+ if(typeString[i] == 'l')
+ {
+ // mr dst+1, src+1
+ *code++ = 0x7c000378
+ | ((-dst_locs[i]+1) << 16)
+ | ((-src_locs[i]+1) << 11)
+ | ((-src_locs[i]+1) << 21);
+ }
+ // mr dst, src
+ *code++ = 0x7c000378
+ | ((-dst_locs[i]) << 16)
+ | ((-src_locs[i]) << 11)
+ | ((-src_locs[i]) << 21);
+ }
+ else
+ {
+ if(typeString[i] == 'l')
+ {
+ // stw src+1, dst_offset+4(r1)
+ *code++ = 0x90010000
+ | ((-src_locs[i]+1) << 21)
+ | (dst_locs[i] + 4);
+ }
+
+ // stw src, dst_offset(r1)
+ *code++ = 0x90010000
+ | ((-src_locs[i]) << 21)
+ | (dst_locs[i] + 8);
+ }
+ }
+ else
+ {
+ ASSERT(dst_locs[i] >= 0);
+ ASSERT(typeString[i] != 'f' && typeString[i] != 'd');
+
+ if(typeString[i] == 'l')
+ {
+ // lwz r0, src_offset(r1)
+ *code++ = 0x80010000
+ | (src_locs[i] + frameSize + 8 + 4);
+ // stw r0, dst_offset(r1)
+ *code++ = 0x90010000
+ | (dst_locs[i] + 8 + 4);
+ }
+ // lwz r0, src_offset(r1)
+ *code++ = 0x80010000
+ | (src_locs[i] + frameSize + 8);
+ // stw r0, dst_offset(r1)
+ *code++ = 0x90010000
+ | (dst_locs[i] + 8);
+ }
+ }
+
+ // * hptr will be the new first argument.
+ // lis r3, hi(hptr)
+ *code++ = OP_HI(0x3c60, hptr);
+ // ori r3,r3,lo(hptr)
+ *code++ = OP_LO(0x6063, hptr);
+
+ // * we need to return to a piece of code
+ // which will tear down the stack frame.
+ // lis r11,hi(obscure_ccall_ret_code)
+ *code++ = OP_HI(0x3d60, obscure_ccall_ret_code);
+ // ori r11,r11,lo(obscure_ccall_ret_code)
+ *code++ = OP_LO(0x616b, obscure_ccall_ret_code);
+ // mtlr r11
+ *code++ = 0x7d6803a6;
+
+ // * jump to wptr
+ // lis r11,hi(wptr)
+ *code++ = OP_HI(0x3d60, wptr);
+ // ori r11,r11,lo(wptr)
+ *code++ = OP_LO(0x616b, wptr);
+ // mtctr r11
+ *code++ = 0x7d6903a6;
+ // bctr
+ *code++ = 0x4e800420;
+
+ // Flush the Instruction cache:
+ {
+ unsigned *p = adjustor;
+ while(p < code)
+ {
+ __asm__ volatile ("dcbf 0,%0\n\tsync\n\ticbi 0,%0"
+ : : "r" (p));
+ p++;
+ }
+ __asm__ volatile ("sync\n\tisync");
+ }
+ }
+
+#elif defined(powerpc_HOST_ARCH) || defined(powerpc64_HOST_ARCH)
+
+#define OP_LO(op,lo) ((((unsigned)(op)) << 16) | (((unsigned)(lo)) & 0xFFFF))
+#define OP_HI(op,hi) ((((unsigned)(op)) << 16) | (((unsigned)(hi)) >> 16))
+ {
+ /* The following code applies to all PowerPC and PowerPC64 platforms
+ whose stack layout is based on the AIX ABI.
+
+ Besides (obviously) AIX, this includes
+ Mac OS 9 and BeOS/PPC (may they rest in peace),
+ which use the 32-bit AIX ABI
+ powerpc64-linux,
+ which uses the 64-bit AIX ABI
+ and Darwin (Mac OS X),
+ which uses the same stack layout as AIX,
+ but no function descriptors.
+
+ The actual stack-frame shuffling is implemented out-of-line
+ in the function adjustorCode, in AdjustorAsm.S.
+ Here, we set up an AdjustorStub structure, which
+ is a function descriptor (on platforms that have function
+ descriptors) or a short piece of stub code (on Darwin) to call
+ adjustorCode with a pointer to the AdjustorStub struct loaded
+ into register r2.
+
+ One nice thing about this is that there is _no_ code generated at
+ runtime on the platforms that have function descriptors.
+ */
+ AdjustorStub *adjustorStub;
+ int sz = 0, extra_sz, total_sz;
+
+ // from AdjustorAsm.s
+ // not declared as a function so that AIX-style
+ // fundescs can never get in the way.
+ extern void *adjustorCode;
+
+#ifdef FUNDESCS
+ adjustorStub = stgMallocBytes(sizeof(AdjustorStub), "createAdjustor");
+#else
+ adjustorStub = stgMallocBytesRWX(sizeof(AdjustorStub));
+#endif
+ adjustor = adjustorStub;
+
+ adjustorStub->code = (void*) &adjustorCode;
+
+#ifdef FUNDESCS
+ // function descriptors are a cool idea.
+ // We don't need to generate any code at runtime.
+ adjustorStub->toc = adjustorStub;
+#else
+
+ // no function descriptors :-(
+ // We need to do things "by hand".
+#if defined(powerpc_HOST_ARCH)
+ // lis r2, hi(adjustorStub)
+ adjustorStub->lis = OP_HI(0x3c40, adjustorStub);
+ // ori r2, r2, lo(adjustorStub)
+ adjustorStub->ori = OP_LO(0x6042, adjustorStub);
+ // lwz r0, code(r2)
+ adjustorStub->lwz = OP_LO(0x8002, (char*)(&adjustorStub->code)
+ - (char*)adjustorStub);
+ // mtctr r0
+ adjustorStub->mtctr = 0x7c0903a6;
+ // bctr
+ adjustorStub->bctr = 0x4e800420;
+#else
+ barf("adjustor creation not supported on this platform");
+#endif
+
+ // Flush the Instruction cache:
+ {
+ int n = sizeof(AdjustorStub)/sizeof(unsigned);
+ unsigned *p = (unsigned*)adjustor;
+ while(n--)
+ {
+ __asm__ volatile ("dcbf 0,%0\n\tsync\n\ticbi 0,%0"
+ : : "r" (p));
+ p++;
+ }
+ __asm__ volatile ("sync\n\tisync");
+ }
+#endif
+
+ // Calculate the size of the stack frame, in words.
+ sz = totalArgumentSize(typeString);
+
+ // The first eight words of the parameter area
+ // are just "backing store" for the parameters passed in
+ // the GPRs. extra_sz is the number of words beyond those first
+ // 8 words.
+ extra_sz = sz - 8;
+ if(extra_sz < 0)
+ extra_sz = 0;
+
+ // Calculate the total size of the stack frame.
+ total_sz = (6 /* linkage area */
+ + 8 /* minimum parameter area */
+ + 2 /* two extra arguments */
+ + extra_sz)*sizeof(StgWord);
+
+ // align to 16 bytes.
+ // AIX only requires 8 bytes, but who cares?
+ total_sz = (total_sz+15) & ~0xF;
+
+ // Fill in the information that adjustorCode in AdjustorAsm.S
+ // will use to create a new stack frame with the additional args.
+ adjustorStub->hptr = hptr;
+ adjustorStub->wptr = wptr;
+ adjustorStub->negative_framesize = -total_sz;
+ adjustorStub->extrawords_plus_one = extra_sz + 1;
+ }