README.txt revision a3c4454c5fb70ab94c7de4faf2600ed729fffda0
1TODO: 2* gpr0 allocation 3* implement do-loop -> bdnz transform 4* implement powerpc-64 for darwin 5* use stfiwx in float->int 6* be able to combine sequences like the following into 2 instructions: 7 lis r2, ha16(l2__ZTV4Cell) 8 la r2, lo16(l2__ZTV4Cell)(r2) 9 addi r2, r2, 8 10 11* Teach LLVM how to codegen this: 12unsigned short foo(float a) { return a; } 13as: 14_foo: 15 fctiwz f0,f1 16 stfd f0,-8(r1) 17 lhz r3,-2(r1) 18 blr 19not: 20_foo: 21 fctiwz f0, f1 22 stfd f0, -8(r1) 23 lwz r2, -4(r1) 24 rlwinm r3, r2, 0, 16, 31 25 blr 26 27 28* Support 'update' load/store instructions. These are cracked on the G5, but 29 are still a codesize win. 30 31* Add a custom legalizer for the GlobalAddress node, to move the funky darwin 32 stub stuff from the instruction selector to the legalizer (exposing low-level 33 operations to the dag for optzn. For example, we want to codegen this: 34 35 int A = 0; 36 void B() { A++; } 37 as: 38 lis r9,ha16(_A) 39 lwz r2,lo16(_A)(r9) 40 addi r2,r2,1 41 stw r2,lo16(_A)(r9) 42 not: 43 lis r2, ha16(_A) 44 lwz r2, lo16(_A)(r2) 45 addi r2, r2, 1 46 lis r3, ha16(_A) 47 stw r2, lo16(_A)(r3) 48 49* should hint to the branch select pass that it doesn't need to print the 50 second unconditional branch, so we don't end up with things like: 51 b .LBBl42__2E_expand_function_8_674 ; loopentry.24 52 b .LBBl42__2E_expand_function_8_42 ; NewDefault 53 b .LBBl42__2E_expand_function_8_42 ; NewDefault 54 55===-------------------------------------------------------------------------=== 56 57* Codegen this: 58 59 void test2(int X) { 60 if (X == 0x12345678) bar(); 61 } 62 63 as: 64 65 xoris r0,r3,0x1234 66 cmpwi cr0,r0,0x5678 67 beq cr0,L6 68 69 not: 70 71 lis r2, 4660 72 ori r2, r2, 22136 73 cmpw cr0, r3, r2 74 bne .LBB_test2_2 75 76===-------------------------------------------------------------------------=== 77 78Lump the constant pool for each function into ONE pic object, and reference 79pieces of it as offsets from the start. For functions like this (contrived 80to have lots of constants obviously): 81 82double X(double Y) { return (Y*1.23 + 4.512)*2.34 + 14.38; } 83 84We generate: 85 86_X: 87 lis r2, ha16(.CPI_X_0) 88 lfd f0, lo16(.CPI_X_0)(r2) 89 lis r2, ha16(.CPI_X_1) 90 lfd f2, lo16(.CPI_X_1)(r2) 91 fmadd f0, f1, f0, f2 92 lis r2, ha16(.CPI_X_2) 93 lfd f1, lo16(.CPI_X_2)(r2) 94 lis r2, ha16(.CPI_X_3) 95 lfd f2, lo16(.CPI_X_3)(r2) 96 fmadd f1, f0, f1, f2 97 blr 98 99It would be better to materialize .CPI_X into a register, then use immediates 100off of the register to avoid the lis's. This is even more important in PIC 101mode. 102 103===-------------------------------------------------------------------------=== 104