mirror of
https://github.com/c64scene-ar/llvm-6502.git
synced 2024-12-15 04:30:12 +00:00
d7ec518927
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27963 91177308-0d34-0410-b5e6-96231b3b80d8 |
||
---|---|---|
.. | ||
.cvsignore | ||
Makefile | ||
README.txt | ||
X86.h | ||
X86.td | ||
X86AsmPrinter.cpp | ||
X86AsmPrinter.h | ||
X86ATTAsmPrinter.cpp | ||
X86ATTAsmPrinter.h | ||
X86CodeEmitter.cpp | ||
X86ELFWriter.cpp | ||
X86FloatingPoint.cpp | ||
X86InstrBuilder.h | ||
X86InstrFPStack.td | ||
X86InstrInfo.cpp | ||
X86InstrInfo.h | ||
X86InstrInfo.td | ||
X86InstrMMX.td | ||
X86InstrSSE.td | ||
X86IntelAsmPrinter.cpp | ||
X86IntelAsmPrinter.h | ||
X86ISelDAGToDAG.cpp | ||
X86ISelLowering.cpp | ||
X86ISelLowering.h | ||
X86JITInfo.cpp | ||
X86JITInfo.h | ||
X86RegisterInfo.cpp | ||
X86RegisterInfo.h | ||
X86RegisterInfo.td | ||
X86Relocations.h | ||
X86Subtarget.cpp | ||
X86Subtarget.h | ||
X86TargetMachine.cpp | ||
X86TargetMachine.h |
//===---------------------------------------------------------------------===// // Random ideas for the X86 backend. //===---------------------------------------------------------------------===// Add a MUL2U and MUL2S nodes to represent a multiply that returns both the Hi and Lo parts (combination of MUL and MULH[SU] into one node). Add this to X86, & make the dag combiner produce it when needed. This will eliminate one imul from the code generated for: long long test(long long X, long long Y) { return X*Y; } by using the EAX result from the mul. We should add a similar node for DIVREM. another case is: long long test(int X, int Y) { return (long long)X*Y; } ... which should only be one imul instruction. //===---------------------------------------------------------------------===// This should be one DIV/IDIV instruction, not a libcall: unsigned test(unsigned long long X, unsigned Y) { return X/Y; } This can be done trivially with a custom legalizer. What about overflow though? http://gcc.gnu.org/bugzilla/show_bug.cgi?id=14224 //===---------------------------------------------------------------------===// Some targets (e.g. athlons) prefer freep to fstp ST(0): http://gcc.gnu.org/ml/gcc-patches/2004-04/msg00659.html //===---------------------------------------------------------------------===// This should use fiadd on chips where it is profitable: double foo(double P, int *I) { return P+*I; } We have fiadd patterns now but the followings have the same cost and complexity. We need a way to specify the later is more profitable. def FpADD32m : FpI<(ops RFP:$dst, RFP:$src1, f32mem:$src2), OneArgFPRW, [(set RFP:$dst, (fadd RFP:$src1, (extloadf64f32 addr:$src2)))]>; // ST(0) = ST(0) + [mem32] def FpIADD32m : FpI<(ops RFP:$dst, RFP:$src1, i32mem:$src2), OneArgFPRW, [(set RFP:$dst, (fadd RFP:$src1, (X86fild addr:$src2, i32)))]>; // ST(0) = ST(0) + [mem32int] //===---------------------------------------------------------------------===// The FP stackifier needs to be global. Also, it should handle simple permutates to reduce number of shuffle instructions, e.g. turning: fld P -> fld Q fld Q fld P fxch or: fxch -> fucomi fucomi jl X jg X Ideas: http://gcc.gnu.org/ml/gcc-patches/2004-11/msg02410.html //===---------------------------------------------------------------------===// Improvements to the multiply -> shift/add algorithm: http://gcc.gnu.org/ml/gcc-patches/2004-08/msg01590.html //===---------------------------------------------------------------------===// Improve code like this (occurs fairly frequently, e.g. in LLVM): long long foo(int x) { return 1LL << x; } http://gcc.gnu.org/ml/gcc-patches/2004-09/msg01109.html http://gcc.gnu.org/ml/gcc-patches/2004-09/msg01128.html http://gcc.gnu.org/ml/gcc-patches/2004-09/msg01136.html Another useful one would be ~0ULL >> X and ~0ULL << X. //===---------------------------------------------------------------------===// Compile this: _Bool f(_Bool a) { return a!=1; } into: movzbl %dil, %eax xorl $1, %eax ret //===---------------------------------------------------------------------===// Some isel ideas: 1. Dynamic programming based approach when compile time if not an issue. 2. Code duplication (addressing mode) during isel. 3. Other ideas from "Register-Sensitive Selection, Duplication, and Sequencing of Instructions". 4. Scheduling for reduced register pressure. E.g. "Minimum Register Instruction Sequence Problem: Revisiting Optimal Code Generation for DAGs" and other related papers. http://citeseer.ist.psu.edu/govindarajan01minimum.html //===---------------------------------------------------------------------===// Should we promote i16 to i32 to avoid partial register update stalls? //===---------------------------------------------------------------------===// Leave any_extend as pseudo instruction and hint to register allocator. Delay codegen until post register allocation. //===---------------------------------------------------------------------===// Add a target specific hook to DAG combiner to handle SINT_TO_FP and FP_TO_SINT when the source operand is already in memory. //===---------------------------------------------------------------------===// Model X86 EFLAGS as a real register to avoid redudant cmp / test. e.g. cmpl $1, %eax setg %al testb %al, %al # unnecessary jne .BB7 //===---------------------------------------------------------------------===// Count leading zeros and count trailing zeros: int clz(int X) { return __builtin_clz(X); } int ctz(int X) { return __builtin_ctz(X); } $ gcc t.c -S -o - -O3 -fomit-frame-pointer -masm=intel clz: bsr %eax, DWORD PTR [%esp+4] xor %eax, 31 ret ctz: bsf %eax, DWORD PTR [%esp+4] ret however, check that these are defined for 0 and 32. Our intrinsics are, GCC's aren't. //===---------------------------------------------------------------------===// Use push/pop instructions in prolog/epilog sequences instead of stores off ESP (certain code size win, perf win on some [which?] processors). Also, it appears icc use push for parameter passing. Need to investigate. //===---------------------------------------------------------------------===// Only use inc/neg/not instructions on processors where they are faster than add/sub/xor. They are slower on the P4 due to only updating some processor flags. //===---------------------------------------------------------------------===// Open code rint,floor,ceil,trunc: http://gcc.gnu.org/ml/gcc-patches/2004-08/msg02006.html http://gcc.gnu.org/ml/gcc-patches/2004-08/msg02011.html //===---------------------------------------------------------------------===// Combine: a = sin(x), b = cos(x) into a,b = sincos(x). Expand these to calls of sin/cos and stores: double sincos(double x, double *sin, double *cos); float sincosf(float x, float *sin, float *cos); long double sincosl(long double x, long double *sin, long double *cos); Doing so could allow SROA of the destination pointers. See also: http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17687 //===---------------------------------------------------------------------===// The instruction selector sometimes misses folding a load into a compare. The pattern is written as (cmp reg, (load p)). Because the compare isn't commutative, it is not matched with the load on both sides. The dag combiner should be made smart enough to cannonicalize the load into the RHS of a compare when it can invert the result of the compare for free. How about intrinsics? An example is: *res = _mm_mulhi_epu16(*A, _mm_mul_epu32(*B, *C)); compiles to pmuludq (%eax), %xmm0 movl 8(%esp), %eax movdqa (%eax), %xmm1 pmulhuw %xmm0, %xmm1 The transformation probably requires a X86 specific pass or a DAG combiner target specific hook. //===---------------------------------------------------------------------===// LSR should be turned on for the X86 backend and tuned to take advantage of its addressing modes. //===---------------------------------------------------------------------===// When compiled with unsafemath enabled, "main" should enable SSE DAZ mode and other fast SSE modes. //===---------------------------------------------------------------------===// Think about doing i64 math in SSE regs. //===---------------------------------------------------------------------===// The DAG Isel doesn't fold the loads into the adds in this testcase. The pattern selector does. This is because the chain value of the load gets selected first, and the loads aren't checking to see if they are only used by and add. .ll: int %test(int* %x, int* %y, int* %z) { %X = load int* %x %Y = load int* %y %Z = load int* %z %a = add int %X, %Y %b = add int %a, %Z ret int %b } dag isel: _test: movl 4(%esp), %eax movl (%eax), %eax movl 8(%esp), %ecx movl (%ecx), %ecx addl %ecx, %eax movl 12(%esp), %ecx movl (%ecx), %ecx addl %ecx, %eax ret pattern isel: _test: movl 12(%esp), %ecx movl 4(%esp), %edx movl 8(%esp), %eax movl (%eax), %eax addl (%edx), %eax addl (%ecx), %eax ret This is bad for register pressure, though the dag isel is producing a better schedule. :) //===---------------------------------------------------------------------===// This testcase should have no SSE instructions in it, and only one load from a constant pool: double %test3(bool %B) { %C = select bool %B, double 123.412, double 523.01123123 ret double %C } Currently, the select is being lowered, which prevents the dag combiner from turning 'select (load CPI1), (load CPI2)' -> 'load (select CPI1, CPI2)' The pattern isel got this one right. //===---------------------------------------------------------------------===// We need to lower switch statements to tablejumps when appropriate instead of always into binary branch trees. //===---------------------------------------------------------------------===// SSE doesn't have [mem] op= reg instructions. If we have an SSE instruction like this: X += y and the register allocator decides to spill X, it is cheaper to emit this as: Y += [xslot] store Y -> [xslot] than as: tmp = [xslot] tmp += y store tmp -> [xslot] ..and this uses one fewer register (so this should be done at load folding time, not at spiller time). *Note* however that this can only be done if Y is dead. Here's a testcase: %.str_3 = external global [15 x sbyte] ; <[15 x sbyte]*> [#uses=0] implementation ; Functions: declare void %printf(int, ...) void %main() { build_tree.exit: br label %no_exit.i7 no_exit.i7: ; preds = %no_exit.i7, %build_tree.exit %tmp.0.1.0.i9 = phi double [ 0.000000e+00, %build_tree.exit ], [ %tmp.34.i18, %no_exit.i7 ] ; <double> [#uses=1] %tmp.0.0.0.i10 = phi double [ 0.000000e+00, %build_tree.exit ], [ %tmp.28.i16, %no_exit.i7 ] ; <double> [#uses=1] %tmp.28.i16 = add double %tmp.0.0.0.i10, 0.000000e+00 %tmp.34.i18 = add double %tmp.0.1.0.i9, 0.000000e+00 br bool false, label %Compute_Tree.exit23, label %no_exit.i7 Compute_Tree.exit23: ; preds = %no_exit.i7 tail call void (int, ...)* %printf( int 0 ) store double %tmp.34.i18, double* null ret void } We currently emit: .BBmain_1: xorpd %XMM1, %XMM1 addsd %XMM0, %XMM1 *** movsd %XMM2, QWORD PTR [%ESP + 8] *** addsd %XMM2, %XMM1 *** movsd QWORD PTR [%ESP + 8], %XMM2 jmp .BBmain_1 # no_exit.i7 This is a bugpoint reduced testcase, which is why the testcase doesn't make much sense (e.g. its an infinite loop). :) //===---------------------------------------------------------------------===// None of the FPStack instructions are handled in X86RegisterInfo::foldMemoryOperand, which prevents the spiller from folding spill code into the instructions. //===---------------------------------------------------------------------===// In many cases, LLVM generates code like this: _test: movl 8(%esp), %eax cmpl %eax, 4(%esp) setl %al movzbl %al, %eax ret on some processors (which ones?), it is more efficient to do this: _test: movl 8(%esp), %ebx xor %eax, %eax cmpl %ebx, 4(%esp) setl %al ret Doing this correctly is tricky though, as the xor clobbers the flags. //===---------------------------------------------------------------------===// We should generate 'test' instead of 'cmp' in various cases, e.g.: bool %test(int %X) { %Y = shl int %X, ubyte 1 %C = seteq int %Y, 0 ret bool %C } bool %test(int %X) { %Y = and int %X, 8 %C = seteq int %Y, 0 ret bool %C } This may just be a matter of using 'test' to write bigger patterns for X86cmp. //===---------------------------------------------------------------------===// SSE should implement 'select_cc' using 'emulated conditional moves' that use pcmp/pand/pandn/por to do a selection instead of a conditional branch: double %X(double %Y, double %Z, double %A, double %B) { %C = setlt double %A, %B %z = add double %Z, 0.0 ;; select operand is not a load %D = select bool %C, double %Y, double %z ret double %D } We currently emit: _X: subl $12, %esp xorpd %xmm0, %xmm0 addsd 24(%esp), %xmm0 movsd 32(%esp), %xmm1 movsd 16(%esp), %xmm2 ucomisd 40(%esp), %xmm1 jb LBB_X_2 LBB_X_1: movsd %xmm0, %xmm2 LBB_X_2: movsd %xmm2, (%esp) fldl (%esp) addl $12, %esp ret //===---------------------------------------------------------------------===// We should generate bts/btr/etc instructions on targets where they are cheap or when codesize is important. e.g., for: void setbit(int *target, int bit) { *target |= (1 << bit); } void clearbit(int *target, int bit) { *target &= ~(1 << bit); } //===---------------------------------------------------------------------===// Instead of the following for memset char*, 1, 10: movl $16843009, 4(%edx) movl $16843009, (%edx) movw $257, 8(%edx) It might be better to generate movl $16843009, %eax movl %eax, 4(%edx) movl %eax, (%edx) movw al, 8(%edx) when we can spare a register. It reduces code size. //===---------------------------------------------------------------------===// It's not clear whether we should use pxor or xorps / xorpd to clear XMM registers. The choice may depend on subtarget information. We should do some more experiments on different x86 machines. //===---------------------------------------------------------------------===// Evaluate what the best way to codegen sdiv X, (2^C) is. For X/8, we currently get this: int %test1(int %X) { %Y = div int %X, 8 ret int %Y } _test1: movl 4(%esp), %eax movl %eax, %ecx sarl $31, %ecx shrl $29, %ecx addl %ecx, %eax sarl $3, %eax ret GCC knows several different ways to codegen it, one of which is this: _test1: movl 4(%esp), %eax cmpl $-1, %eax leal 7(%eax), %ecx cmovle %ecx, %eax sarl $3, %eax ret which is probably slower, but it's interesting at least :) //===---------------------------------------------------------------------===// Currently the x86 codegen isn't very good at mixing SSE and FPStack code: unsigned int foo(double x) { return x; } foo: subl $20, %esp movsd 24(%esp), %xmm0 movsd %xmm0, 8(%esp) fldl 8(%esp) fisttpll (%esp) movl (%esp), %eax addl $20, %esp ret This will be solved when we go to a dynamic programming based isel. //===---------------------------------------------------------------------===// Should generate min/max for stuff like: void minf(float a, float b, float *X) { *X = a <= b ? a : b; } Make use of floating point min / max instructions. Perhaps introduce ISD::FMIN and ISD::FMAX node types? //===---------------------------------------------------------------------===// The first BB of this code: declare bool %foo() int %bar() { %V = call bool %foo() br bool %V, label %T, label %F T: ret int 1 F: call bool %foo() ret int 12 } compiles to: _bar: subl $12, %esp call L_foo$stub xorb $1, %al testb %al, %al jne LBB_bar_2 # F It would be better to emit "cmp %al, 1" than a xor and test. //===---------------------------------------------------------------------===// Enable X86InstrInfo::convertToThreeAddress(). //===---------------------------------------------------------------------===// Investigate whether it is better to codegen the following %tmp.1 = mul int %x, 9 to movl 4(%esp), %eax leal (%eax,%eax,8), %eax as opposed to what llc is currently generating: imull $9, 4(%esp), %eax Currently the load folding imull has a higher complexity than the LEA32 pattern. //===---------------------------------------------------------------------===// We are currently lowering large (1MB+) memmove/memcpy to rep/stosl and rep/movsl We should leave these as libcalls for everything over a much lower threshold, since libc is hand tuned for medium and large mem ops (avoiding RFO for large stores, TLB preheating, etc) //===---------------------------------------------------------------------===// Lower memcpy / memset to a series of SSE 128 bit move instructions when it's feasible. //===---------------------------------------------------------------------===// Teach the coalescer to commute 2-addr instructions, allowing us to eliminate the reg-reg copy in this example: float foo(int *x, float *y, unsigned c) { float res = 0.0; unsigned i; for (i = 0; i < c; i++) { float xx = (float)x[i]; xx = xx * y[i]; xx += res; res = xx; } return res; } LBB_foo_3: # no_exit cvtsi2ss %XMM0, DWORD PTR [%EDX + 4*%ESI] mulss %XMM0, DWORD PTR [%EAX + 4*%ESI] addss %XMM0, %XMM1 inc %ESI cmp %ESI, %ECX **** movaps %XMM1, %XMM0 jb LBB_foo_3 # no_exit //===---------------------------------------------------------------------===// Codegen: if (copysign(1.0, x) == copysign(1.0, y)) into: if (x^y & mask) when using SSE. //===---------------------------------------------------------------------===// Optimize this into something reasonable: x * copysign(1.0, y) * copysign(1.0, z) //===---------------------------------------------------------------------===// Optimize copysign(x, *y) to use an integer load from y. //===---------------------------------------------------------------------===// %X = weak global int 0 void %foo(int %N) { %N = cast int %N to uint %tmp.24 = setgt int %N, 0 br bool %tmp.24, label %no_exit, label %return no_exit: %indvar = phi uint [ 0, %entry ], [ %indvar.next, %no_exit ] %i.0.0 = cast uint %indvar to int volatile store int %i.0.0, int* %X %indvar.next = add uint %indvar, 1 %exitcond = seteq uint %indvar.next, %N br bool %exitcond, label %return, label %no_exit return: ret void } compiles into: .text .align 4 .globl _foo _foo: movl 4(%esp), %eax cmpl $1, %eax jl LBB_foo_4 # return LBB_foo_1: # no_exit.preheader xorl %ecx, %ecx LBB_foo_2: # no_exit movl L_X$non_lazy_ptr, %edx movl %ecx, (%edx) incl %ecx cmpl %eax, %ecx jne LBB_foo_2 # no_exit LBB_foo_3: # return.loopexit LBB_foo_4: # return ret We should hoist "movl L_X$non_lazy_ptr, %edx" out of the loop after remateralization is implemented. This can be accomplished with 1) a target dependent LICM pass or 2) makeing SelectDAG represent the whole function. //===---------------------------------------------------------------------===// The following tests perform worse with LSR: lambda, siod, optimizer-eval, ackermann, hash2, nestedloop, strcat, and Treesor. //===---------------------------------------------------------------------===// Teach the coalescer to coalesce vregs of different register classes. e.g. FR32 / FR64 to VR128. //===---------------------------------------------------------------------===// mov $reg, 48(%esp) ... leal 48(%esp), %eax mov %eax, (%esp) call _foo Obviously it would have been better for the first mov (or any op) to store directly %esp[0] if there are no other uses. //===---------------------------------------------------------------------===// Use movhps to update upper 64-bits of a v4sf value. Also movlps on lower half of a v4sf value. //===---------------------------------------------------------------------===// Better codegen for vector_shuffles like this { x, 0, 0, 0 } or { x, 0, x, 0}. Perhaps use pxor / xorp* to clear a XMM register first? //===---------------------------------------------------------------------===// Better codegen for: void f(float a, float b, vector float * out) { *out = (vector float){ a, 0.0, 0.0, b}; } void f(float a, float b, vector float * out) { *out = (vector float){ a, b, 0.0, 0}; } For the later we generate: _f: pxor %xmm0, %xmm0 movss 8(%esp), %xmm1 movaps %xmm0, %xmm2 unpcklps %xmm1, %xmm2 movss 4(%esp), %xmm1 unpcklps %xmm0, %xmm1 unpcklps %xmm2, %xmm1 movl 12(%esp), %eax movaps %xmm1, (%eax) ret This seems like it should use shufps, one for each of a & b. //===---------------------------------------------------------------------===// Adding to the list of cmp / test poor codegen issues: int test(__m128 *A, __m128 *B) { if (_mm_comige_ss(*A, *B)) return 3; else return 4; } _test: movl 8(%esp), %eax movaps (%eax), %xmm0 movl 4(%esp), %eax movaps (%eax), %xmm1 comiss %xmm0, %xmm1 setae %al movzbl %al, %ecx movl $3, %eax movl $4, %edx cmpl $0, %ecx cmove %edx, %eax ret Note the setae, movzbl, cmpl, cmove can be replaced with a single cmovae. There are a number of issues. 1) We are introducing a setcc between the result of the intrisic call and select. 2) The intrinsic is expected to produce a i32 value so a any extend (which becomes a zero extend) is added. We probably need some kind of target DAG combine hook to fix this. //===---------------------------------------------------------------------===// How to decide when to use the "floating point version" of logical ops? Here are some code fragments: movaps LCPI5_5, %xmm2 divps %xmm1, %xmm2 mulps %xmm2, %xmm3 mulps 8656(%ecx), %xmm3 addps 8672(%ecx), %xmm3 andps LCPI5_6, %xmm2 andps LCPI5_1, %xmm3 por %xmm2, %xmm3 movdqa %xmm3, (%edi) movaps LCPI5_5, %xmm1 divps %xmm0, %xmm1 mulps %xmm1, %xmm3 mulps 8656(%ecx), %xmm3 addps 8672(%ecx), %xmm3 andps LCPI5_6, %xmm1 andps LCPI5_1, %xmm3 orps %xmm1, %xmm3 movaps %xmm3, 112(%esp) movaps %xmm3, (%ebx) Due to some minor source change, the later case ended up using orps and movaps instead of por and movdqa. Does it matter? //===---------------------------------------------------------------------===// Use movddup to splat a v2f64 directly from a memory source. e.g. #include <emmintrin.h> void test(__m128d *r, double A) { *r = _mm_set1_pd(A); } llc: _test: movsd 8(%esp), %xmm0 unpcklpd %xmm0, %xmm0 movl 4(%esp), %eax movapd %xmm0, (%eax) ret icc: _test: movl 4(%esp), %eax movddup 8(%esp), %xmm0 movapd %xmm0, (%eax) ret //===---------------------------------------------------------------------===// A Mac OS X IA-32 specific ABI bug wrt returning value > 8 bytes: http://llvm.org/bugs/show_bug.cgi?id=729 //===---------------------------------------------------------------------===// X86RegisterInfo::copyRegToReg() returns X86::MOVAPSrr for VR128. Is it possible to choose between movaps, movapd, and movdqa based on types of source and destination? How about andps, andpd, and pand? Do we really care about the type of the packed elements? If not, why not always use the "ps" variants which are likely to be shorter. //===---------------------------------------------------------------------===// We are emitting bad code for this: float %test(float* %V, int %I, int %D, float %V) { entry: %tmp = seteq int %D, 0 br bool %tmp, label %cond_true, label %cond_false23 cond_true: %tmp3 = getelementptr float* %V, int %I %tmp = load float* %tmp3 %tmp5 = setgt float %tmp, %V %tmp6 = tail call bool %llvm.isunordered.f32( float %tmp, float %V ) %tmp7 = or bool %tmp5, %tmp6 br bool %tmp7, label %UnifiedReturnBlock, label %cond_next cond_next: %tmp10 = add int %I, 1 %tmp12 = getelementptr float* %V, int %tmp10 %tmp13 = load float* %tmp12 %tmp15 = setle float %tmp13, %V %tmp16 = tail call bool %llvm.isunordered.f32( float %tmp13, float %V ) %tmp17 = or bool %tmp15, %tmp16 %retval = select bool %tmp17, float 0.000000e+00, float 1.000000e+00 ret float %retval cond_false23: %tmp28 = tail call float %foo( float* %V, int %I, int %D, float %V ) ret float %tmp28 UnifiedReturnBlock: ; preds = %cond_true ret float 0.000000e+00 } declare bool %llvm.isunordered.f32(float, float) declare float %foo(float*, int, int, float) It exposes a known load folding problem: movss (%edx,%ecx,4), %xmm1 ucomiss %xmm1, %xmm0 As well as this: LBB_test_2: # cond_next movss LCPI1_0, %xmm2 pxor %xmm3, %xmm3 ucomiss %xmm0, %xmm1 jbe LBB_test_6 # cond_next LBB_test_5: # cond_next movaps %xmm2, %xmm3 LBB_test_6: # cond_next movss %xmm3, 40(%esp) flds 40(%esp) addl $44, %esp ret Clearly it's unnecessary to clear %xmm3. It's also not clear why we are emitting three moves (movss, movaps, movss). //===---------------------------------------------------------------------===// External test Nurbs exposed some problems. Look for __ZN15Nurbs_SSE_Cubic17TessellateSurfaceE, bb cond_next140. This is what icc emits: movaps (%edx), %xmm2 #59.21 movaps (%edx), %xmm5 #60.21 movaps (%edx), %xmm4 #61.21 movaps (%edx), %xmm3 #62.21 movl 40(%ecx), %ebp #69.49 shufps $0, %xmm2, %xmm5 #60.21 movl 100(%esp), %ebx #69.20 movl (%ebx), %edi #69.20 imull %ebp, %edi #69.49 addl (%eax), %edi #70.33 shufps $85, %xmm2, %xmm4 #61.21 shufps $170, %xmm2, %xmm3 #62.21 shufps $255, %xmm2, %xmm2 #63.21 lea (%ebp,%ebp,2), %ebx #69.49 negl %ebx #69.49 lea -3(%edi,%ebx), %ebx #70.33 shll $4, %ebx #68.37 addl 32(%ecx), %ebx #68.37 testb $15, %bl #91.13 jne L_B1.24 # Prob 5% #91.13 This is the llvm code after instruction scheduling: cond_next140 (0xa910740, LLVM BB @0xa90beb0): %reg1078 = MOV32ri -3 %reg1079 = ADD32rm %reg1078, %reg1068, 1, %NOREG, 0 %reg1037 = MOV32rm %reg1024, 1, %NOREG, 40 %reg1080 = IMUL32rr %reg1079, %reg1037 %reg1081 = MOV32rm %reg1058, 1, %NOREG, 0 %reg1038 = LEA32r %reg1081, 1, %reg1080, -3 %reg1036 = MOV32rm %reg1024, 1, %NOREG, 32 %reg1082 = SHL32ri %reg1038, 4 %reg1039 = ADD32rr %reg1036, %reg1082 %reg1083 = MOVAPSrm %reg1059, 1, %NOREG, 0 %reg1034 = SHUFPSrr %reg1083, %reg1083, 170 %reg1032 = SHUFPSrr %reg1083, %reg1083, 0 %reg1035 = SHUFPSrr %reg1083, %reg1083, 255 %reg1033 = SHUFPSrr %reg1083, %reg1083, 85 %reg1040 = MOV32rr %reg1039 %reg1084 = AND32ri8 %reg1039, 15 CMP32ri8 %reg1084, 0 JE mbb<cond_next204,0xa914d30> Still ok. After register allocation: cond_next140 (0xa910740, LLVM BB @0xa90beb0): %EAX = MOV32ri -3 %EDX = MOV32rm <fi#3>, 1, %NOREG, 0 ADD32rm %EAX<def&use>, %EDX, 1, %NOREG, 0 %EDX = MOV32rm <fi#7>, 1, %NOREG, 0 %EDX = MOV32rm %EDX, 1, %NOREG, 40 IMUL32rr %EAX<def&use>, %EDX %ESI = MOV32rm <fi#5>, 1, %NOREG, 0 %ESI = MOV32rm %ESI, 1, %NOREG, 0 MOV32mr <fi#4>, 1, %NOREG, 0, %ESI %EAX = LEA32r %ESI, 1, %EAX, -3 %ESI = MOV32rm <fi#7>, 1, %NOREG, 0 %ESI = MOV32rm %ESI, 1, %NOREG, 32 %EDI = MOV32rr %EAX SHL32ri %EDI<def&use>, 4 ADD32rr %EDI<def&use>, %ESI %XMM0 = MOVAPSrm %ECX, 1, %NOREG, 0 %XMM1 = MOVAPSrr %XMM0 SHUFPSrr %XMM1<def&use>, %XMM1, 170 %XMM2 = MOVAPSrr %XMM0 SHUFPSrr %XMM2<def&use>, %XMM2, 0 %XMM3 = MOVAPSrr %XMM0 SHUFPSrr %XMM3<def&use>, %XMM3, 255 SHUFPSrr %XMM0<def&use>, %XMM0, 85 %EBX = MOV32rr %EDI AND32ri8 %EBX<def&use>, 15 CMP32ri8 %EBX, 0 JE mbb<cond_next204,0xa914d30> This looks really bad. The problem is shufps is a destructive opcode. Since it appears as operand two in more than one shufps ops. It resulted in a number of copies. Note icc also suffers from the same problem. Either the instruction selector should select pshufd or The register allocator can made the two-address to three-address transformation. It also exposes some other problems. See MOV32ri -3 and the spills. //===---------------------------------------------------------------------===// http://gcc.gnu.org/bugzilla/show_bug.cgi?id=25500 LLVM is producing bad code. LBB_main_4: # cond_true44 addps %xmm1, %xmm2 subps %xmm3, %xmm2 movaps (%ecx), %xmm4 movaps %xmm2, %xmm1 addps %xmm4, %xmm1 addl $16, %ecx incl %edx cmpl $262144, %edx movaps %xmm3, %xmm2 movaps %xmm4, %xmm3 jne LBB_main_4 # cond_true44 There are two problems. 1) No need to two loop induction variables. We can compare against 262144 * 16. 2) Known register coalescer issue. We should be able eliminate one of the movaps: addps %xmm2, %xmm1 <=== Commute! subps %xmm3, %xmm1 movaps (%ecx), %xmm4 movaps %xmm1, %xmm1 <=== Eliminate! addps %xmm4, %xmm1 addl $16, %ecx incl %edx cmpl $262144, %edx movaps %xmm3, %xmm2 movaps %xmm4, %xmm3 jne LBB_main_4 # cond_true44 //===---------------------------------------------------------------------===// Consider: __m128 test(float a) { return _mm_set_ps(0.0, 0.0, 0.0, a*a); } This compiles into: movss 4(%esp), %xmm1 mulss %xmm1, %xmm1 xorps %xmm0, %xmm0 movss %xmm1, %xmm0 ret Because mulss doesn't modify the top 3 elements, the top elements of xmm1 are already zero'd. We could compile this to: movss 4(%esp), %xmm0 mulss %xmm0, %xmm0 ret //===---------------------------------------------------------------------===// Here's a sick and twisted idea. Consider code like this: __m128 test(__m128 a) { float b = *(float*)&A; ... return _mm_set_ps(0.0, 0.0, 0.0, b); } This might compile to this code: movaps c(%esp), %xmm1 xorps %xmm0, %xmm0 movss %xmm1, %xmm0 ret Now consider if the ... code caused xmm1 to get spilled. This might produce this code: movaps c(%esp), %xmm1 movaps %xmm1, c2(%esp) ... xorps %xmm0, %xmm0 movaps c2(%esp), %xmm1 movss %xmm1, %xmm0 ret However, since the reload is only used by these instructions, we could "fold" it into the uses, producing something like this: movaps c(%esp), %xmm1 movaps %xmm1, c2(%esp) ... movss c2(%esp), %xmm0 ret ... saving two instructions. The basic idea is that a reload from a spill slot, can, if only one 4-byte chunk is used, bring in 3 zeros the the one element instead of 4 elements. This can be used to simplify a variety of shuffle operations, where the elements are fixed zeros. //===---------------------------------------------------------------------===// We generate significantly worse code for this than GCC: http://gcc.gnu.org/bugzilla/show_bug.cgi?id=21150 http://gcc.gnu.org/bugzilla/attachment.cgi?id=8701 There is also one case we do worse on PPC. //===---------------------------------------------------------------------===// For this: #include <emmintrin.h> void test(__m128d *r, __m128d *A, double B) { *r = _mm_loadl_pd(*A, &B); } We generates: subl $12, %esp movsd 24(%esp), %xmm0 movsd %xmm0, (%esp) movl 20(%esp), %eax movapd (%eax), %xmm0 movlpd (%esp), %xmm0 movl 16(%esp), %eax movapd %xmm0, (%eax) addl $12, %esp ret icc generates: movl 4(%esp), %edx #3.6 movl 8(%esp), %eax #3.6 movapd (%eax), %xmm0 #4.22 movlpd 12(%esp), %xmm0 #4.8 movapd %xmm0, (%edx) #4.3 ret #5.1 So icc is smart enough to know that B is in memory so it doesn't load it and store it back to stack.