===// + +int a(int a, int b, int c) {return (a&&b) || (!a&&c);} +Should fold to "a ? b : c", or at least something sane. Currently not +optimized with "clang -emit-llvm-bc | opt -std-compile-opts". + +//===---------------------------------------------------------------------===// + +int a(int a, int b, int c) {return (a&&b) || (a&&c) || (a&&b&&c);} +Should fold to a && (b || c). Currently not optimized with "clang +-emit-llvm-bc | opt -std-compile-opts". + +//===---------------------------------------------------------------------===// + +int a(int x) {return x | ((x & 8) ^ 8);} +Should combine to x | 8. Currently not optimized with "clang +-emit-llvm-bc | opt -std-compile-opts". + +//===---------------------------------------------------------------------===// + +int a(int x) {return x ^ ((x & 8) ^ 8);} +Should also combine to x | 8. Currently not optimized with "clang +-emit-llvm-bc | opt -std-compile-opts". + +//===---------------------------------------------------------------------===// + +int a(int x) {return (x & 8) == 0 ? -1 : -9;} +Should combine to (x | -9) ^ 8. Currently not optimized with "clang +-emit-llvm-bc | opt -std-compile-opts". + +//===---------------------------------------------------------------------===// + +int a(int x) {return (x & 8) == 0 ? -9 : -1;} +Should combine to x | -9. Currently not optimized with "clang +-emit-llvm-bc | opt -std-compile-opts". + +//===---------------------------------------------------------------------===// + +int a(int x) {return ((x | -9) ^ 8) & x;} +Should combine to x & -9. Currently not optimized with "clang +-emit-llvm-bc | opt -std-compile-opts". + +//===---------------------------------------------------------------------===// + +unsigned a(unsigned a) {return a * 0x11111111 >> 28 & 1;} +Should combine to "a * 0x88888888 >> 31". Currently not optimized +with "clang -emit-llvm-bc | opt -std-compile-opts". + +//===---------------------------------------------------------------------===// + +unsigned a(char* x) {if ((*x & 32) == 0) return b();} +There's an unnecessary zext in the generated code with "clang +-emit-llvm-bc | opt -std-compile-opts". + +//===---------------------------------------------------------------------===// + +unsigned a(unsigned long long x) {return 40 * (x >> 1);} +Should combine to "20 * (((unsigned)x) & -2)". Currently not +optimized with "clang -emit-llvm-bc | opt -std-compile-opts". + +//===---------------------------------------------------------------------===// + +We would like to do the following transform in the instcombiner: + + -X/C -> X/-C + +However, this isn't valid if (-X) overflows. We can implement this when we +have the concept of a "C signed subtraction" operator that which is undefined +on overflow. + +//===---------------------------------------------------------------------===// + +This was noticed in the entryblock for grokdeclarator in 403.gcc: + + %tmp = icmp eq i32 %decl_context, 4 + %decl_context_addr.0 = select i1 %tmp, i32 3, i32 %decl_context + %tmp1 = icmp eq i32 %decl_context_addr.0, 1 + %decl_context_addr.1 = select i1 %tmp1, i32 0, i32 %decl_context_addr.0 + +tmp1 should be simplified to something like: + (!tmp || decl_context == 1) + +This allows recursive simplifications, tmp1 is used all over the place in +the function, e.g. by: + + %tmp23 = icmp eq i32 %decl_context_addr.1, 0 ; [#uses=1] + %tmp24 = xor i1 %tmp1, true ; [#uses=1] + %or.cond8 = and i1 %tmp23, %tmp24 ; [#uses=1] + +later. + +//===---------------------------------------------------------------------===// + +Store sinking: This code: + +void f (int n, int *cond, int *res) { + int i; + *res = 0; + for (i = 0; i < n; i++) + if (*cond) + *res ^= 234; /* (*) */ +} + +On this function GVN hoists the fully redundant value of *res, but nothing +moves the store out. This gives us this code: + +bb: ; preds = %bb2, %entry + %.rle = phi i32 [ 0, %entry ], [ %.rle6, %bb2 ] + %i.05 = phi i32 [ 0, %entry ], [ %indvar.next, %bb2 ] + %1 = load i32* %cond, align 4 + %2 = icmp eq i32 %1, 0 + br i1 %2, label %bb2, label %bb1 + +bb1: ; preds = %bb + %3 = xor i32 %.rle, 234 + store i32 %3, i32* %res, align 4 + br label %bb2 + +bb2: ; preds = %bb, %bb1 + %.rle6 = phi i32 [ %3, %bb1 ], [ %.rle, %bb ] + %indvar.next = add i32 %i.05, 1 + %exitcond = icmp eq i32 %indvar.next, %n + br i1 %exitcond, label %return, label %bb + +DSE should sink partially dead stores to get the store out of the loop. + +Here's another partial dead case: +http://gcc.gnu.org/bugzilla/show_bug.cgi?id=12395 + +//===---------------------------------------------------------------------===// + +Scalar PRE hoists the mul in the common block up to the else: + +int test (int a, int b, int c, int g) { + int d, e; + if (a) + d = b * c; + else + d = b - c; + e = b * c + g; + return d + e; +} + +It would be better to do the mul once to reduce codesize above the if. +This is GCC PR38204. + +//===---------------------------------------------------------------------===// + +GCC PR37810 is an interesting case where we should sink load/store reload +into the if block and outside the loop, so we don't reload/store it on the +non-call path. + +for () { + *P += 1; + if () + call(); + else + ... +-> +tmp = *P +for () { + tmp += 1; + if () { + *P = tmp; + call(); + tmp = *P; + } else ... +} +*P = tmp; + +We now hoist the reload after the call (Transforms/GVN/lpre-call-wrap.ll), but +we don't sink the store. We need partially dead store sinking. + +//===---------------------------------------------------------------------===// + +[PHI TRANSLATE GEPs] + +GCC PR37166: Sinking of loads prevents SROA'ing the "g" struct on the stack +leading to excess stack traffic. This could be handled by GVN with some crazy +symbolic phi translation. The code we get looks like (g is on the stack): + +bb2: ; preds = %bb1 +.. + %9 = getelementptr %struct.f* %g, i32 0, i32 0 + store i32 %8, i32* %9, align bel %bb3 + +bb3: ; preds = %bb1, %bb2, %bb + %c_addr.0 = phi %struct.f* [ %g, %bb2 ], [ %c, %bb ], [ %c, %bb1 ] + %b_addr.0 = phi %struct.f* [ %b, %bb2 ], [ %g, %bb ], [ %b, %bb1 ] + %10 = getelementptr %struct.f* %c_addr.0, i32 0, i32 0 + %11 = load i32* %10, align 4 + +%11 is fully redundant, an in BB2 it should have the value %8. + +GCC PR33344 is a similar case. + +//===---------------------------------------------------------------------===// + +There are many load PRE testcases in testsuite/gcc.dg/tree-ssa/loadpre* in the +GCC testsuite. There are many pre testcases as ssa-pre-*.c + +//===---------------------------------------------------------------------===// + +There are some interesting cases in testsuite/gcc.dg/tree-ssa/pred-comm* in the +GCC testsuite. For example, predcom-1.c is: + + for (i = 2; i < 1000; i++) + fib[i] = (fib[i-1] + fib[i - 2]) & 0xffff; + +which compiles into: + +bb1: ; preds = %bb1, %bb1.thread + %indvar = phi i32 [ 0, %bb1.thread ], [ %0, %bb1 ] + %i.0.reg2mem.0 = add i32 %indvar, 2 + %0 = add i32 %indvar, 1 ; [#uses=3] + %1 = getelementptr [1000 x i32]* @fib, i32 0, i32 %0 + %2 = load i32* %1, align 4 ; [#uses=1] + %3 = getelementptr [1000 x i32]* @fib, i32 0, i32 %indvar + %4 = load i32* %3, align 4 ; [#uses=1] + %5 = add i32 %4, %2 ; [#uses=1] + %6 = and i32 %5, 65535 ; [#uses=1] + %7 = getelementptr [1000 x i32]* @fib, i32 0, i32 %i.0.reg2mem.0 + store i32 %6, i32* %7, align 4 + %exitcond = icmp eq i32 %0, 998 ; [#uses=1] + br i1 %exitcond, label %return, label %bb1 + +This is basically: + LOAD fib[i+1] + LOAD fib[i] + STORE fib[i+2] + +instead of handling this as a loop or other xform, all we'd need to do is teach +load PRE to phi translate the %0 add (i+1) into the predecessor as (i'+1+1) = +(i'+2) (where i' is the previous iteration of i). This would find the store +which feeds it. + +predcom-2.c is apparently the same as predcom-1.c +predcom-3.c is very similar but needs loads feeding each other instead of +store->load. +predcom-4.c seems the same as the rest. + + +//===---------------------------------------------------------------------===// + +Other simple load PRE cases: +http://gcc.gnu.org/bugzilla/show_bug.cgi?id=35287 [LPRE crit edge splitting] + +http://gcc.gnu.org/bugzilla/show_bug.cgi?id=34677 (licm does this, LPRE crit edge) + llvm-gcc t2.c -S -o - -O0 -emit-llvm | llvm-as | opt -mem2reg -simplifycfg -gvn | llvm-dis + +//===---------------------------------------------------------------------===// + +Type based alias analysis: +http://gcc.gnu.org/bugzilla/show_bug.cgi?id=14705 + +//===---------------------------------------------------------------------===// + +When GVN/PRE finds a store of float* to a must aliases pointer when expecting +an int*, it should turn it into a bitcast. This is a nice generalization of +the SROA hack that would apply to other cases, e.g.: + +int foo(int C, int *P, float X) { + if (C) { + bar(); + *P = 42; + } else + *(float*)P = X; + + return *P; +} + + +One example (that requires crazy phi translation) is: +http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16799 [BITCAST PHI TRANS] + +//===---------------------------------------------------------------------===// + +A/B get pinned to the stack because we turn an if/then into a select instead +of PRE'ing the load/store. This may be fixable in instcombine: +http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37892 + + + +Interesting missed case because of control flow flattening (should be 2 loads): +http://gcc.gnu.org/bugzilla/show_bug.cgi?id=26629 +With: llvm-gcc t2.c -S -o - -O0 -emit-llvm | llvm-as | + opt -mem2reg -gvn -instcombine | llvm-dis +we miss it because we need 1) GEP PHI TRAN, 2) CRIT EDGE 3) MULTIPLE DIFFERENT +VALS PRODUCED BY ONE BLOCK OVER DIFFERENT PATHS + +//===---------------------------------------------------------------------===// + +http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19633 +We could eliminate the branch condition here, loading from null is undefined: + +struct S { int w, x, y, z; }; +struct T { int r; struct S s; }; +void bar (struct S, int); +void foo (int a, struct T b) +{ + struct S *c = 0; + if (a) + c = &b.s; + bar (*c, a); +} + +//===---------------------------------------------------------------------===// + +simplifylibcalls should do several optimizations for strspn/strcspn: + +strcspn(x, "") -> strlen(x) +strcspn("", x) -> 0 +strspn("", x) -> 0 +strspn(x, "") -> strlen(x) +strspn(x, "a") -> strchr(x, 'a')-x + +strcspn(x, "a") -> inlined loop for up to 3 letters (similarly for strspn): + +size_t __strcspn_c3 (__const char *__s, int __reject1, int __reject2, + int __reject3) { + register size_t __result = 0; + while (__s[__result] != '\0' && __s[__result] != __reject1 && + __s[__result] != __reject2 && __s[__result] != __reject3) + ++__result; + return __result; +} + +This should turn into a switch on the character. See PR3253 for some notes on +codegen. + +456.hmmer apparently uses strcspn and strspn a lot. 471.omnetpp uses strspn. + +//===---------------------------------------------------------------------===// + +"gas" uses this idiom: + else if (strchr ("+-/*%|&^:[]()~", *intel_parser.op_string)) +.. + else if (strchr ("<>", *intel_parser.op_string) + +Those should be turned into a switch. + +//===---------------------------------------------------------------------===// + +252.eon contains this interesting code: + + %3072 = getelementptr [100 x i8]* %tempString, i32 0, i32 0 + %3073 = call i8* @strcpy(i8* %3072, i8* %3071) nounwind + %strlen = call i32 @strlen(i8* %3072) ; uses = 1 + %endptr = getelementptr [100 x i8]* %tempString, i32 0, i32 %strlen + call void @llvm.memcpy.i32(i8* %endptr, + i8* getelementptr ([5 x i8]* @"\01LC42", i32 0, i32 0), i32 5, i32 1) + %3074 = call i32 @strlen(i8* %endptr) nounwind readonly + +This is interesting for a couple reasons. First, in this: + + %3073 = call i8* @strcpy(i8* %3072, i8* %3071) nounwind + %strlen = call i32 @strlen(i8* %3072) + +The strlen could be replaced with: %strlen = sub %3072, %3073, because the +strcpy call returns a pointer to the end of the string. Based on that, the +endptr GEP just becomes equal to 3073, which eliminates a strlen call and GEP. + +Second, the memcpy+strlen strlen can be replaced with: + + %3074 = call i32 @strlen([5 x i8]* @"\01LC42") nounwind readonly + +Because the destination was just copied into the specified memory buffer. This, +in turn, can be constant folded to "4". + +In other code, it contains: + + %endptr6978 = bitcast i8* %endptr69 to i32* + store i32 7107374, i32* %endptr6978, align 1 + %3167 = call i32 @strlen(i8* %endptr69) nounwind readonly + +Which could also be constant folded. Whatever is producing this should probably +be fixed to leave this as a memcpy from a string. + +Further, eon also has an interesting partially redundant strlen call: + +bb8: ; preds = %_ZN18eonImageCalculatorC1Ev.exit + %682 = getelementptr i8** %argv, i32 6 ; [#uses=2] + %683 = load i8** %682, align 4 ; [#uses=4] + %684 = load i8* %683, align 1 ; [#uses=1] + %685 = icmp eq i8 %684, 0 ; [#uses=1] + br i1 %685, label %bb10, label %bb9 + +bb9: ; preds = %bb8 + %686 = call i32 @strlen(i8* %683) nounwind readonly + %687 = icmp ugt i32 %686, 254 ; [#uses=1] + br i1 %687, label %bb10, label %bb11 + +bb10: ; preds = %bb9, %bb8 + %688 = call i32 @strlen(i8* %683) nounwind readonly + +This could be eliminated by doing the strlen once in bb8, saving code size and +improving perf on the bb8->9->10 path. + +//===---------------------------------------------------------------------===// + +I see an interesting fully redundant call to strlen left in 186.crafty:InputMove +which looks like: + %movetext11 = getelementptr [128 x i8]* %movetext, i32 0, i32 0 + + +bb62: ; preds = %bb55, %bb53 + %promote.0 = phi i32 [ %169, %bb55 ], [ 0, %bb53 ] + %171 = call i32 @strlen(i8* %movetext11) nounwind readonly align 1 + %172 = add i32 %171, -1 ; [#uses=1] + %173 = getelementptr [128 x i8]* %movetext, i32 0, i32 %172 + +... no stores ... + br i1 %or.cond, label %bb65, label %bb72 + +bb65: ; preds = %bb62 + store i8 0, i8* %173, align 1 + br label %bb72 + +bb72: ; preds = %bb65, %bb62 + %trank.1 = phi i32 [ %176, %bb65 ], [ -1, %bb62 ] + %177 = call i32 @strlen(i8* %movetext11) nounwind readonly align 1 + +Note that on the bb62->bb72 path, that the %177 strlen call is partially +redundant with the %171 call. At worst, we could shove the %177 strlen call +up into the bb65 block moving it out of the bb62->bb72 path. However, note +that bb65 stores to the string, zeroing out the last byte. This means that on +that path the value of %177 is actually just %171-1. A sub is cheaper than a +strlen! + +This pattern repeats several times, basically doing: + + A = strlen(P); + P[A-1] = 0; + B = strlen(P); + where it is "obvious" that B = A-1. + +//===---------------------------------------------------------------------===// + +186.crafty contains this interesting pattern: + +%77 = call i8* @strstr(i8* getelementptr ([6 x i8]* @"\01LC5", i32 0, i32 0), + i8* %30) +%phitmp648 = icmp eq i8* %77, getelementptr ([6 x i8]* @"\01LC5", i32 0, i32 0) +br i1 %phitmp648, label %bb70, label %bb76 + +bb70: ; preds = %OptionMatch.exit91, %bb69 + %78 = call i32 @strlen(i8* %30) nounwind readonly align 1 ; [#uses=1] + +This is basically: + cststr = "abcdef"; + if (strstr(cststr, P) == cststr) { + x = strlen(P); + ... + +The strstr call would be significantly cheaper written as: + +cststr = "abcdef"; +if (memcmp(P, str, strlen(P))) + x = strlen(P); + +This is memcmp+strlen instead of strstr. This also makes the strlen fully +redundant. + +//===---------------------------------------------------------------------===// + +186.crafty also contains this code: + +%1906 = call i32 @strlen(i8* getelementptr ([32 x i8]* @pgn_event, i32 0,i32 0)) +%1907 = getelementptr [32 x i8]* @pgn_event, i32 0, i32 %1906 +%1908 = call i8* @strcpy(i8* %1907, i8* %1905) nounwind align 1 +%1909 = call i32 @strlen(i8* getelementptr ([32 x i8]* @pgn_event, i32 0,i32 0)) +%1910 = getelementptr [32 x i8]* @pgn_event, i32 0, i32 %1909 + +The last strlen is computable as 1908-@pgn_event, which means 1910=1908. + +//===---------------------------------------------------------------------===// + +186.crafty has this interesting pattern with the "out.4543" variable: + +call void @llvm.memcpy.i32( + i8* getelementptr ([10 x i8]* @out.4543, i32 0, i32 0), + i8* getelementptr ([7 x i8]* @"\01LC28700", i32 0, i32 0), i32 7, i32 1) +%101 = call@printf(i8* ... @out.4543, i32 0, i32 0)) nounwind + +It is basically doing: + + memcpy(globalarray, "string"); + printf(..., globalarray); + +Anyway, by knowing that printf just reads the memory and forward substituting +the string directly into the printf, this eliminates reads from globalarray. +Since this pattern occurs frequently in crafty (due to the "DisplayTime" and +other similar functions) there are many stores to "out". Once all the printfs +stop using "out", all that is left is the memcpy's into it. This should allow +globalopt to remove the "stored only" global. + +//===---------------------------------------------------------------------===// + +This code: + +define inreg i32 @foo(i8* inreg %p) nounwind { + %tmp0 = load i8* %p + %tmp1 = ashr i8 %tmp0, 5 + %tmp2 = sext i8 %tmp1 to i32 + ret i32 %tmp2 +} + +could be dagcombine'd to a sign-extending load with a shift. +For example, on x86 this currently gets this: + + movb (%eax), %al + sarb $5, %al + movsbl %al, %eax + +while it could get this: + + movsbl (%eax), %eax + sarl $5, %eax + +//===---------------------------------------------------------------------===// + +GCC PR31029: + +int test(int x) { return 1-x == x; } // --> return false +int test2(int x) { return 2-x == x; } // --> return x == 1 ? + +Always foldable for odd constants, what is the rule for even? + +//===---------------------------------------------------------------------===// + +PR 3381: GEP to field of size 0 inside a struct could be turned into GEP +for next field in struct (which is at same address). + +For example: store of float into { {{}}, float } could be turned into a store to +the float directly. + +//===---------------------------------------------------------------------===// +#include +double foo(double a) { return sin(a); } + +This compiles into this on x86-64 Linux: +foo: + subq $8, %rsp + call sin + addq $8, %rsp + ret +vs: + +foo: + jmp sin +