// Random ideas for the X86 backend: SSE-specific stuff.
//===---------------------------------------------------------------------===//
-- Consider eliminating the unaligned SSE load intrinsics, replacing them with
- unaligned LLVM load instructions.
-
-//===---------------------------------------------------------------------===//
-
-Expand libm rounding functions inline: Significant speedups possible.
-http://gcc.gnu.org/ml/gcc-patches/2006-10/msg00909.html
-
//===---------------------------------------------------------------------===//
-When compiled with unsafemath enabled, "main" should enable SSE DAZ mode and
-other fast SSE modes.
+SSE Variable shift can be custom lowered to something like this, which uses a
+small table + unaligned load + shuffle instead of going through memory.
-//===---------------------------------------------------------------------===//
-
-Think about doing i64 math in SSE regs.
-
-//===---------------------------------------------------------------------===//
-
-This testcase should have no SSE instructions in it, and only one load from
-a constant pool:
+__m128i_shift_right:
+ .byte 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15
+ .byte -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1
-double %test3(bool %B) {
- %C = select bool %B, double 123.412, double 523.01123123
- ret double %C
+...
+__m128i shift_right(__m128i value, unsigned long offset) {
+ return _mm_shuffle_epi8(value,
+ _mm_loadu_si128((__m128 *) (___m128i_shift_right + offset)));
}
-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.
-
//===---------------------------------------------------------------------===//
-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:
+SSE has instructions for doing operations on complex numbers, we should pattern
+match them. For example, this should turn into a horizontal add:
-Y += [xslot]
-store Y -> [xslot]
-
-than as:
+typedef float __attribute__((vector_size(16))) v4f32;
+float f32(v4f32 A) {
+ return A[0]+A[1]+A[2]+A[3];
+}
-tmp = [xslot]
-tmp += y
-store tmp -> [xslot]
+Instead we get this:
+
+_f32: ## @f32
+ pshufd $1, %xmm0, %xmm1 ## xmm1 = xmm0[1,0,0,0]
+ addss %xmm0, %xmm1
+ pshufd $3, %xmm0, %xmm2 ## xmm2 = xmm0[3,0,0,0]
+ movhlps %xmm0, %xmm0 ## xmm0 = xmm0[1,1]
+ movaps %xmm0, %xmm3
+ addss %xmm1, %xmm3
+ movdqa %xmm2, %xmm0
+ addss %xmm3, %xmm0
+ ret
-..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:
+Also, there are cases where some simple local SLP would improve codegen a bit.
+compiling this:
-%.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
+_Complex float f32(_Complex float A, _Complex float B) {
+ return A+B;
}
-We currently emit:
+into:
+
+_f32: ## @f32
+ movdqa %xmm0, %xmm2
+ addss %xmm1, %xmm2
+ pshufd $1, %xmm1, %xmm1 ## xmm1 = xmm1[1,0,0,0]
+ pshufd $1, %xmm0, %xmm3 ## xmm3 = xmm0[1,0,0,0]
+ addss %xmm1, %xmm3
+ movaps %xmm2, %xmm0
+ unpcklps %xmm3, %xmm0 ## xmm0 = xmm0[0],xmm3[0],xmm0[1],xmm3[1]
+ ret
-.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
+seems silly when it could just be one addps.
-This is a bugpoint reduced testcase, which is why the testcase doesn't make
-much sense (e.g. its an infinite loop). :)
//===---------------------------------------------------------------------===//
-SSE should implement 'select_cc' using 'emulated conditional moves' that use
-pcmp/pand/pandn/por to do a selection instead of a conditional branch:
+Expand libm rounding functions inline: Significant speedups possible.
+http://gcc.gnu.org/ml/gcc-patches/2006-10/msg00909.html
-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
+When compiled with unsafemath enabled, "main" should enable SSE DAZ mode and
+other fast SSE modes.
//===---------------------------------------------------------------------===//
-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.
+Think about doing i64 math in SSE regs on x86-32.
//===---------------------------------------------------------------------===//
-Currently the x86 codegen isn't very good at mixing SSE and FPStack
-code:
+This testcase should have no SSE instructions in it, and only one load from
+a constant pool:
-unsigned int foo(double x) { return x; }
+double %test3(bool %B) {
+ %C = select bool %B, double 123.412, double 523.01123123
+ ret double %C
+}
-foo:
- subl $20, %esp
- movsd 24(%esp), %xmm0
- movsd %xmm0, 8(%esp)
- fldl 8(%esp)
- fisttpll (%esp)
- movl (%esp), %eax
- addl $20, %esp
- ret
+Currently, the select is being lowered, which prevents the dag combiner from
+turning 'select (load CPI1), (load CPI2)' -> 'load (select CPI1, CPI2)'
-This will be solved when we go to a dynamic programming based isel.
+The pattern isel got this one right.
//===---------------------------------------------------------------------===//
//===---------------------------------------------------------------------===//
-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:
//===---------------------------------------------------------------------===//
-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?
-
-//===---------------------------------------------------------------------===//
-
-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.
-
-//===---------------------------------------------------------------------===//
-
External test Nurbs exposed some problems. Look for
__ZN15Nurbs_SSE_Cubic17TessellateSurfaceE, bb cond_next140. This is what icc
emits:
//===---------------------------------------------------------------------===//
-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) {
... 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.
+chunk is used, bring in 3 zeros the one element instead of 4 elements.
This can be used to simplify a variety of shuffle operations, where the
elements are fixed zeros.
//===---------------------------------------------------------------------===//
-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.
-
-//===---------------------------------------------------------------------===//
-
-__m128d test1( __m128d A, __m128d B) {
- return _mm_shuffle_pd(A, B, 0x3);
-}
-
-compiles to
-
-shufpd $3, %xmm1, %xmm0
-
-Perhaps it's better to use unpckhpd instead?
-
-unpckhpd %xmm1, %xmm0
-
-Don't know if unpckhpd is faster. But it is shorter.
-
-//===---------------------------------------------------------------------===//
-
This code generates ugly code, probably due to costs being off or something:
-void %test(float* %P, <4 x float>* %P2 ) {
+define void @test(float* %P, <4 x float>* %P2 ) {
%xFloat0.688 = load float* %P
- %loadVector37.712 = load <4 x float>* %P2
- %inFloat3.713 = insertelement <4 x float> %loadVector37.712, float 0.000000e+00, uint 3
+ %tmp = load <4 x float>* %P2
+ %inFloat3.713 = insertelement <4 x float> %tmp, float 0.0, i32 3
store <4 x float> %inFloat3.713, <4 x float>* %P2
ret void
}
Generates:
_test:
- pxor %xmm0, %xmm0
- movd %xmm0, %eax ;; EAX = 0!
- movl 8(%esp), %ecx
- movaps (%ecx), %xmm0
- pinsrw $6, %eax, %xmm0
- shrl $16, %eax ;; EAX = 0 again!
- pinsrw $7, %eax, %xmm0
- movaps %xmm0, (%ecx)
- ret
+ movl 8(%esp), %eax
+ movaps (%eax), %xmm0
+ pxor %xmm1, %xmm1
+ movaps %xmm0, %xmm2
+ shufps $50, %xmm1, %xmm2
+ shufps $132, %xmm2, %xmm0
+ movaps %xmm0, (%eax)
+ ret
-It would be better to generate:
+Would it be better to generate:
_test:
movl 8(%esp), %ecx
movaps %xmm0, (%ecx)
ret
-or use pxor (to make a zero vector) and shuffle (to insert it).
+?
//===---------------------------------------------------------------------===//
//===---------------------------------------------------------------------===//
-We should compile this:
+We should materialize vector constants like "all ones" and "signbit" with
+code like:
+
+ cmpeqps xmm1, xmm1 ; xmm1 = all-ones
+
+and:
+ cmpeqps xmm1, xmm1 ; xmm1 = all-ones
+ psrlq xmm1, 31 ; xmm1 = all 100000000000...
+
+instead of using a load from the constant pool. The later is important for
+ABS/NEG/copysign etc.
+
+//===---------------------------------------------------------------------===//
+
+These functions:
+
#include <xmmintrin.h>
-typedef union {
- int i[4];
- float f[4];
- __m128 v;
-} vector4_t;
-void swizzle (const void *a, vector4_t * b, vector4_t * c) {
- b->v = _mm_loadl_pi (b->v, (__m64 *) a);
- c->v = _mm_loadl_pi (c->v, ((__m64 *) a) + 1);
+__m128i a;
+void x(unsigned short n) {
+ a = _mm_slli_epi32 (a, n);
+}
+void y(unsigned n) {
+ a = _mm_slli_epi32 (a, n);
}
-to:
-
-_swizzle:
- movl 4(%esp), %eax
- movl 8(%esp), %edx
- movl 12(%esp), %ecx
- movlps (%eax), %xmm0
- movlps %xmm0, (%edx)
- movlps 8(%eax), %xmm0
- movlps %xmm0, (%ecx)
+compile to ( -O3 -static -fomit-frame-pointer):
+_x:
+ movzwl 4(%esp), %eax
+ movd %eax, %xmm0
+ movaps _a, %xmm1
+ pslld %xmm0, %xmm1
+ movaps %xmm1, _a
ret
-
-not:
-
-swizzle:
- movl 8(%esp), %eax
- movaps (%eax), %xmm0
- movl 4(%esp), %ecx
- movlps (%ecx), %xmm0
- movaps %xmm0, (%eax)
- movl 12(%esp), %eax
- movaps (%eax), %xmm0
- movlps 8(%ecx), %xmm0
- movaps %xmm0, (%eax)
+_y:
+ movd 4(%esp), %xmm0
+ movaps _a, %xmm1
+ pslld %xmm0, %xmm1
+ movaps %xmm1, _a
ret
-//===---------------------------------------------------------------------===//
+"y" looks good, but "x" does silly movzwl stuff around into a GPR. It seems
+like movd would be sufficient in both cases as the value is already zero
+extended in the 32-bit stack slot IIRC. For signed short, it should also be
+save, as a really-signed value would be undefined for pslld.
-This code:
-#include <emmintrin.h>
-__m128i test(long long i) { return _mm_cvtsi64x_si128(i); }
+//===---------------------------------------------------------------------===//
-Should turn into a single 'movq %rdi, %xmm0' instruction. Instead, we
-get this (on x86-64):
+#include <math.h>
+int t1(double d) { return signbit(d); }
-_test:
- movd %rdi, %xmm1
- xorps %xmm0, %xmm0
- movsd %xmm1, %xmm0
+This currently compiles to:
+ subl $12, %esp
+ movsd 16(%esp), %xmm0
+ movsd %xmm0, (%esp)
+ movl 4(%esp), %eax
+ shrl $31, %eax
+ addl $12, %esp
ret
-The LLVM IR is:
+We should use movmskp{s|d} instead.
-target triple = "x86_64-apple-darwin8"
-define <2 x i64> @test(i64 %i) {
+//===---------------------------------------------------------------------===//
+
+CodeGen/X86/vec_align.ll tests whether we can turn 4 scalar loads into a single
+(aligned) vector load. This functionality has a couple of problems.
+
+1. The code to infer alignment from loads of globals is in the X86 backend,
+ not the dag combiner. This is because dagcombine2 needs to be able to see
+ through the X86ISD::Wrapper node, which DAGCombine can't really do.
+2. The code for turning 4 x load into a single vector load is target
+ independent and should be moved to the dag combiner.
+3. The code for turning 4 x load into a vector load can only handle a direct
+ load from a global or a direct load from the stack. It should be generalized
+ to handle any load from P, P+4, P+8, P+12, where P can be anything.
+4. The alignment inference code cannot handle loads from globals in non-static
+ mode because it doesn't look through the extra dyld stub load. If you try
+ vec_align.ll without -relocation-model=static, you'll see what I mean.
+
+//===---------------------------------------------------------------------===//
+
+We should lower store(fneg(load p), q) into an integer load+xor+store, which
+eliminates a constant pool load. For example, consider:
+
+define i64 @ccosf(float %z.0, float %z.1) nounwind readonly {
entry:
- %tmp10 = insertelement <2 x i64> undef, i64 %i, i32 0
- %tmp11 = insertelement <2 x i64> %tmp10, i64 0, i32 1
- ret <2 x i64> %tmp11
+ %tmp6 = fsub float -0.000000e+00, %z.1 ; <float> [#uses=1]
+ %tmp20 = tail call i64 @ccoshf( float %tmp6, float %z.0 ) nounwind readonly
+ ret i64 %tmp20
}
+declare i64 @ccoshf(float %z.0, float %z.1) nounwind readonly
+
+This currently compiles to:
+
+LCPI1_0: # <4 x float>
+ .long 2147483648 # float -0
+ .long 2147483648 # float -0
+ .long 2147483648 # float -0
+ .long 2147483648 # float -0
+_ccosf:
+ subl $12, %esp
+ movss 16(%esp), %xmm0
+ movss %xmm0, 4(%esp)
+ movss 20(%esp), %xmm0
+ xorps LCPI1_0, %xmm0
+ movss %xmm0, (%esp)
+ call L_ccoshf$stub
+ addl $12, %esp
+ ret
+
+Note the load into xmm0, then xor (to negate), then store. In PIC mode,
+this code computes the pic base and does two loads to do the constant pool
+load, so the improvement is much bigger.
+
+The tricky part about this xform is that the argument load/store isn't exposed
+until post-legalize, and at that point, the fneg has been custom expanded into
+an X86 fxor. This means that we need to handle this case in the x86 backend
+instead of in target independent code.
+
+//===---------------------------------------------------------------------===//
+
+Non-SSE4 insert into 16 x i8 is atrociously bad.
+
+//===---------------------------------------------------------------------===//
+
+<2 x i64> extract is substantially worse than <2 x f64>, even if the destination
+is memory.
//===---------------------------------------------------------------------===//
-These functions should produce the same code:
+INSERTPS can match any insert (extract, imm1), imm2 for 4 x float, and insert
+any number of 0.0 simultaneously. Currently we only use it for simple
+insertions.
+
+See comments in LowerINSERT_VECTOR_ELT_SSE4.
+
+//===---------------------------------------------------------------------===//
+On a random note, SSE2 should declare insert/extract of 2 x f64 as legal, not
+Custom. All combinations of insert/extract reg-reg, reg-mem, and mem-reg are
+legal, it'll just take a few extra patterns written in the .td file.
+
+Note: this is not a code quality issue; the custom lowered code happens to be
+right, but we shouldn't have to custom lower anything. This is probably related
+to <2 x i64> ops being so bad.
+
+//===---------------------------------------------------------------------===//
+
+LLVM currently generates stack realignment code, when it is not necessary
+needed. The problem is that we need to know about stack alignment too early,
+before RA runs.
+
+At that point we don't know, whether there will be vector spill, or not.
+Stack realignment logic is overly conservative here, but otherwise we can
+produce unaligned loads/stores.
+
+Fixing this will require some huge RA changes.
+
+Testcase:
#include <emmintrin.h>
-typedef long long __m128i __attribute__ ((__vector_size__ (16)));
+typedef short vSInt16 __attribute__ ((__vector_size__ (16)));
+
+static const vSInt16 a = {- 22725, - 12873, - 22725, - 12873, - 22725, - 12873,
+- 22725, - 12873};;
+
+vSInt16 madd(vSInt16 b)
+{
+ return _mm_madd_epi16(a, b);
+}
+
+Generated code (x86-32, linux):
+madd:
+ pushl %ebp
+ movl %esp, %ebp
+ andl $-16, %esp
+ movaps .LCPI1_0, %xmm1
+ pmaddwd %xmm1, %xmm0
+ movl %ebp, %esp
+ popl %ebp
+ ret
+
+//===---------------------------------------------------------------------===//
+
+Consider:
+#include <emmintrin.h>
+__m128 foo2 (float x) {
+ return _mm_set_ps (0, 0, x, 0);
+}
+
+In x86-32 mode, we generate this spiffy code:
+
+_foo2:
+ movss 4(%esp), %xmm0
+ pshufd $81, %xmm0, %xmm0
+ ret
+
+in x86-64 mode, we generate this code, which could be better:
+
+_foo2:
+ xorps %xmm1, %xmm1
+ movss %xmm0, %xmm1
+ pshufd $81, %xmm1, %xmm0
+ ret
+
+In sse4 mode, we could use insertps to make both better.
+
+Here's another testcase that could use insertps [mem]:
+
+#include <xmmintrin.h>
+extern float x2, x3;
+__m128 foo1 (float x1, float x4) {
+ return _mm_set_ps (x2, x1, x3, x4);
+}
+
+gcc mainline compiles it to:
+
+foo1:
+ insertps $0x10, x2(%rip), %xmm0
+ insertps $0x10, x3(%rip), %xmm1
+ movaps %xmm1, %xmm2
+ movlhps %xmm0, %xmm2
+ movaps %xmm2, %xmm0
+ ret
+
+//===---------------------------------------------------------------------===//
+
+We compile vector multiply-by-constant into poor code:
-int foo(__m128i* val) {
- return __builtin_ia32_vec_ext_v4si(*val, 1);
+define <4 x i32> @f(<4 x i32> %i) nounwind {
+ %A = mul <4 x i32> %i, < i32 10, i32 10, i32 10, i32 10 >
+ ret <4 x i32> %A
}
-int bar(__m128i* val) {
- union vs {
- __m128i *_v;
- int* _s;
- } v = {val};
- return v._s[1];
+
+On targets without SSE4.1, this compiles into:
+
+LCPI1_0: ## <4 x i32>
+ .long 10
+ .long 10
+ .long 10
+ .long 10
+ .text
+ .align 4,0x90
+ .globl _f
+_f:
+ pshufd $3, %xmm0, %xmm1
+ movd %xmm1, %eax
+ imull LCPI1_0+12, %eax
+ movd %eax, %xmm1
+ pshufd $1, %xmm0, %xmm2
+ movd %xmm2, %eax
+ imull LCPI1_0+4, %eax
+ movd %eax, %xmm2
+ punpckldq %xmm1, %xmm2
+ movd %xmm0, %eax
+ imull LCPI1_0, %eax
+ movd %eax, %xmm1
+ movhlps %xmm0, %xmm0
+ movd %xmm0, %eax
+ imull LCPI1_0+8, %eax
+ movd %eax, %xmm0
+ punpckldq %xmm0, %xmm1
+ movaps %xmm1, %xmm0
+ punpckldq %xmm2, %xmm0
+ ret
+
+It would be better to synthesize integer vector multiplication by constants
+using shifts and adds, pslld and paddd here. And even on targets with SSE4.1,
+simple cases such as multiplication by powers of two would be better as
+vector shifts than as multiplications.
+
+//===---------------------------------------------------------------------===//
+
+We compile this:
+
+__m128i
+foo2 (char x)
+{
+ return _mm_set_epi8 (1, 0, 0, 0, 0, 0, 0, 0, 0, x, 0, 1, 0, 0, 0, 0);
}
-We currently produce (with -m64):
+into:
+ movl $1, %eax
+ xorps %xmm0, %xmm0
+ pinsrw $2, %eax, %xmm0
+ movzbl 4(%esp), %eax
+ pinsrw $3, %eax, %xmm0
+ movl $256, %eax
+ pinsrw $7, %eax, %xmm0
+ ret
+
+
+gcc-4.2:
+ subl $12, %esp
+ movzbl 16(%esp), %eax
+ movdqa LC0, %xmm0
+ pinsrw $3, %eax, %xmm0
+ addl $12, %esp
+ ret
+ .const
+ .align 4
+LC0:
+ .word 0
+ .word 0
+ .word 1
+ .word 0
+ .word 0
+ .word 0
+ .word 0
+ .word 256
+
+With SSE4, it should be
+ movdqa .LC0(%rip), %xmm0
+ pinsrb $6, %edi, %xmm0
+
+//===---------------------------------------------------------------------===//
+
+We should transform a shuffle of two vectors of constants into a single vector
+of constants. Also, insertelement of a constant into a vector of constants
+should also result in a vector of constants. e.g. 2008-06-25-VecISelBug.ll.
+
+We compiled it to something horrible:
+
+ .align 4
+LCPI1_1: ## float
+ .long 1065353216 ## float 1
+ .const
+
+ .align 4
+LCPI1_0: ## <4 x float>
+ .space 4
+ .long 1065353216 ## float 1
+ .space 4
+ .long 1065353216 ## float 1
+ .text
+ .align 4,0x90
+ .globl _t
+_t:
+ xorps %xmm0, %xmm0
+ movhps LCPI1_0, %xmm0
+ movss LCPI1_1, %xmm1
+ movaps %xmm0, %xmm2
+ shufps $2, %xmm1, %xmm2
+ shufps $132, %xmm2, %xmm0
+ movaps %xmm0, 0
+
+//===---------------------------------------------------------------------===//
+rdar://5907648
+
+This function:
+
+float foo(unsigned char x) {
+ return x;
+}
+
+compiles to (x86-32):
+
+define float @foo(i8 zeroext %x) nounwind {
+ %tmp12 = uitofp i8 %x to float ; <float> [#uses=1]
+ ret float %tmp12
+}
+
+compiles to:
_foo:
- pshufd $1, (%rdi), %xmm0
- movd %xmm0, %eax
- ret
-_bar:
- movl 4(%rdi), %eax
- ret
+ subl $4, %esp
+ movzbl 8(%esp), %eax
+ cvtsi2ss %eax, %xmm0
+ movss %xmm0, (%esp)
+ flds (%esp)
+ addl $4, %esp
+ ret
+
+We should be able to use:
+ cvtsi2ss 8($esp), %xmm0
+since we know the stack slot is already zext'd.
//===---------------------------------------------------------------------===//
-We should materialize vector constants like "all ones" and "signbit" with
-code like:
+Consider using movlps instead of movsd to implement (scalar_to_vector (loadf64))
+when code size is critical. movlps is slower than movsd on core2 but it's one
+byte shorter.
- cmpeqps xmm1, xmm1 ; xmm1 = all-ones
+//===---------------------------------------------------------------------===//
-and:
- cmpeqps xmm1, xmm1 ; xmm1 = all-ones
- psrlq xmm1, 31 ; xmm1 = all 100000000000...
+We should use a dynamic programming based approach to tell when using FPStack
+operations is cheaper than SSE. SciMark montecarlo contains code like this
+for example:
-instead of using a load from the constant pool. The later is important for
-ABS/NEG/copysign etc.
+double MonteCarlo_num_flops(int Num_samples) {
+ return ((double) Num_samples)* 4.0;
+}
+
+In fpstack mode, this compiles into:
+
+LCPI1_0:
+ .long 1082130432 ## float 4.000000e+00
+_MonteCarlo_num_flops:
+ subl $4, %esp
+ movl 8(%esp), %eax
+ movl %eax, (%esp)
+ fildl (%esp)
+ fmuls LCPI1_0
+ addl $4, %esp
+ ret
+
+in SSE mode, it compiles into significantly slower code:
+
+_MonteCarlo_num_flops:
+ subl $12, %esp
+ cvtsi2sd 16(%esp), %xmm0
+ mulsd LCPI1_0, %xmm0
+ movsd %xmm0, (%esp)
+ fldl (%esp)
+ addl $12, %esp
+ ret
+
+There are also other cases in scimark where using fpstack is better, it is
+cheaper to do fld1 than load from a constant pool for example, so
+"load, add 1.0, store" is better done in the fp stack, etc.
+
+//===---------------------------------------------------------------------===//
+
+These should compile into the same code (PR6214): Perhaps instcombine should
+canonicalize the former into the later?
+
+define float @foo(float %x) nounwind {
+ %t = bitcast float %x to i32
+ %s = and i32 %t, 2147483647
+ %d = bitcast i32 %s to float
+ ret float %d
+}
+
+declare float @fabsf(float %n)
+define float @bar(float %x) nounwind {
+ %d = call float @fabsf(float %x)
+ ret float %d
+}
//===---------------------------------------------------------------------===//
-"converting 64-bit constant pool entry to 32-bit not necessarily beneficial"
-http://llvm.org/PR1264
+This IR (from PR6194):
+
+target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64-S128"
+target triple = "x86_64-apple-darwin10.0.0"
-For this test case:
+%0 = type { double, double }
+%struct.float3 = type { float, float, float }
-define double @foo(double %x) {
- %y = mul double %x, 5.000000e-01
- ret double %y
+define void @test(%0, %struct.float3* nocapture %res) nounwind noinline ssp {
+entry:
+ %tmp18 = extractvalue %0 %0, 0 ; <double> [#uses=1]
+ %tmp19 = bitcast double %tmp18 to i64 ; <i64> [#uses=1]
+ %tmp20 = zext i64 %tmp19 to i128 ; <i128> [#uses=1]
+ %tmp10 = lshr i128 %tmp20, 32 ; <i128> [#uses=1]
+ %tmp11 = trunc i128 %tmp10 to i32 ; <i32> [#uses=1]
+ %tmp12 = bitcast i32 %tmp11 to float ; <float> [#uses=1]
+ %tmp5 = getelementptr inbounds %struct.float3* %res, i64 0, i32 1 ; <float*> [#uses=1]
+ store float %tmp12, float* %tmp5
+ ret void
}
-llc -march=x86-64 currently produces a 32-bit constant pool entry and this code:
+Compiles to:
+
+_test: ## @test
+ movd %xmm0, %rax
+ shrq $32, %rax
+ movl %eax, 4(%rdi)
+ ret
+
+This would be better kept in the SSE unit by treating XMM0 as a 4xfloat and
+doing a shuffle from v[1] to v[0] then a float store.
+
+//===---------------------------------------------------------------------===//
- cvtss2sd .LCPI1_0(%rip), %xmm1
- mulsd %xmm1, %xmm0
+[UNSAFE FP]
-instead of just using a 64-bit constant pool entry with this:
+void foo(double, double, double);
+void norm(double x, double y, double z) {
+ double scale = __builtin_sqrt(x*x + y*y + z*z);
+ foo(x/scale, y/scale, z/scale);
+}
- mulsd .LCPI1_0(%rip), %xmm0
+We currently generate an sqrtsd and 3 divsd instructions. This is bad, fp div is
+slow and not pipelined. In -ffast-math mode we could compute "1.0/scale" first
+and emit 3 mulsd in place of the divs. This can be done as a target-independent
+transform.
-This is due to the code in ExpandConstantFP in LegalizeDAG.cpp. It notices that
-x86-64 indeed has an instruction to load a 32-bit float from memory and convert
-it into a 64-bit float in a register, however it doesn't notice that this isn't
-beneficial because it prevents the load from being folded into the multiply.
+If we're dealing with floats instead of doubles we could even replace the sqrtss
+and inversion with an rsqrtss instruction, which computes 1/sqrt faster at the
+cost of reduced accuracy.
//===---------------------------------------------------------------------===//
-In this loop:
+This function should be matched to haddpd when the appropriate CPU is enabled:
-LBB4_6: # bb47.preheader
- shlw $2, %si
- decw %si
- movzwl %si, %eax
- incl %eax
- xorl %ecx, %ecx
-LBB4_7: # bb47
- xorps %xmm0, %xmm0 # (1)
- movaps %xmm0, (%edx)
- movaps %xmm0, 16(%edx)
- addl $32, %edx
- incl %ecx
- cmpl %eax, %ecx
- jne LBB4_7 # bb47
-LBB4_8: # bb42.return_crit_edge
- xorw %si, %si
- jmp LBB4_12 # return
+#include <x86intrin.h>
+double f (__m128d p) {
+ return p[0] + p[1];
+}
-The instruction at (1) can be moved out of the main body of the loop.
+similarly, v[0]-v[1] should match to hsubpd, and {v[0]-v[1], w[0]-w[1]} should
+turn into hsubpd also.
//===---------------------------------------------------------------------===//
+define <2 x i32> @foo(<2 x double> %in) {
+ %x = fptosi <2 x double> %in to <2 x i32>
+ ret <2 x i32> %x
+}
+
+Should compile into cvttpd2dq instead of being scalarized into 2 cvttsd2si.
+
+//===---------------------------------------------------------------------===//