Adding VZEROUPPER to avoid potential AVX<->SSE transitions.

Change-Id: I6a23998f59086549957dc056fd2a42c273c8e78b
intel · Mar 20, 2018 · e1c591d · e1c591d
1 parent 83cfcbf
commit e1c591d
Show file tree

Hide file tree

Showing 15 changed files with 19 additions and 10 deletions.
diff --git a/avx2/mb_mgr_hmac_flush_avx2.asm b/avx2/mb_mgr_hmac_flush_avx2.asm
@@ -255,7 +255,7 @@ end_loop:
 	mov	[p + 2*4], DWORD(tmp3)
 
 return:
-
+        vzeroupper
 	mov	rbp, [rsp + _gpr_save + 8*0]
 	mov	r12, [rsp + _gpr_save + 8*1]
 	mov	r13, [rsp + _gpr_save + 8*2]

diff --git a/avx2/mb_mgr_hmac_md5_flush_avx2.asm b/avx2/mb_mgr_hmac_md5_flush_avx2.asm
@@ -300,6 +300,7 @@ end_loop:
 
 return:
         DBGPRINTL "---------- exit md5 flush -----------"
+        vzeroupper
 
 	mov	rbx, [rsp + _gpr_save + 8*0]
 	mov	rbp, [rsp + _gpr_save + 8*1]

diff --git a/avx2/mb_mgr_hmac_md5_submit_avx2.asm b/avx2/mb_mgr_hmac_md5_submit_avx2.asm
@@ -316,8 +316,9 @@ end_loop:
         mov	[p + 1*4], DWORD(tmp2)
         mov	[p + 2*4], DWORD(tmp3)
 
+        vzeroupper
 return:
-  DBGPRINTL "---------- exit md5 submit -----------"
+        DBGPRINTL "---------- exit md5 submit -----------"
 
 	mov	rbx, [rsp + _gpr_save + 8*0]
 	mov	rbp, [rsp + _gpr_save + 8*1]

diff --git a/avx2/mb_mgr_hmac_sha_256_flush_avx2.asm b/avx2/mb_mgr_hmac_sha_256_flush_avx2.asm
@@ -278,6 +278,7 @@ end_loop:
 %endif
 
 return:
+        vzeroupper
 
 	mov	rbx, [rsp + _gpr_save + 8*0]
 	mov	rbp, [rsp + _gpr_save + 8*1]

diff --git a/avx2/mb_mgr_hmac_sha_256_submit_avx2.asm b/avx2/mb_mgr_hmac_sha_256_submit_avx2.asm
@@ -318,8 +318,8 @@ end_loop:
 %else
 	mov	[p + 3*4], DWORD(tmp4)
 %endif
+        vzeroupper
 return:
-
 	mov	rbx, [rsp + _gpr_save + 8*0]
 	mov	rbp, [rsp + _gpr_save + 8*1]
 	mov	r12, [rsp + _gpr_save + 8*2]

diff --git a/avx2/mb_mgr_hmac_sha_512_flush_avx2.asm b/avx2/mb_mgr_hmac_sha_512_flush_avx2.asm
@@ -252,6 +252,7 @@ end_loop:
 %endif
 
 return:
+        vzeroupper
 
 	mov	rbx, [rsp + _gpr_save + 8*0]
 	mov	rbp, [rsp + _gpr_save + 8*1]

diff --git a/avx2/mb_mgr_hmac_sha_512_submit_avx2.asm b/avx2/mb_mgr_hmac_sha_512_submit_avx2.asm
@@ -309,8 +309,8 @@ end_loop:
 	mov	[p + 3*8], QWORD(tmp4)
 %endif
 
+        vzeroupper
 return:
-
 	mov	rbx, [rsp + _gpr_save + 8*0]
 	mov	rbp, [rsp + _gpr_save + 8*1]
 	mov	r12, [rsp + _gpr_save + 8*2]

diff --git a/avx2/mb_mgr_hmac_submit_avx2.asm b/avx2/mb_mgr_hmac_submit_avx2.asm
@@ -297,9 +297,9 @@ end_loop:
         mov	[p + 0*SHA1_DIGEST_WORD_SIZE], DWORD(tmp)
         mov	[p + 1*SHA1_DIGEST_WORD_SIZE], DWORD(tmp2)
         mov	[p + 2*SHA1_DIGEST_WORD_SIZE], DWORD(tmp3)
-        DBGPRINTL "---------- exit sha1 submit -----------"
+        vzeroupper
 return:
-
+        DBGPRINTL "---------- exit sha1 submit -----------"
         mov	rbp, [rsp + _gpr_save + 8*0]
         mov	r12, [rsp + _gpr_save + 8*1]
         mov	r13, [rsp + _gpr_save + 8*2]

diff --git a/avx512/mb_mgr_des_avx512.asm b/avx512/mb_mgr_des_avx512.asm
@@ -243,6 +243,7 @@ extern des3_x16_cbc_dec_avx512
         mov     rax, [STATE + _des_job_in_lane + MIN_IDX*8]
         mov     qword [STATE + _des_job_in_lane + MIN_IDX*8], 0
         or	dword [rax + _status], STS_COMPLETED_AES
+        vzeroupper
 %%_des_submit_return:
 %endmacro
 
@@ -386,6 +387,7 @@ extern des3_x16_cbc_dec_avx512
         ;; - clear job pointer
         mov     qword [STATE + _des_job_in_lane + MIN_IDX*8], 0
 %%_des_flush_return:
+        vzeroupper
 %endmacro
 
 ;;; ========================================================

diff --git a/avx512/mb_mgr_hmac_flush_avx512.asm b/avx512/mb_mgr_hmac_flush_avx512.asm
@@ -306,6 +306,7 @@ end_loop:
 
 return:
         DBGPRINTL "---------- exit hmac flush avx512 -----------"
+        vzeroupper
 
         mov	rbp, [rsp + _gpr_save + 8*0]
         mov	r12, [rsp + _gpr_save + 8*1]

diff --git a/avx512/mb_mgr_hmac_sha_256_flush_avx512.asm b/avx512/mb_mgr_hmac_sha_256_flush_avx512.asm
@@ -329,6 +329,7 @@ end_loop:
 %endif
 
 return:
+        vzeroupper
 
 	mov	rbx, [rsp + _gpr_save + 8*0]
 	mov	rbp, [rsp + _gpr_save + 8*1]

diff --git a/avx512/mb_mgr_hmac_sha_256_submit_avx512.asm b/avx512/mb_mgr_hmac_sha_256_submit_avx512.asm
@@ -352,8 +352,8 @@ end_loop:
 %else
 	mov	[p + 3*4], DWORD(tmp4)
 %endif
+        vzeroupper
 return:
-
 	mov	rbx, [rsp + _gpr_save + 8*0]
 	mov	rbp, [rsp + _gpr_save + 8*1]
 	mov	r12, [rsp + _gpr_save + 8*2]

diff --git a/avx512/mb_mgr_hmac_sha_512_flush_avx512.asm b/avx512/mb_mgr_hmac_sha_512_flush_avx512.asm
@@ -281,6 +281,7 @@ end_loop:
 %endif
 
 return:
+        vzeroupper
 
 	mov	rbx, [rsp + _gpr_save + 8*0]
 	mov	rbp, [rsp + _gpr_save + 8*1]

diff --git a/avx512/mb_mgr_hmac_sha_512_submit_avx512.asm b/avx512/mb_mgr_hmac_sha_512_submit_avx512.asm
@@ -320,7 +320,7 @@ end_loop:
 %if (SHA_X_DIGEST_SIZE != 384)
 	mov	[p + 3*8], QWORD(tmp4)
 %endif
-
+        vzeroupper
 return:
 	mov	rbx, [rsp + _gpr_save + 8*0]
 	mov	rbp, [rsp + _gpr_save + 8*1]

diff --git a/avx512/mb_mgr_hmac_submit_avx512.asm b/avx512/mb_mgr_hmac_submit_avx512.asm
@@ -346,9 +346,9 @@ end_loop:
         mov	[p + 1*SHA1_DIGEST_WORD_SIZE], DWORD(tmp2)
         mov	[p + 2*SHA1_DIGEST_WORD_SIZE], DWORD(tmp3)
 
-        DBGPRINTL "---------- exit sha1 submit -----------"
+        vzeroupper
 return:
-
+        DBGPRINTL "---------- exit sha1 submit -----------"
         mov	rbp, [rsp + _gpr_save + 8*0]
         mov	r12, [rsp + _gpr_save + 8*1]
         mov	r13, [rsp + _gpr_save + 8*2]
-Original file line number
+Diff line change
@@ Expand Up / @@ -278,6 +278,7 @@ end_loop: @@
     %endif
     return:
+            vzeroupper
     	mov	rbx, [rsp + _gpr_save + 8*0]
     	mov	rbp, [rsp + _gpr_save + 8*1]
@@ Expand Down @@