[PATCH 2/5] Enable AMD64 SHA256 implementations for WIN64
Jussi Kivilinna
jussi.kivilinna at iki.fi
Sat May 2 15:11:48 CEST 2015
* cipher/sha256-avx-amd64.S: Enable when
HAVE_COMPATIBLE_GCC_WIN64_PLATFORM_AS defined.
(ELF): New macro to mask lines with ELF specific commands.
* cipher/sha256-avx2-bmi2-amd64.S: Ditto.
* cipher/sha256-ssse3-amd64.S: Ditto.
* cipher/sha256.c (USE_SSSE3, USE_AVX, USE_AVX2): Enable when
HAVE_COMPATIBLE_GCC_WIN64_PLATFORM_AS defined.
[USE_SSSE3 || USE_AVX || USE_AVX2] (ASM_FUNC_ABI)
(ASM_EXTRA_STACK): New.
(_gcry_sha256_transform_amd64_ssse3, _gcry_sha256_transform_amd64_avx)
(_gcry_sha256_transform_amd64_avx2): Add ASM_FUNC_ABI to prototypes.
(transform): Add ASM_EXTRA_STACK to stack burn value.
--
Signed-off-by: Jussi Kivilinna <jussi.kivilinna at iki.fi>
---
cipher/sha256-avx-amd64.S | 11 ++++++-
cipher/sha256-avx2-bmi2-amd64.S | 11 ++++++-
cipher/sha256-ssse3-amd64.S | 11 ++++++-
cipher/sha256.c | 60 +++++++++++++++++++++++++++++----------
4 files changed, 72 insertions(+), 21 deletions(-)
diff --git a/cipher/sha256-avx-amd64.S b/cipher/sha256-avx-amd64.S
index 3912db7..8bf26bd 100644
--- a/cipher/sha256-avx-amd64.S
+++ b/cipher/sha256-avx-amd64.S
@@ -54,7 +54,8 @@
#ifdef __x86_64
#include <config.h>
-#if defined(HAVE_COMPATIBLE_GCC_AMD64_PLATFORM_AS) && \
+#if (defined(HAVE_COMPATIBLE_GCC_AMD64_PLATFORM_AS) || \
+ defined(HAVE_COMPATIBLE_GCC_WIN64_PLATFORM_AS)) && \
defined(HAVE_INTEL_SYNTAX_PLATFORM_AS) && \
defined(HAVE_GCC_INLINE_ASM_AVX) && defined(USE_SHA256)
@@ -64,6 +65,12 @@
# define ADD_RIP
#endif
+#ifdef HAVE_COMPATIBLE_GCC_AMD64_PLATFORM_AS
+# define ELF(...) __VA_ARGS__
+#else
+# define ELF(...) /*_*/
+#endif
+
.intel_syntax noprefix
#define VMOVDQ vmovdqu /* assume buffers not aligned */
@@ -370,7 +377,7 @@ rotate_Xs
*/
.text
.globl _gcry_sha256_transform_amd64_avx
-.type _gcry_sha256_transform_amd64_avx, at function;
+ELF(.type _gcry_sha256_transform_amd64_avx, at function;)
.align 16
_gcry_sha256_transform_amd64_avx:
vzeroupper
diff --git a/cipher/sha256-avx2-bmi2-amd64.S b/cipher/sha256-avx2-bmi2-amd64.S
index 09df711..74b6063 100644
--- a/cipher/sha256-avx2-bmi2-amd64.S
+++ b/cipher/sha256-avx2-bmi2-amd64.S
@@ -54,7 +54,8 @@
#ifdef __x86_64
#include <config.h>
-#if defined(HAVE_COMPATIBLE_GCC_AMD64_PLATFORM_AS) && \
+#if (defined(HAVE_COMPATIBLE_GCC_AMD64_PLATFORM_AS) || \
+ defined(HAVE_COMPATIBLE_GCC_WIN64_PLATFORM_AS)) && \
defined(HAVE_INTEL_SYNTAX_PLATFORM_AS) && \
defined(HAVE_GCC_INLINE_ASM_AVX2) && defined(HAVE_GCC_INLINE_ASM_BMI2) && \
defined(USE_SHA256)
@@ -65,6 +66,12 @@
# define ADD_RIP
#endif
+#ifdef HAVE_COMPATIBLE_GCC_AMD64_PLATFORM_AS
+# define ELF(...) __VA_ARGS__
+#else
+# define ELF(...) /*_*/
+#endif
+
.intel_syntax noprefix
#define VMOVDQ vmovdqu /* ; assume buffers not aligned */
@@ -555,7 +562,7 @@ rotate_Xs
*/
.text
.globl _gcry_sha256_transform_amd64_avx2
-.type _gcry_sha256_transform_amd64_avx2, at function
+ELF(.type _gcry_sha256_transform_amd64_avx2, at function)
.align 32
_gcry_sha256_transform_amd64_avx2:
push rbx
diff --git a/cipher/sha256-ssse3-amd64.S b/cipher/sha256-ssse3-amd64.S
index 80b1cec..9ec87e4 100644
--- a/cipher/sha256-ssse3-amd64.S
+++ b/cipher/sha256-ssse3-amd64.S
@@ -55,7 +55,8 @@
#ifdef __x86_64
#include <config.h>
-#if defined(HAVE_COMPATIBLE_GCC_AMD64_PLATFORM_AS) && \
+#if (defined(HAVE_COMPATIBLE_GCC_AMD64_PLATFORM_AS) || \
+ defined(HAVE_COMPATIBLE_GCC_WIN64_PLATFORM_AS)) && \
defined(HAVE_INTEL_SYNTAX_PLATFORM_AS) && \
defined(HAVE_GCC_INLINE_ASM_SSSE3) && defined(USE_SHA256)
@@ -65,6 +66,12 @@
# define ADD_RIP
#endif
+#ifdef HAVE_COMPATIBLE_GCC_AMD64_PLATFORM_AS
+# define ELF(...) __VA_ARGS__
+#else
+# define ELF(...) /*_*/
+#endif
+
.intel_syntax noprefix
#define MOVDQ movdqu /* assume buffers not aligned */
@@ -376,7 +383,7 @@ rotate_Xs
*/
.text
.globl _gcry_sha256_transform_amd64_ssse3
-.type _gcry_sha256_transform_amd64_ssse3, at function;
+ELF(.type _gcry_sha256_transform_amd64_ssse3, at function;)
.align 16
_gcry_sha256_transform_amd64_ssse3:
push rbx
diff --git a/cipher/sha256.c b/cipher/sha256.c
index d3af172..59ffa43 100644
--- a/cipher/sha256.c
+++ b/cipher/sha256.c
@@ -49,25 +49,29 @@
/* USE_SSSE3 indicates whether to compile with Intel SSSE3 code. */
#undef USE_SSSE3
-#if defined(__x86_64__) && defined(HAVE_COMPATIBLE_GCC_AMD64_PLATFORM_AS) && \
- defined(HAVE_GCC_INLINE_ASM_SSSE3) && \
- defined(HAVE_INTEL_SYNTAX_PLATFORM_AS)
+#if defined(__x86_64__) && defined(HAVE_GCC_INLINE_ASM_SSSE3) && \
+ defined(HAVE_INTEL_SYNTAX_PLATFORM_AS) && \
+ (defined(HAVE_COMPATIBLE_GCC_AMD64_PLATFORM_AS) || \
+ defined(HAVE_COMPATIBLE_GCC_WIN64_PLATFORM_AS))
# define USE_SSSE3 1
#endif
/* USE_AVX indicates whether to compile with Intel AVX code. */
#undef USE_AVX
-#if defined(__x86_64__) && defined(HAVE_COMPATIBLE_GCC_AMD64_PLATFORM_AS) && \
- defined(HAVE_GCC_INLINE_ASM_AVX) && \
- defined(HAVE_INTEL_SYNTAX_PLATFORM_AS)
+#if defined(__x86_64__) && defined(HAVE_GCC_INLINE_ASM_AVX) && \
+ defined(HAVE_INTEL_SYNTAX_PLATFORM_AS) && \
+ (defined(HAVE_COMPATIBLE_GCC_AMD64_PLATFORM_AS) || \
+ defined(HAVE_COMPATIBLE_GCC_WIN64_PLATFORM_AS))
# define USE_AVX 1
#endif
/* USE_AVX2 indicates whether to compile with Intel AVX2/BMI2 code. */
#undef USE_AVX2
-#if defined(__x86_64__) && defined(HAVE_COMPATIBLE_GCC_AMD64_PLATFORM_AS) && \
- defined(HAVE_GCC_INLINE_ASM_AVX2) && defined(HAVE_GCC_INLINE_ASM_BMI2) && \
- defined(HAVE_INTEL_SYNTAX_PLATFORM_AS)
+#if defined(__x86_64__) && defined(HAVE_GCC_INLINE_ASM_AVX2) && \
+ defined(HAVE_GCC_INLINE_ASM_BMI2) && \
+ defined(HAVE_INTEL_SYNTAX_PLATFORM_AS) && \
+ (defined(HAVE_COMPATIBLE_GCC_AMD64_PLATFORM_AS) || \
+ defined(HAVE_COMPATIBLE_GCC_WIN64_PLATFORM_AS))
# define USE_AVX2 1
#endif
@@ -322,19 +326,37 @@ transform_blk (void *ctx, const unsigned char *data)
#undef R
+/* Assembly implementations use SystemV ABI, ABI conversion and additional
+ * stack to store XMM6-XMM15 needed on Win64. */
+#undef ASM_FUNC_ABI
+#undef ASM_EXTRA_STACK
+#if defined(USE_SSSE3) || defined(USE_AVX) || defined(USE_AVX2)
+# ifdef HAVE_COMPATIBLE_GCC_WIN64_PLATFORM_AS
+# define ASM_FUNC_ABI __attribute__((sysv_abi))
+# define ASM_EXTRA_STACK (10 * 16)
+# else
+# define ASM_FUNC_ABI
+# define ASM_EXTRA_STACK 0
+# endif
+#endif
+
+
#ifdef USE_SSSE3
unsigned int _gcry_sha256_transform_amd64_ssse3(const void *input_data,
- u32 state[8], size_t num_blks);
+ u32 state[8],
+ size_t num_blks) ASM_FUNC_ABI;
#endif
#ifdef USE_AVX
unsigned int _gcry_sha256_transform_amd64_avx(const void *input_data,
- u32 state[8], size_t num_blks);
+ u32 state[8],
+ size_t num_blks) ASM_FUNC_ABI;
#endif
#ifdef USE_AVX2
unsigned int _gcry_sha256_transform_amd64_avx2(const void *input_data,
- u32 state[8], size_t num_blks);
+ u32 state[8],
+ size_t num_blks) ASM_FUNC_ABI;
#endif
@@ -347,19 +369,19 @@ transform (void *ctx, const unsigned char *data, size_t nblks)
#ifdef USE_AVX2
if (hd->use_avx2)
return _gcry_sha256_transform_amd64_avx2 (data, &hd->h0, nblks)
- + 4 * sizeof(void*);
+ + 4 * sizeof(void*) + ASM_EXTRA_STACK;
#endif
#ifdef USE_AVX
if (hd->use_avx)
return _gcry_sha256_transform_amd64_avx (data, &hd->h0, nblks)
- + 4 * sizeof(void*);
+ + 4 * sizeof(void*) + ASM_EXTRA_STACK;
#endif
#ifdef USE_SSSE3
if (hd->use_ssse3)
return _gcry_sha256_transform_amd64_ssse3 (data, &hd->h0, nblks)
- + 4 * sizeof(void*);
+ + 4 * sizeof(void*) + ASM_EXTRA_STACK;
#endif
do
@@ -369,6 +391,14 @@ transform (void *ctx, const unsigned char *data, size_t nblks)
}
while (--nblks);
+#ifdef ASM_EXTRA_STACK
+ /* 'transform_blk' is typically inlined and XMM6-XMM15 are stored at
+ * the prologue of this function. Therefore need to add ASM_EXTRA_STACK to
+ * here too.
+ */
+ burn += ASM_EXTRA_STACK;
+#endif
+
return burn;
}
More information about the Gcrypt-devel
mailing list