mirror of
https://github.com/RPCS3/llvm.git
synced 2025-01-25 03:55:25 +00:00
ba05f728b5
movd always clear the top 96 bits and movss does so when it's loading the value from memory. The net result is codegen for 4-wide shuffles is much improved. It is near optimal if one or more elements is a zero. e.g. __m128i test(int a, int b) { return _mm_set_epi32(0, 0, b, a); } compiles to _test: movd 8(%esp), %xmm1 movd 4(%esp), %xmm0 punpckldq %xmm1, %xmm0 ret compare to gcc: _test: subl $12, %esp movd 20(%esp), %xmm0 movd 16(%esp), %xmm1 punpckldq %xmm0, %xmm1 movq %xmm1, %xmm0 movhps LC0, %xmm0 addl $12, %esp ret or icc: _test: movd 4(%esp), %xmm0 #5.10 movd 8(%esp), %xmm3 #5.10 xorl %eax, %eax #5.10 movd %eax, %xmm1 #5.10 punpckldq %xmm1, %xmm0 #5.10 movd %eax, %xmm2 #5.10 punpckldq %xmm2, %xmm3 #5.10 punpckldq %xmm3, %xmm0 #5.10 ret #5.10 There are still room for improvement, for example the FP variant of the above example: __m128 test(float a, float b) { return _mm_set_ps(0.0, 0.0, b, a); } _test: movss 8(%esp), %xmm1 movss 4(%esp), %xmm0 unpcklps %xmm1, %xmm0 xorps %xmm1, %xmm1 movlhps %xmm1, %xmm0 ret The xorps and movlhps are unnecessary. This will require post legalizer optimization to handle. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27939 91177308-0d34-0410-b5e6-96231b3b80d8
Target Independent Opportunities: ===-------------------------------------------------------------------------=== FreeBench/mason contains code like this: static p_type m0u(p_type p) { int m[]={0, 8, 1, 2, 16, 5, 13, 7, 14, 9, 3, 4, 11, 12, 15, 10, 17, 6}; p_type pu; pu.a = m[p.a]; pu.b = m[p.b]; pu.c = m[p.c]; return pu; } We currently compile this into a memcpy from a static array into 'm', then a bunch of loads from m. It would be better to avoid the memcpy and just do loads from the static array. //===---------------------------------------------------------------------===// 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. :) //===---------------------------------------------------------------------===// Solve this DAG isel folding deficiency: int X, Y; void fn1(void) { X = X | (Y << 3); } compiles to fn1: movl Y, %eax shll $3, %eax orl X, %eax movl %eax, X ret The problem is the store's chain operand is not the load X but rather a TokenFactor of the load X and load Y, which prevents the folding. There are two ways to fix this: 1. The dag combiner can start using alias analysis to realize that y/x don't alias, making the store to X not dependent on the load from Y. 2. The generated isel could be made smarter in the case it can't disambiguate the pointers. Number 1 is the preferred solution. This has been "fixed" by a TableGen hack. But that is a short term workaround which will be removed once the proper fix is made. //===---------------------------------------------------------------------===// Turn this into a signed shift right in instcombine: int f(unsigned x) { return x >> 31 ? -1 : 0; } http://gcc.gnu.org/bugzilla/show_bug.cgi?id=25600 http://gcc.gnu.org/ml/gcc-patches/2006-02/msg01492.html //===---------------------------------------------------------------------===// On targets with expensive 64-bit multiply, we could LSR this: for (i = ...; ++i) { x = 1ULL << i; into: long long tmp = 1; for (i = ...; ++i, tmp+=tmp) x = tmp; This would be a win on ppc32, but not x86 or ppc64. //===---------------------------------------------------------------------===// Shrink: (setlt (loadi32 P), 0) -> (setlt (loadi8 Phi), 0) //===---------------------------------------------------------------------===// Reassociate should turn: X*X*X*X -> t=(X*X) (t*t) to eliminate a multiply. //===---------------------------------------------------------------------===// Interesting? testcase for add/shift/mul reassoc: int bar(int x, int y) { return x*x*x+y+x*x*x*x*x*y*y*y*y; } int foo(int z, int n) { return bar(z, n) + bar(2*z, 2*n); } //===---------------------------------------------------------------------===// These two functions should generate the same code on big-endian systems: int g(int *j,int *l) { return memcmp(j,l,4); } int h(int *j, int *l) { return *j - *l; } this could be done in SelectionDAGISel.cpp, along with other special cases, for 1,2,4,8 bytes. //===---------------------------------------------------------------------===// This code: int rot(unsigned char b) { int a = ((b>>1) ^ (b<<7)) & 0xff; return a; } Can be improved in two ways: 1. The instcombiner should eliminate the type conversions. 2. The X86 backend should turn this into a rotate by one bit. //===---------------------------------------------------------------------===// Add LSR exit value substitution. It'll probably be a win for Ackermann, etc. //===---------------------------------------------------------------------===// It would be nice to revert this patch: http://lists.cs.uiuc.edu/pipermail/llvm-commits/Week-of-Mon-20060213/031986.html And teach the dag combiner enough to simplify the code expanded before legalize. It seems plausible that this knowledge would let it simplify other stuff too. //===---------------------------------------------------------------------===// The loop unroller should be enhanced to be able to unroll loops that aren't single basic blocks. It should be able to handle stuff like this: for (i = 0; i < c1; ++i) if (c2 & (1 << i)) foo where c1/c2 are constants. //===---------------------------------------------------------------------===// For packed types, TargetData.cpp::getTypeInfo() returns alignment that is equal to the type size. It works but can be overly conservative as the alignment of specific packed types are target dependent. //===---------------------------------------------------------------------===// We should add 'unaligned load/store' nodes, and produce them from code like this: v4sf example(float *P) { return (v4sf){P[0], P[1], P[2], P[3] }; } //===---------------------------------------------------------------------===// We should constant fold packed type casts at the LLVM level, regardless of the cast. Currently we cannot fold some casts because we don't have TargetData information in the constant folder, so we don't know the endianness of the target! //===---------------------------------------------------------------------===// Consider this: unsigned short swap_16(unsigned short v) { return (v>>8) | (v<<8); } Compiled with the ppc backend: _swap_16: slwi r2, r3, 8 srwi r3, r3, 8 or r2, r3, r2 rlwinm r3, r2, 0, 16, 31 blr The rlwinm (an and by 65535) is dead. The dag combiner should propagate bits better than that to see this. //===---------------------------------------------------------------------===//