base/asm-common.h, *.S: Add `INTFUNC' macro for internal subroutines.

This provides correct alignment, and scoping for Windows SEH
annotations.

diff --git a/base/asm-common.h b/base/asm-common.h
index 083643e7cef21fa291c897aa11c68da038b54929..fdd7fad14deefedbbffecc8c016c885acad06e58 100644 (file)
--- a/base/asm-common.h
+++ b/base/asm-common.h
@@ -62,15 +62,19 @@
 #endif
 #define DATA .data
 
-// Announcing an external function.
-#define FUNC(name)                                                     \
-       .globl  F(name);                                                \
+// Announcing an internal function.
+#define INTFUNC(name)                                                  \
        TYPE_FUNC(name);                                                \
        .macro ENDFUNC; _ENDFUNC(name); .endm;                          \
        FUNC_PREHOOK(name);                                             \
-F(name):                                                               \
+name:                                                                  \
        FUNC_POSTHOOK(name)
 
+// Announcing an external function.
+#define FUNC(name)                                                     \
+       .globl  F(name);                                                \
+INTFUNC(F(name))
+
 // Marking the end of a function.
 #define _ENDFUNC(name)                                                 \
        .purgem ENDFUNC;                                                \

diff --git a/math/mpx-mul4-x86-sse2.S b/math/mpx-mul4-x86-sse2.S
index 922de33d49649d8acda847056a17b4d92f8c43d4..8f69a559dfdd96b4b2d4dd2d2212cbad5be61e9e 100644 (file)
--- a/math/mpx-mul4-x86-sse2.S
+++ b/math/mpx-mul4-x86-sse2.S
@@ -310,8 +310,7 @@
 ///--------------------------------------------------------------------------
 /// Primitive multipliers and related utilities.
 
-       .p2align 4
-carryprop:
+INTFUNC(carryprop)
        // On entry, XMM4, XMM5, and XMM6 hold a 144-bit carry in an expanded
        // form.  Store the low 128 bits of the represented carry to [EDI] as
        // a packed 128-bit value, and leave the remaining 16 bits in the low
@@ -322,8 +321,9 @@ carryprop:
        endprop [edi + 12], xmm6, xmm4
        ret
 
-       .p2align 4
-dmul4:
+ENDFUNC
+
+INTFUNC(dmul4)
        // On entry, EDI points to the destination buffer; EAX and EBX point
        // to the packed operands U and X; ECX and EDX point to the expanded
        // operands V and Y; and XMM4, XMM5, XMM6 hold the incoming carry
@@ -351,8 +351,9 @@ dmul4:
 
        ret
 
-       .p2align 4
-dmla4:
+ENDFUNC
+
+INTFUNC(dmla4)
        // On entry, EDI points to the destination buffer, which also
        // contains an addend A to accumulate; EAX and EBX point to the
        // packed operands U and X; ECX and EDX point to the expanded
@@ -384,8 +385,9 @@ dmla4:
 
        ret
 
-       .p2align 4
-mul4zc:
+ENDFUNC
+
+INTFUNC(mul4zc)
        // On entry, EDI points to the destination buffer; EBX points to a
        // packed operand X; and EDX points to an expanded operand Y.
        //
@@ -407,8 +409,9 @@ mul4zc:
 
        ret
 
-       .p2align 4
-mul4:
+ENDFUNC
+
+INTFUNC(mul4)
        // On entry, EDI points to the destination buffer; EBX points to a
        // packed operand X; EDX points to an expanded operand Y; and XMM4,
        // XMM5, XMM6 hold the incoming carry registers c0, c1, and c2,
@@ -432,8 +435,9 @@ mul4:
 
        ret
 
-       .p2align 4
-mla4zc:
+ENDFUNC
+
+INTFUNC(mla4zc)
        // On entry, EDI points to the destination buffer, which also
        // contains an addend A to accumulate; EBX points to a packed operand
        // X; and EDX points to an expanded operand Y.
@@ -461,8 +465,9 @@ mla4zc:
 
        ret
 
-       .p2align 4
-mla4:
+ENDFUNC
+
+INTFUNC(mla4)
        // On entry, EDI points to the destination buffer, which also
        // contains an addend A to accumulate; EBX points to a packed operand
        // X; EDX points to an expanded operand Y; and XMM4, XMM5, XMM6 hold
@@ -489,8 +494,9 @@ mla4:
 
        ret
 
-       .p2align 4
-mmul4:
+ENDFUNC
+
+INTFUNC(mmul4)
        // On entry, EDI points to the destination buffer; EAX and EBX point
        // to the packed operands U and N; ECX and ESI point to the expanded
        // operands V and M; and EDX points to a place to store an expanded
@@ -510,8 +516,9 @@ mmul4:
        propout [edi +  0],      xmm4, xmm5
        jmp     5f
 
-       .p2align 4
-mmla4:
+ENDFUNC
+
+INTFUNC(mmla4)
        // On entry, EDI points to the destination buffer, which also
        // contains an addend A to accumulate; EAX and EBX point
        // to the packed operands U and N; ECX and ESI point to the expanded
@@ -595,8 +602,9 @@ mmla4:
        add     esp, 64
        ret
 
-       .p2align 4
-mont4:
+ENDFUNC
+
+INTFUNC(mont4)
        // On entry, EDI points to the destination buffer holding a packed
        // value A; EBX points to a packed operand N; ESI points to an
        // expanded operand M; and EDX points to a place to store an expanded
@@ -650,6 +658,8 @@ mont4:
        // And, with that, we're done.
        ret
 
+ENDFUNC
+
 ///--------------------------------------------------------------------------
 /// Bulk multipliers.
 

diff --git a/symm/rijndael-arm-crypto.S b/symm/rijndael-arm-crypto.S
index 5ca516e331ba2975502c5808d1eaeb444d2aa3b6..4d7312d4d85144b01a643a1a0b3676353f395484 100644 (file)
--- a/symm/rijndael-arm-crypto.S
+++ b/symm/rijndael-arm-crypto.S
@@ -30,8 +30,8 @@
 #include "config.h"
 #include "asm-common.h"
 
-       .globl  F(abort)
-       .globl  F(rijndael_rcon)
+       .extern F(abort)
+       .extern F(rijndael_rcon)
 
 ///--------------------------------------------------------------------------
 /// Main code.
@@ -211,9 +211,12 @@ FUNC(rijndael_setup_arm_crypto)
        // All done.
 9:     ldmfd   sp!, {r4-r9, pc}
 
-endswap_block:
+ENDFUNC
+
+INTFUNC(endswap_block)
        // End-swap R2 words starting at R1.  R1 is clobbered; R2 is not.
        // It's OK to work in 16-byte chunks.
+
        mov     r4, r2
 0:     vldmia  r1, {d0, d1}
        vrev32.8 q0, q0

diff --git a/symm/rijndael-x86ish-aesni.S b/symm/rijndael-x86ish-aesni.S
index 8090bca675a6332ab17100798f73f2e2cc5357e9..b0b880a44e7c7623021dad92f28d44c9a1ae6848 100644 (file)
--- a/symm/rijndael-x86ish-aesni.S
+++ b/symm/rijndael-x86ish-aesni.S
@@ -332,10 +332,15 @@ FUNC(rijndael_setup_x86ish_aesni)
 #endif
        ret
 
-       .align  16
-endswap_block:
+ENDFUNC
+
+INTFUNC(endswap_block)
        // End-swap NKW words starting at SI.  The end-swapping table is
        // already loaded into XMM5; and it's OK to work in 16-byte chunks.
+#if CPUFAM_AMD64 && ABI_WIN
+  .seh_endprologue
+#endif
+
        mov     ecx, NKW
 0:     movdqu  xmm1, [SI]
        pshufb  xmm1, xmm5
@@ -343,8 +348,11 @@ endswap_block:
        add     SI, 16
        sub     ecx, 4
        ja      0b
+
        ret
 
+ENDFUNC
+
 #undef CTX
 #undef BLKSZ
 #undef SI
@@ -356,8 +364,6 @@ endswap_block:
 #undef LRK
 #undef BLKOFF
 
-ENDFUNC
-
 ///--------------------------------------------------------------------------
 /// Encrypting and decrypting blocks.
 
@@ -509,17 +515,23 @@ ENDFUNC
 ///--------------------------------------------------------------------------
 /// Random utilities.
 
-       .align  16
+INTFUNC(bogus)
        // Abort the process because of a programming error.  Indirecting
        // through this point serves several purposes: (a) by CALLing, rather
        // than branching to, `abort', we can save the return address, which
        // might at least provide a hint as to what went wrong; (b) we don't
        // have conditional CALLs (and they'd be big anyway); and (c) we can
        // write a HLT here as a backstop against `abort' being mad.
-bogus: callext F(abort)
+#if CPUFAM_AMD64 && ABI_WIN
+  .seh_endprologue
+#endif
+
+       callext F(abort)
 0:     hlt
        jmp     0b
 
+ENDFUNC
+
 ///--------------------------------------------------------------------------
 /// Data tables.