From 13fbb8a541db75484af7a457b2c892e7e0b5cbca Mon Sep 17 00:00:00 2001
From: psychocrypt <psychocryptHPC@gmail.com>
Date: Mon, 17 Sep 2018 09:16:06 +0200
Subject: [PATCH] asm compiler compatibility
- add special asm version for win64 and linux
- add cmake path for MSVC and other systems
---
CMakeLists.txt | 40 ++--
.../cpu/crypto/asm/cryptonigh_v8_main_loop.S | 22 ++-
.../crypto/asm/cryptonigh_v8_main_loop.asm | 8 +-
...yptonigh_v8_main_loop_ivybridge_linux.inc} | 22 +--
...ryptonigh_v8_main_loop_ivybridge_win64.inc | 176 ++++++++++++++++++
...> cryptonigh_v8_main_loop_ryzen_linux.inc} | 22 +--
.../cryptonigh_v8_main_loop_ryzen_win64.inc | 174 +++++++++++++++++
7 files changed, 415 insertions(+), 49 deletions(-)
rename xmrstak/backend/cpu/crypto/asm/{cryptonigh_v8_main_loop_ivybridge.inc => cryptonigh_v8_main_loop_ivybridge_linux.inc} (91%)
create mode 100644 xmrstak/backend/cpu/crypto/asm/cryptonigh_v8_main_loop_ivybridge_win64.inc
rename xmrstak/backend/cpu/crypto/asm/{cryptonigh_v8_main_loop_ryzen.inc => cryptonigh_v8_main_loop_ryzen_linux.inc} (92%)
create mode 100644 xmrstak/backend/cpu/crypto/asm/cryptonigh_v8_main_loop_ryzen_win64.inc
diff --git a/CMakeLists.txt b/CMakeLists.txt
index cf43922..b51eb2a 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -445,6 +445,26 @@ if(CMAKE_LINK_STATIC)
endif()
endif()
+if(CMAKE_C_COMPILER_ID MATCHES "MSVC")
+ # asm optimized monero v8 code
+ enable_language(ASM_MASM)
+ set_property(SOURCE "xmrstak/backend/cpu/crypto/asm/cryptonigh_v8_main_loop.asm" PROPERTY ASM_MASM)
+ add_library(xmr-stak-asm
+ STATIC
+ "xmrstak/backend/cpu/crypto/asm/cryptonigh_v8_main_loop.asm"
+ )
+else()
+ # asm optimized monero v8 code
+ enable_language(ASM)
+ set_property(SOURCE "xmrstak/backend/cpu/crypto/asm/cryptonigh_v8_main_loop.S" PROPERTY C)
+ add_library(xmr-stak-asm
+ STATIC
+ "xmrstak/backend/cpu/crypto/asm/cryptonigh_v8_main_loop.S"
+ )
+endif()
+
+set_property(TARGET xmr-stak-asm PROPERTY LINKER_LANGUAGE C)
+
# compile C files
file(GLOB SRCFILES_C "xmrstak/backend/cpu/crypto/*.c")
@@ -456,17 +476,7 @@ set_property(TARGET xmr-stak-c PROPERTY C_STANDARD 99)
if(MICROHTTPD_ENABLE)
target_link_libraries(xmr-stak-c ${MHTD})
endif()
-target_link_libraries(xmr-stak-c ${LIBS})
-
-enable_language(ASM)
-set_property(SOURCE "xmrstak/backend/cpu/crypto/asm/cryptonigh_v8_main_loop.S" PROPERTY LANGUAGE C)
-# asm optimized monero v8 code
-add_library(xmr-stak-asm
- STATIC
- "xmrstak/backend/cpu/crypto/asm/cryptonigh_v8_main_loop.S"
-)
-set_property(TARGET xmr-stak-asm PROPERTY LINKER_LANGUAGE C)
-
+target_link_libraries(xmr-stak-c ${LIBS} xmr-stak-asm)
# compile generic backend files
file(GLOB BACKEND_CPP
@@ -509,7 +519,7 @@ if(CUDA_FOUND)
)
endif()
target_link_libraries(xmrstak_cuda_backend ${CUDA_LIBRARIES})
- target_link_libraries(xmrstak_cuda_backend xmr-stak-backend)
+ target_link_libraries(xmrstak_cuda_backend xmr-stak-backend xmr-stak-asm)
endif()
# compile AMD backend
@@ -522,7 +532,7 @@ if(OpenCL_FOUND)
${OPENCLSRCFILES}
)
target_link_libraries(xmrstak_opencl_backend ${OpenCL_LIBRARY} )
- target_link_libraries(xmrstak_opencl_backend xmr-stak-backend)
+ target_link_libraries(xmrstak_opencl_backend xmr-stak-backend xmr-stak-asm)
endif()
# compile final binary
@@ -538,7 +548,7 @@ endif()
set(EXECUTABLE_OUTPUT_PATH "bin" CACHE STRING "Path to place executables relative to ${CMAKE_INSTALL_PREFIX}")
set(LIBRARY_OUTPUT_PATH "bin" CACHE STRING "Path to place libraries relative to ${CMAKE_INSTALL_PREFIX}")
-target_link_libraries(xmr-stak ${LIBS} xmr-stak-c xmr-stak-backend)
+target_link_libraries(xmr-stak ${LIBS} xmr-stak-c xmr-stak-backend xmr-stak-asm)
################################################################################
# Install
@@ -569,4 +579,4 @@ if( NOT CMAKE_INSTALL_PREFIX STREQUAL PROJECT_BINARY_DIR )
else()
# this rule is used if the install prefix is the build directory
install(CODE "MESSAGE(\"xmr-stak installed to folder 'bin'\")")
-endif()
+endif()
\ No newline at end of file
diff --git a/xmrstak/backend/cpu/crypto/asm/cryptonigh_v8_main_loop.S b/xmrstak/backend/cpu/crypto/asm/cryptonigh_v8_main_loop.S
index cd747f7..736dac7 100644
--- a/xmrstak/backend/cpu/crypto/asm/cryptonigh_v8_main_loop.S
+++ b/xmrstak/backend/cpu/crypto/asm/cryptonigh_v8_main_loop.S
@@ -1,21 +1,27 @@
#define ALIGN .align
.intel_syntax noprefix
+#ifdef __APPLE__
+# define FN_PREFIX(fn) _ ## fn
+.text
+#else
+# define FN_PREFIX(fn) fn
.section .text
-.global cryptonigh_v8_mainloop_ivybridge_asm
-.global cryptonigh_v8_mainloop_ryzen_asm
+#endif
+.global FN_PREFIX(cryptonigh_v8_mainloop_ivybridge_asm)
+.global FN_PREFIX(cryptonigh_v8_mainloop_ryzen_asm)
-ALIGN 64
-cryptonigh_v8_mainloop_ivybridge_asm:
+ALIGN 8
+FN_PREFIX(cryptonigh_v8_mainloop_ivybridge_asm):
sub rsp, 48
mov rcx, rdi
- #include "cryptonigh_v8_main_loop_ivybridge.inc"
+ #include "cryptonigh_v8_main_loop_ivybridge_linux.inc"
add rsp, 48
ret 0
-ALIGN 64
-cryptonigh_v8_mainloop_ryzen_asm:
+ALIGN 8
+FN_PREFIX(cryptonigh_v8_mainloop_ryzen_asm):
sub rsp, 48
mov rcx, rdi
- #include "cryptonigh_v8_main_loop_ryzen.inc"
+ #include "cryptonigh_v8_main_loop_ryzen_linux.inc"
add rsp, 48
ret 0
diff --git a/xmrstak/backend/cpu/crypto/asm/cryptonigh_v8_main_loop.asm b/xmrstak/backend/cpu/crypto/asm/cryptonigh_v8_main_loop.asm
index 2101a59..7f2d6a5 100644
--- a/xmrstak/backend/cpu/crypto/asm/cryptonigh_v8_main_loop.asm
+++ b/xmrstak/backend/cpu/crypto/asm/cryptonigh_v8_main_loop.asm
@@ -2,15 +2,15 @@ _TEXT_CNV8_MAINLOOP SEGMENT PAGE READ EXECUTE
PUBLIC cryptonigh_v8_mainloop_ivybridge_asm
PUBLIC cryptonigh_v8_mainloop_ryzen_asm
-ALIGN 64
+ALIGN 8
cryptonigh_v8_mainloop_ivybridge_asm PROC
- INCLUDE cryptonigh_v8_main_loop_ivybridge.inc
+ INCLUDE cryptonigh_v8_main_loop_ivybridge_win64.inc
ret 0
cryptonigh_v8_mainloop_ivybridge_asm ENDP
-ALIGN 64
+ALIGN 8
cryptonigh_v8_mainloop_ryzen_asm PROC
- INCLUDE cryptonigh_v8_main_loop_ryzen.inc
+ INCLUDE cryptonigh_v8_main_loop_ryzen_win64.inc
ret 0
cryptonigh_v8_mainloop_ryzen_asm ENDP
diff --git a/xmrstak/backend/cpu/crypto/asm/cryptonigh_v8_main_loop_ivybridge.inc b/xmrstak/backend/cpu/crypto/asm/cryptonigh_v8_main_loop_ivybridge_linux.inc
similarity index 91%
rename from xmrstak/backend/cpu/crypto/asm/cryptonigh_v8_main_loop_ivybridge.inc
rename to xmrstak/backend/cpu/crypto/asm/cryptonigh_v8_main_loop_ivybridge_linux.inc
index 1cc20b3..23f6cc0 100644
--- a/xmrstak/backend/cpu/crypto/asm/cryptonigh_v8_main_loop_ivybridge.inc
+++ b/xmrstak/backend/cpu/crypto/asm/cryptonigh_v8_main_loop_ivybridge_linux.inc
@@ -49,8 +49,8 @@
movq xmm0, rcx
punpcklqdq xmm5, xmm0
- ALIGN 64
-$main_loop_ivybridge:
+ ALIGN 8
+main_loop_ivybridge:
movdqu xmm6, XMMWORD PTR [r10+rbx]
lea rdx, QWORD PTR [r10+rbx]
mov ecx, r10d
@@ -105,10 +105,10 @@ $main_loop_ivybridge:
sqrtsd xmm3, xmm0
movq rdx, xmm3
test rdx, 524287
- je $sqrt_fixup_ivybridge
+ je sqrt_fixup_ivybridge
psrlq xmm3, 19
psubq xmm3, XMMWORD PTR [rsp+16]
-$sqrt_fixup_ivybridge_ret:
+sqrt_fixup_ivybridge_ret:
mov r9, r10
mov rax, rdi
@@ -138,7 +138,7 @@ $sqrt_fixup_ivybridge_ret:
and r10d, 2097136
xor r11, r12
dec rsi
- jne $main_loop_ivybridge
+ jne main_loop_ivybridge
ldmxcsr DWORD PTR [rsp]
mov rbx, QWORD PTR [rsp+160]
@@ -153,24 +153,24 @@ $sqrt_fixup_ivybridge_ret:
pop rdi
pop rsi
pop rbp
- jmp $cnv2_main_loop_ivybridge_endp
+ jmp cnv2_main_loop_ivybridge_endp
-$sqrt_fixup_ivybridge:
+sqrt_fixup_ivybridge:
dec rdx
- movq r13, -4389456576512
+ movq r13, -4389456576512
mov rax, rdx
shr rdx, 19
shr rax, 20
mov rcx, rdx
sub rcx, rax
add rax, r13
- movq r13, 4389456576511
+ movq r13, 4389456576511
sub rcx, r13
mov r13d, -2147483647
imul rcx, rax
sub rcx, r9
adc rdx, 0
movq xmm3, rdx
- jmp $sqrt_fixup_ivybridge_ret
+ jmp sqrt_fixup_ivybridge_ret
-$cnv2_main_loop_ivybridge_endp:
+cnv2_main_loop_ivybridge_endp:
diff --git a/xmrstak/backend/cpu/crypto/asm/cryptonigh_v8_main_loop_ivybridge_win64.inc b/xmrstak/backend/cpu/crypto/asm/cryptonigh_v8_main_loop_ivybridge_win64.inc
new file mode 100644
index 0000000..ee7f317
--- /dev/null
+++ b/xmrstak/backend/cpu/crypto/asm/cryptonigh_v8_main_loop_ivybridge_win64.inc
@@ -0,0 +1,176 @@
+ mov QWORD PTR [rsp+24], rbx
+ push rbp
+ push rsi
+ push rdi
+ push r12
+ push r13
+ push r14
+ push r15
+ sub rsp, 80
+
+ stmxcsr DWORD PTR [rsp]
+ mov DWORD PTR [rsp+4], 24448
+ ldmxcsr DWORD PTR [rsp+4]
+
+ mov rax, QWORD PTR [rcx+48]
+ mov r9, rcx
+ xor rax, QWORD PTR [rcx+16]
+ mov esi, 524288
+ mov r8, QWORD PTR [rcx+32]
+ mov r13d, -2147483647
+ xor r8, QWORD PTR [rcx]
+ mov r11, QWORD PTR [rcx+40]
+ mov r10, r8
+ mov rdx, QWORD PTR [rcx+56]
+ movd xmm4, rax
+ xor rdx, QWORD PTR [rcx+24]
+ xor r11, QWORD PTR [rcx+8]
+ mov rbx, QWORD PTR [rcx+224]
+ mov rax, QWORD PTR [r9+80]
+ xor rax, QWORD PTR [r9+64]
+ movd xmm0, rdx
+ mov rcx, QWORD PTR [rcx+88]
+ xor rcx, QWORD PTR [r9+72]
+ movq xmm3, QWORD PTR [r9+104]
+ movaps XMMWORD PTR [rsp+64], xmm6
+ movaps XMMWORD PTR [rsp+48], xmm7
+ movaps XMMWORD PTR [rsp+32], xmm8
+ and r10d, 2097136
+ movd xmm5, rax
+
+ xor eax, eax
+ mov QWORD PTR [rsp+16], rax
+
+ mov ax, 1023
+ shl rax, 52
+ movd xmm8, rax
+ mov r15, QWORD PTR [r9+96]
+ punpcklqdq xmm4, xmm0
+ movd xmm0, rcx
+ punpcklqdq xmm5, xmm0
+
+ ALIGN 8
+main_loop_ivybridge:
+ movdqu xmm6, XMMWORD PTR [r10+rbx]
+ lea rdx, QWORD PTR [r10+rbx]
+ mov ecx, r10d
+ mov eax, r10d
+ mov rdi, r15
+ xor ecx, 16
+ xor eax, 32
+ xor r10d, 48
+ movd xmm0, r11
+ movd xmm7, r8
+ punpcklqdq xmm7, xmm0
+ aesenc xmm6, xmm7
+ movdqu xmm1, XMMWORD PTR [rax+rbx]
+ movdqu xmm0, XMMWORD PTR [r10+rbx]
+ paddq xmm1, xmm7
+ movdqu xmm2, XMMWORD PTR [rcx+rbx]
+ paddq xmm0, xmm5
+ paddq xmm2, xmm4
+ movdqu XMMWORD PTR [rcx+rbx], xmm0
+ movd rcx, xmm3
+ movdqu XMMWORD PTR [rax+rbx], xmm2
+ mov rax, rcx
+ movdqu XMMWORD PTR [r10+rbx], xmm1
+ shl rax, 32
+ xor rdi, rax
+ movd rbp, xmm6
+ movdqa xmm0, xmm6
+ pxor xmm0, xmm4
+ mov r10, rbp
+ and r10d, 2097136
+ movdqu XMMWORD PTR [rdx], xmm0
+ xor rdi, QWORD PTR [r10+rbx]
+ lea r14, QWORD PTR [r10+rbx]
+ mov r12, QWORD PTR [r10+rbx+8]
+ xor edx, edx
+ lea r9d, DWORD PTR [ecx+ecx]
+ add r9d, ebp
+ movdqa xmm0, xmm6
+ psrldq xmm0, 8
+ or r9d, r13d
+ movd rax, xmm0
+ div r9
+ mov eax, eax
+ shl rdx, 32
+ add rdx, rax
+ lea r9, QWORD PTR [rdx+rbp]
+ mov r15, rdx
+ mov rax, r9
+ shr rax, 12
+ movd xmm0, rax
+ paddq xmm0, xmm8
+ sqrtsd xmm3, xmm0
+ movd rdx, xmm3
+ test rdx, 524287
+ je sqrt_fixup_ivybridge
+ psrlq xmm3, 19
+ psubq xmm3, XMMWORD PTR [rsp+16]
+sqrt_fixup_ivybridge_ret:
+
+ mov r9, r10
+ mov rax, rdi
+ mul rbp
+
+ xor r9, 16
+ mov rcx, r10
+ xor rcx, 32
+ xor r10, 48
+ add r8, rdx
+ add r11, rax
+ movdqu xmm0, XMMWORD PTR [r10+rbx]
+ movdqu xmm2, XMMWORD PTR [r9+rbx]
+ paddq xmm0, xmm5
+ movdqu xmm1, XMMWORD PTR [rcx+rbx]
+ paddq xmm2, xmm4
+ paddq xmm1, xmm7
+ movdqa xmm5, xmm4
+ movdqu XMMWORD PTR [r9+rbx], xmm0
+ movdqa xmm4, xmm6
+ movdqu XMMWORD PTR [rcx+rbx], xmm2
+ movdqu XMMWORD PTR [r10+rbx], xmm1
+ mov QWORD PTR [r14], r8
+ xor r8, rdi
+ mov r10, r8
+ mov QWORD PTR [r14+8], r11
+ and r10d, 2097136
+ xor r11, r12
+ dec rsi
+ jne main_loop_ivybridge
+
+ ldmxcsr DWORD PTR [rsp]
+ mov rbx, QWORD PTR [rsp+160]
+ movaps xmm6, XMMWORD PTR [rsp+64]
+ movaps xmm7, XMMWORD PTR [rsp+48]
+ movaps xmm8, XMMWORD PTR [rsp+32]
+ add rsp, 80
+ pop r15
+ pop r14
+ pop r13
+ pop r12
+ pop rdi
+ pop rsi
+ pop rbp
+ jmp cnv2_main_loop_ivybridge_endp
+
+sqrt_fixup_ivybridge:
+ dec rdx
+ mov r13, -4389456576512
+ mov rax, rdx
+ shr rdx, 19
+ shr rax, 20
+ mov rcx, rdx
+ sub rcx, rax
+ add rax, r13
+ mov r13, 4389456576511
+ sub rcx, r13
+ mov r13d, -2147483647
+ imul rcx, rax
+ sub rcx, r9
+ adc rdx, 0
+ movd xmm3, rdx
+ jmp sqrt_fixup_ivybridge_ret
+
+cnv2_main_loop_ivybridge_endp:
diff --git a/xmrstak/backend/cpu/crypto/asm/cryptonigh_v8_main_loop_ryzen.inc b/xmrstak/backend/cpu/crypto/asm/cryptonigh_v8_main_loop_ryzen_linux.inc
similarity index 92%
rename from xmrstak/backend/cpu/crypto/asm/cryptonigh_v8_main_loop_ryzen.inc
rename to xmrstak/backend/cpu/crypto/asm/cryptonigh_v8_main_loop_ryzen_linux.inc
index c564d89..551ee85 100644
--- a/xmrstak/backend/cpu/crypto/asm/cryptonigh_v8_main_loop_ryzen.inc
+++ b/xmrstak/backend/cpu/crypto/asm/cryptonigh_v8_main_loop_ryzen_linux.inc
@@ -45,8 +45,8 @@
movq xmm0, rcx
punpcklqdq xmm4, xmm0
- ALIGN 64
-$main_loop_ryzen:
+ ALIGN 8
+main_loop_ryzen:
movdqa xmm5, XMMWORD PTR [r10+rbx]
movq xmm0, r11
movq xmm6, r8
@@ -103,10 +103,10 @@ $main_loop_ryzen:
sqrtsd xmm1, xmm0
movq rdi, xmm1
test rdi, 524287
- je $sqrt_fixup_ryzen
+ je sqrt_fixup_ryzen
shr rdi, 19
-$sqrt_fixup_ryzen_ret:
+sqrt_fixup_ryzen_ret:
mov rax, rsi
mul r14
@@ -136,7 +136,7 @@ $sqrt_fixup_ryzen_ret:
and r10d, 2097136
movdqa xmm3, xmm5
dec ebp
- jne $main_loop_ryzen
+ jne main_loop_ryzen
ldmxcsr DWORD PTR [rsp]
movaps xmm6, XMMWORD PTR [rsp+48]
@@ -152,23 +152,23 @@ $sqrt_fixup_ryzen_ret:
pop r13
pop r12
pop rdi
- jmp $cnv2_main_loop_ryzen_endp
+ jmp cnv2_main_loop_ryzen_endp
-$sqrt_fixup_ryzen:
+sqrt_fixup_ryzen:
movq r9, xmm2
dec rdi
- movq rdx, 4389456576511
+ movq rdx, 4389456576511
mov rax, rdi
shr rdi, 19
shr rax, 20
mov rcx, rdi
sub rcx, rax
sub rcx, rdx
- movq rdx, -4389456576512
+ movq rdx, -4389456576512
add rax, rdx
imul rcx, rax
sub rcx, r9
adc rdi, 0
- jmp $sqrt_fixup_ryzen_ret
+ jmp sqrt_fixup_ryzen_ret
-$cnv2_main_loop_ryzen_endp:
+cnv2_main_loop_ryzen_endp:
diff --git a/xmrstak/backend/cpu/crypto/asm/cryptonigh_v8_main_loop_ryzen_win64.inc b/xmrstak/backend/cpu/crypto/asm/cryptonigh_v8_main_loop_ryzen_win64.inc
new file mode 100644
index 0000000..f70dcce
--- /dev/null
+++ b/xmrstak/backend/cpu/crypto/asm/cryptonigh_v8_main_loop_ryzen_win64.inc
@@ -0,0 +1,174 @@
+ mov QWORD PTR [rsp+16], rbx
+ mov QWORD PTR [rsp+24], rbp
+ mov QWORD PTR [rsp+32], rsi
+ push rdi
+ push r12
+ push r13
+ push r14
+ push r15
+ sub rsp, 64
+
+ stmxcsr DWORD PTR [rsp]
+ mov DWORD PTR [rsp+4], 24448
+ ldmxcsr DWORD PTR [rsp+4]
+
+ mov rax, QWORD PTR [rcx+48]
+ mov r9, rcx
+ xor rax, QWORD PTR [rcx+16]
+ mov ebp, 524288
+ mov r8, QWORD PTR [rcx+32]
+ xor r8, QWORD PTR [rcx]
+ mov r11, QWORD PTR [rcx+40]
+ mov r10, r8
+ mov rdx, QWORD PTR [rcx+56]
+ movd xmm3, rax
+ xor rdx, QWORD PTR [rcx+24]
+ xor r11, QWORD PTR [rcx+8]
+ mov rbx, QWORD PTR [rcx+224]
+ mov rax, QWORD PTR [r9+80]
+ xor rax, QWORD PTR [r9+64]
+ movd xmm0, rdx
+ mov rcx, QWORD PTR [rcx+88]
+ xor rcx, QWORD PTR [r9+72]
+ mov rdi, QWORD PTR [r9+104]
+ and r10d, 2097136
+ movaps XMMWORD PTR [rsp+48], xmm6
+ movd xmm4, rax
+ movaps XMMWORD PTR [rsp+32], xmm7
+ movaps XMMWORD PTR [rsp+16], xmm8
+ xorps xmm8, xmm8
+ mov ax, 1023
+ shl rax, 52
+ movd xmm7, rax
+ mov r15, QWORD PTR [r9+96]
+ punpcklqdq xmm3, xmm0
+ movd xmm0, rcx
+ punpcklqdq xmm4, xmm0
+
+ ALIGN 8
+main_loop_ryzen:
+ movdqa xmm5, XMMWORD PTR [r10+rbx]
+ movd xmm0, r11
+ movd xmm6, r8
+ punpcklqdq xmm6, xmm0
+ lea rdx, QWORD PTR [r10+rbx]
+ lea r9, QWORD PTR [rdi+rdi]
+ shl rdi, 32
+
+ mov ecx, r10d
+ mov eax, r10d
+ xor ecx, 16
+ xor eax, 32
+ xor r10d, 48
+ aesenc xmm5, xmm6
+ movdqa xmm2, XMMWORD PTR [rcx+rbx]
+ movdqa xmm1, XMMWORD PTR [rax+rbx]
+ movdqa xmm0, XMMWORD PTR [r10+rbx]
+ paddq xmm2, xmm3
+ paddq xmm1, xmm6
+ paddq xmm0, xmm4
+ movdqa XMMWORD PTR [rcx+rbx], xmm0
+ movdqa XMMWORD PTR [rax+rbx], xmm2
+ movdqa XMMWORD PTR [r10+rbx], xmm1
+
+ movaps xmm1, xmm8
+ mov rsi, r15
+ xor rsi, rdi
+ movd r14, xmm5
+ movdqa xmm0, xmm5
+ pxor xmm0, xmm3
+ mov r10, r14
+ and r10d, 2097136
+ movdqa XMMWORD PTR [rdx], xmm0
+ xor rsi, QWORD PTR [r10+rbx]
+ lea r12, QWORD PTR [r10+rbx]
+ mov r13, QWORD PTR [r10+rbx+8]
+
+ add r9d, r14d
+ or r9d, -2147483647
+ xor edx, edx
+ movdqa xmm0, xmm5
+ psrldq xmm0, 8
+ movd rax, xmm0
+
+ div r9
+ movd xmm0, rax
+ movd xmm1, rdx
+ punpckldq xmm0, xmm1
+ movd r15, xmm0
+ paddq xmm0, xmm5
+ movdqa xmm2, xmm0
+ psrlq xmm0, 12
+ paddq xmm0, xmm7
+ sqrtsd xmm1, xmm0
+ movd rdi, xmm1
+ test rdi, 524287
+ je sqrt_fixup_ryzen
+ shr rdi, 19
+
+sqrt_fixup_ryzen_ret:
+ mov rax, rsi
+ mul r14
+
+ mov r9d, r10d
+ mov ecx, r10d
+ xor r9d, 16
+ xor ecx, 32
+ xor r10d, 48
+ movdqa xmm0, XMMWORD PTR [r10+rbx]
+ movdqa xmm2, XMMWORD PTR [r9+rbx]
+ movdqa xmm1, XMMWORD PTR [rcx+rbx]
+ paddq xmm0, xmm4
+ paddq xmm2, xmm3
+ paddq xmm1, xmm6
+ movdqa XMMWORD PTR [r9+rbx], xmm0
+ movdqa XMMWORD PTR [rcx+rbx], xmm2
+ movdqa XMMWORD PTR [r10+rbx], xmm1
+
+ movdqa xmm4, xmm3
+ add r8, rdx
+ add r11, rax
+ mov QWORD PTR [r12], r8
+ xor r8, rsi
+ mov QWORD PTR [r12+8], r11
+ mov r10, r8
+ xor r11, r13
+ and r10d, 2097136
+ movdqa xmm3, xmm5
+ dec ebp
+ jne main_loop_ryzen
+
+ ldmxcsr DWORD PTR [rsp]
+ movaps xmm6, XMMWORD PTR [rsp+48]
+ lea r11, QWORD PTR [rsp+64]
+ mov rbx, QWORD PTR [r11+56]
+ mov rbp, QWORD PTR [r11+64]
+ mov rsi, QWORD PTR [r11+72]
+ movaps xmm8, XMMWORD PTR [r11-48]
+ movaps xmm7, XMMWORD PTR [rsp+32]
+ mov rsp, r11
+ pop r15
+ pop r14
+ pop r13
+ pop r12
+ pop rdi
+ jmp cnv2_main_loop_ryzen_endp
+
+sqrt_fixup_ryzen:
+ movd r9, xmm2
+ dec rdi
+ mov rdx, 4389456576511
+ mov rax, rdi
+ shr rdi, 19
+ shr rax, 20
+ mov rcx, rdi
+ sub rcx, rax
+ sub rcx, rdx
+ mov rdx, -4389456576512
+ add rax, rdx
+ imul rcx, rax
+ sub rcx, r9
+ adc rdi, 0
+ jmp sqrt_fixup_ryzen_ret
+
+cnv2_main_loop_ryzen_endp:
--
GitLab