Zend/asm: adjust XCOFF asm files for AIX assembler

AIX assembler is a bit more strict than GNU assembler. Thus, adjust
the XCOFF asm files to be able to accept both assembler.

Author: Clément Chigot

Zend/asm/jump_ppc32_sysv_xcoff_gas.S

@@ -5,199 +5,212 @@
           http://www.boost.org/LICENSE_1_0.txt)
 */
 
-/******************************************************
+/*******************************************************
  *                                                     *
  *  -------------------------------------------------  *
- *  |  0  |  1  |  2  |  3  |  4  |  5  |  6  |  7  |  *
- *  -------------------------------------------------  *
  *  |  0  |  4  |  8  |  12 |  16 |  20 |  24 |  28 |  *
  *  -------------------------------------------------  *
- *  |    F14    |    F15    |    F16    |    F17    |  *
- *  -------------------------------------------------  *
+ *  |bchai|  CR | LR  |compl| link| TOC | R14 | R15 |  *
  *  -------------------------------------------------  *
- *  |  8  |  9  |  10 |  11 |  12 |  13 |  14 |  15 |  *
  *  -------------------------------------------------  *
  *  |  32 |  36 |  40 |  44 |  48 |  52 |  56 |  60 |  *
  *  -------------------------------------------------  *
- *  |    F18    |    F19    |    F20    |    F21    |  *
- *  -------------------------------------------------  *
+ *  | R16 | R17 | R18 | R19 | R20 | R21 | R22 | R23 |  *
  *  -------------------------------------------------  *
- *  |  16 |  17 |  18 |  19 |  20 |  21 |  22 |  23 |  *
  *  -------------------------------------------------  *
  *  |  64 |  68 |  72 |  76 |  80 |  84 |  88 |  92 |  *
  *  -------------------------------------------------  *
- *  |    F22    |    F23    |    F24    |    F25    |  *
- *  -------------------------------------------------  *
+ *  | R24 | R25 | R26 | R27 | R28 | R29 | R30 | R31 |  *
  *  -------------------------------------------------  *
- *  |  24 |  25 |  26 |  27 |  28 |  29 |  30 |  31 |  *
  *  -------------------------------------------------  *
  *  |  96 | 100 | 104 | 108 | 112 | 116 | 120 | 124 |  *
  *  -------------------------------------------------  *
- *  |    F26    |    F27    |    F28    |    F29    |  *
- *  -------------------------------------------------  *
+ *  |    F14    |    F15    |    F16    |    F17    |  *
  *  -------------------------------------------------  *
- *  |  32 |  33 |  34 |  35 |  36 |  37 |  38 |  39 |  *
  *  -------------------------------------------------  *
  *  | 128 | 132 | 136 | 140 | 144 | 148 | 152 | 156 |  *
  *  -------------------------------------------------  *
- *  |    F30    |    F31    |   fpscr   | R13 | R14 |  *
- *  -------------------------------------------------  *
+ *  |    F18    |    F19    |    F20    |    F21    |  *
  *  -------------------------------------------------  *
- *  |  40 |  41 |  42 |  43 |  44 |  45 |  46 |  47 |  *
  *  -------------------------------------------------  *
  *  | 160 | 164 | 168 | 172 | 176 | 180 | 184 | 188 |  *
  *  -------------------------------------------------  *
- *  | R15 | R16 | R17 | R18 | R19 | R20 | R21 | R22 |  *
- *  -------------------------------------------------  *
+ *  |    F22    |    F23    |    F24    |    F25    |  *
  *  -------------------------------------------------  *
- *  |  48 |  49 |  50 |  51 |  52 |  53 |  54 |  55 |  *
  *  -------------------------------------------------  *
  *  | 192 | 196 | 200 | 204 | 208 | 212 | 216 | 220 |  *
  *  -------------------------------------------------  *
- *  | R23 | R24 | R25 | R26 | R27 | R28 | R29 | R30 |  *
- *  -------------------------------------------------  *
+ *  |    F26    |    F27    |    F28    |    F29    |  *
  *  -------------------------------------------------  *
- *  |  56 |  57 |  58 |  59 |  60 |  61 |  62 |  63 |  *
  *  -------------------------------------------------  *
  *  | 224 | 228 | 232 | 236 | 240 | 244 | 248 | 252 |  *
  *  -------------------------------------------------  *
- *  | R31 |hiddn|  CR |  LR |  PC |bchai|linkr| FCTX|  *
- *  -------------------------------------------------  *
+ *  |    F30    |    F31    |  PC |hiddn|  fpscr    |  *
  *  -------------------------------------------------  *
- *  |  64 |                                         |  *
  *  -------------------------------------------------  *
- *  | 256 |                                         |  *
+ *  | 256 | 260 | 264 | 268 | 272 | 276 | 280 | 284 |  *
  *  -------------------------------------------------  *
- *  | DATA|                                         |  * 
+ *  |bchai|savCR|savLR|compl| link|svTOC| FCTX| DATA|  *
  *  -------------------------------------------------  *
  *                                                     *
  *******************************************************/
-.globl .jump_fcontext
-.globl  jump_fcontext[DS]
-.align 2 
-.csect	jump_fcontext[DS]
+
+    .file "jump_ppc32_sysv_xcoff_gas.S"
+    .toc
+    .csect .text[PR], 5
+    .globl  jump_fcontext[DS]
+    .globl .jump_fcontext
+    .csect  jump_fcontext[DS]
 jump_fcontext:
-  .long	.jump_fcontext
+    .long .jump_fcontext[PR], TOC[tc0], 0
+    .csect .text[PR], 5
 .jump_fcontext:
     # reserve space on stack
-    subi  r1, r1, 244
-
-    stfd  f14, 0(r1)  # save F14
-    stfd  f15, 8(r1)  # save F15
-    stfd  f16, 16(r1)  # save F16
-    stfd  f17, 24(r1)  # save F17
-    stfd  f18, 32(r1)  # save F18
-    stfd  f19, 40(r1)  # save F19
-    stfd  f20, 48(r1)  # save F20
-    stfd  f21, 56(r1)  # save F21
-    stfd  f22, 64(r1)  # save F22
-    stfd  f23, 72(r1)  # save F23
-    stfd  f24, 80(r1)  # save F24
-    stfd  f25, 88(r1)  # save F25
-    stfd  f26, 96(r1)  # save F26
-    stfd  f27, 104(r1)  # save F27
-    stfd  f28, 112(r1)  # save F28
-    stfd  f29, 120(r1)  # save F29
-    stfd  f30, 128(r1)  # save F30
-    stfd  f31, 136(r1)  # save F31
-    mffs  f0  # load FPSCR
-    stfd  f0, 144(r1)  # save FPSCR
-
-    stw  r13, 152(r1)  # save R13
-    stw  r14, 156(r1)  # save R14
-    stw  r15, 160(r1)  # save R15
-    stw  r16, 164(r1)  # save R16
-    stw  r17, 168(r1)  # save R17
-    stw  r18, 172(r1)  # save R18
-    stw  r19, 176(r1)  # save R19
-    stw  r20, 180(r1)  # save R20
-    stw  r21, 184(r1)  # save R21
-    stw  r22, 188(r1)  # save R22
-    stw  r23, 192(r1)  # save R23
-    stw  r24, 196(r1)  # save R24
-    stw  r25, 200(r1)  # save R25
-    stw  r26, 204(r1)  # save R26
-    stw  r27, 208(r1)  # save R27
-    stw  r28, 212(r1)  # save R28
-    stw  r29, 216(r1)  # save R29
-    stw  r30, 220(r1)  # save R30
-    stw  r31, 224(r1)  # save R31
-    stw  r3,  228(r1)  # save hidden
+    subi  1, 1, 256
 
     # save CR
-    mfcr  r0
-    stw   r0, 232(r1)
+    mfcr  0
+    stw   0, 4(1)
     # save LR
-    mflr  r0
-    stw   r0, 236(r1)
+    mflr  0
+    stw   0, 8(1)
     # save LR as PC
-    stw   r0, 240(r1)
+    stw   0, 240(1)
+    # save TOC
+    stw   2, 20(1)
+
+    # Save registers R14 to R31.
+    stw   14, 24(1)
+    stw   15, 28(1)
+    stw   16, 32(1)
+    stw   17, 36(1)
+    stw   18, 40(1)
+    stw   19, 44(1)
+    stw   20, 48(1)
+    stw   21, 52(1)
+    stw   22, 56(1)
+    stw   23, 60(1)
+    stw   24, 64(1)
+    stw   25, 68(1)
+    stw   26, 72(1)
+    stw   27, 76(1)
+    stw   28, 80(1)
+    stw   29, 84(1)
+    stw   30, 88(1)
+    stw   31, 92(1)
+
+    # Save registers F14 to F31 in slots with 8-byte alignment.
+    # 4-byte alignment may stall the pipeline of some processors.
+    # Less than 4 may cause alignment traps.
+    stfd  14, 96(1)
+    stfd  15, 104(1)
+    stfd  16, 112(1)
+    stfd  17, 120(1)
+    stfd  18, 128(1)
+    stfd  19, 136(1)
+    stfd  20, 144(1)
+    stfd  21, 152(1)
+    stfd  22, 160(1)
+    stfd  23, 168(1)
+    stfd  24, 176(1)
+    stfd  25, 184(1)
+    stfd  26, 192(1)
+    stfd  27, 200(1)
+    stfd  28, 208(1)
+    stfd  29, 216(1)
+    stfd  30, 224(1)
+    stfd  31, 232(1)
+
+    # hidden pointer
+    stw   3, 244(1)
+
+    mffs  0  # load FPSCR
+    stfd  0, 248(1)  # save FPSCR
+
 
     # store RSP (pointing to context-data) in R6
-    mr  r6, r1
+    mr  6, 1
 
     # restore RSP (pointing to context-data) from R4
-    mr  r1, r4
-
-    lfd  f14, 0(r1)  # restore F14
-    lfd  f15, 8(r1)  # restore F15
-    lfd  f16, 16(r1)  # restore F16
-    lfd  f17, 24(r1)  # restore F17
-    lfd  f18, 32(r1)  # restore F18
-    lfd  f19, 40(r1)  # restore F19
-    lfd  f20, 48(r1)  # restore F20
-    lfd  f21, 56(r1)  # restore F21
-    lfd  f22, 64(r1)  # restore F22
-    lfd  f23, 72(r1)  # restore F23
-    lfd  f24, 80(r1)  # restore F24
-    lfd  f25, 88(r1)  # restore F25
-    lfd  f26, 96(r1)  # restore F26
-    lfd  f27, 104(r1)  # restore F27
-    lfd  f28, 112(r1)  # restore F28
-    lfd  f29, 120(r1)  # restore F29
-    lfd  f30, 128(r1)  # restore F30
-    lfd  f31, 136(r1)  # restore F31
-    lfd  f0,  144(r1)  # load FPSCR
-    mtfsf  0xff, f0  # restore FPSCR
-
-    lwz  r13, 152(r1)  # restore R13
-    lwz  r14, 156(r1)  # restore R14
-    lwz  r15, 160(r1)  # restore R15
-    lwz  r16, 164(r1)  # restore R16
-    lwz  r17, 168(r1)  # restore R17
-    lwz  r18, 172(r1)  # restore R18
-    lwz  r19, 176(r1)  # restore R19
-    lwz  r20, 180(r1)  # restore R20
-    lwz  r21, 184(r1)  # restore R21
-    lwz  r22, 188(r1)  # restore R22
-    lwz  r23, 192(r1)  # restore R23
-    lwz  r24, 196(r1)  # restore R24
-    lwz  r25, 200(r1)  # restore R25
-    lwz  r26, 204(r1)  # restore R26
-    lwz  r27, 208(r1)  # restore R27
-    lwz  r28, 212(r1)  # restore R28
-    lwz  r29, 216(r1)  # restore R29
-    lwz  r30, 220(r1)  # restore R30
-    lwz  r31, 224(r1)  # restore R31
-    lwz  r3,  228(r1)  # restore hidden
+    mr  1, 4
 
     # restore CR
-    lwz   r0, 232(r1)
-    mtcr  r0
+    lwz   0, 4(1)
+    mtcr  0
     # restore LR
-    lwz   r0, 236(r1)
-    mtlr  r0
+    lwz   0, 8(1)
+    mtlr  0
     # load PC
-    lwz   r0, 240(r1)
-    # restore CTR
-    mtctr r0
+    lwz   0, 240(1)
+    mtctr 0
+
+    # restore TOC
+    lwz   2, 20(1)
+
+    # restore R14 to R31
+    lwz  14, 24(1)
+    lwz  15, 28(1)
+    lwz  16, 32(1)
+    lwz  17, 36(1)
+    lwz  18, 40(1)
+    lwz  19, 44(1)
+    lwz  20, 48(1)
+    lwz  21, 52(1)
+    lwz  22, 56(1)
+    lwz  23, 60(1)
+    lwz  24, 64(1)
+    lwz  25, 68(1)
+    lwz  26, 72(1)
+    lwz  27, 76(1)
+    lwz  28, 80(1)
+    lwz  29, 84(1)
+    lwz  30, 88(1)
+    lwz  31, 92(1)
+
+    # restore F14 to F31
+    lfd  14, 96(1)
+    lfd  15, 104(1)
+    lfd  16, 112(1)
+    lfd  17, 120(1)
+    lfd  18, 128(1)
+    lfd  19, 136(1)
+    lfd  20, 144(1)
+    lfd  21, 152(1)
+    lfd  22, 160(1)
+    lfd  23, 168(1)
+    lfd  24, 176(1)
+    lfd  25, 184(1)
+    lfd  26, 192(1)
+    lfd  27, 200(1)
+    lfd  28, 208(1)
+    lfd  29, 216(1)
+    lfd  30, 224(1)
+    lfd  31, 232(1)
+
+    # hidden pointer
+    lwz   3, 244(1)
+
+    lfd  0,  248(1)  # load FPSCR
+    mtfsf  0xff, 0  # restore FPSCR
 
     # adjust stack
-    addi  r1, r1, 244
+    addi  1, 1, 256
+
+    # zero in r3 indicates first jump to context-function
+    cmpdi 3, 0
+    beq use_entry_arg
+
+    # return transfer_t
+    stw  6, 0(3)
+    stw  5, 4(3)
+
+    # jump to context
+    bctr
 
-    # return transfer_t 
-    stw  r6, 0(r3)
-    stw  r5, 4(r3)
+use_entry_arg:
+    # copy transfer_t into transfer_fn arg registers
+    mr  3, 6
+    mr  4, 5
 
     # jump to context
     bctr