Fix Win64 ABI conformance issue in AVX2 ISLOW IDCT

xmm8-xmm11 must be saved and restored, since the function uses
ymm8-ymm11.

Closes #211

diff --git a/simd/x86_64/jidctint-avx2.asm b/simd/x86_64/jidctint-avx2.asm
index c009b45e9981442f7d72a5c405f1c6a3216bb948..b00151bbe9e51f1cbf6ed247e726d96cd17b9298 100644 (file)
--- a/simd/x86_64/jidctint-avx2.asm
+++ b/simd/x86_64/jidctint-avx2.asm
@@ -286,6 +286,7 @@ EXTN(jsimd_idct_islow_avx2):
     push        rbp
     mov         rax, rsp                     ; rax = original rbp
     mov         rbp, rsp                     ; rbp = aligned rbp
+    push_xmm    4
     collect_args 4
 
     ; ---- Pass 1: process columns.
@@ -409,6 +410,7 @@ EXTN(jsimd_idct_islow_avx2):
     movq        XMM_MMWORD [rsi+rax*SIZEOF_JSAMPLE], xmm7
 
     uncollect_args 4
+    pop_xmm     4
     pop         rbp
     ret