bugfix for piledriver cgemm-, zgemm- and zgemv-kernel

kernel/x86_64/KERNEL.PILEDRIVER

@@ -2,7 +2,7 @@ SGEMVNKERNEL = sgemv_n.S
 SGEMVTKERNEL = sgemv_t.S
 
 ZGEMVNKERNEL = zgemv_n_dup.S
-ZGEMVTKERNEL = zgemv_t_dup.S
+ZGEMVTKERNEL = zgemv_t.S
 
 DGEMVNKERNEL = dgemv_n_bulldozer.S
 DGEMVTKERNEL = dgemv_t_bulldozer.S

kernel/x86_64/cgemm_kernel_4x2_piledriver.S

@@ -26,7 +26,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 *****************************************************************************/
 /*********************************************************************
 *
-* 2013/10/31 Saar
+* 2014/06/28 Saar
 *        BLASTEST               : OK
 *        CTEST                  : OK
 *        TEST                   : OK
@@ -546,16 +546,16 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #ifdef WINDOWS_ABI
 	movq	%rdi,    48(%rsp)
 	movq	%rsi,    56(%rsp)
-	movups	%xmm6,   64(%rsp)
-	movups	%xmm7,   80(%rsp)
-	movups	%xmm8,   96(%rsp)
-	movups	%xmm9,  112(%rsp)
-	movups	%xmm10, 128(%rsp)
-	movups	%xmm11, 144(%rsp)
-	movups	%xmm12, 160(%rsp)
-	movups	%xmm13, 176(%rsp)
-	movups	%xmm14, 192(%rsp)
-	movups	%xmm15, 208(%rsp)
+	vmovups	%xmm6,   64(%rsp)
+	vmovups	%xmm7,   80(%rsp)
+	vmovups	%xmm8,   96(%rsp)
+	vmovups	%xmm9,  112(%rsp)
+	vmovups	%xmm10, 128(%rsp)
+	vmovups	%xmm11, 144(%rsp)
+	vmovups	%xmm12, 160(%rsp)
+	vmovups	%xmm13, 176(%rsp)
+	vmovups	%xmm14, 192(%rsp)
+	vmovups	%xmm15, 208(%rsp)
 
 	movq	ARG1,      OLD_M
 	movq	ARG2,      OLD_N
@@ -568,6 +568,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	movsd	OLD_OFFSET, %xmm12
 #endif
 	vmovaps	%xmm3, %xmm0
+	vmovsd   OLD_ALPHA_I, %xmm1
 
 #else
 	movq	STACKSIZE +  8(%rsp), LDC
@@ -1889,6 +1890,8 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 
 .L999:
+	vzeroupper
+
 	movq   		SP, %rsp
 	movq	   (%rsp), %rbx
 	movq	  8(%rsp), %rbp
@@ -1900,16 +1903,16 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #ifdef WINDOWS_ABI
 	movq	 48(%rsp), %rdi
 	movq	 56(%rsp), %rsi
-	movups	 64(%rsp), %xmm6
-	movups	 80(%rsp), %xmm7
-	movups	 96(%rsp), %xmm8
-	movups	112(%rsp), %xmm9
-	movups	128(%rsp), %xmm10
-	movups	144(%rsp), %xmm11
-	movups	160(%rsp), %xmm12
-	movups	176(%rsp), %xmm13
-	movups	192(%rsp), %xmm14
-	movups	208(%rsp), %xmm15
+	vmovups	 64(%rsp), %xmm6
+	vmovups	 80(%rsp), %xmm7
+	vmovups	 96(%rsp), %xmm8
+	vmovups	112(%rsp), %xmm9
+	vmovups	128(%rsp), %xmm10
+	vmovups	144(%rsp), %xmm11
+	vmovups	160(%rsp), %xmm12
+	vmovups	176(%rsp), %xmm13
+	vmovups	192(%rsp), %xmm14
+	vmovups	208(%rsp), %xmm15
 #endif
 
 	addq	$STACKSIZE, %rsp

kernel/x86_64/zgemm_kernel_2x2_piledriver.S

@@ -27,7 +27,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 /*********************************************************************
 *
-* 2013/10/30 Saar
+* 2014/06/28 Saar
 *        BLASTEST               : OK
 *        CTEST                  : OK
 *        TEST                   : OK
@@ -437,16 +437,16 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #ifdef WINDOWS_ABI
 	movq	%rdi,    48(%rsp)
 	movq	%rsi,    56(%rsp)
-	movups	%xmm6,   64(%rsp)
-	movups	%xmm7,   80(%rsp)
-	movups	%xmm8,   96(%rsp)
-	movups	%xmm9,  112(%rsp)
-	movups	%xmm10, 128(%rsp)
-	movups	%xmm11, 144(%rsp)
-	movups	%xmm12, 160(%rsp)
-	movups	%xmm13, 176(%rsp)
-	movups	%xmm14, 192(%rsp)
-	movups	%xmm15, 208(%rsp)
+	vmovups	%xmm6,   64(%rsp)
+	vmovups	%xmm7,   80(%rsp)
+	vmovups	%xmm8,   96(%rsp)
+	vmovups	%xmm9,  112(%rsp)
+	vmovups	%xmm10, 128(%rsp)
+	vmovups	%xmm11, 144(%rsp)
+	vmovups	%xmm12, 160(%rsp)
+	vmovups	%xmm13, 176(%rsp)
+	vmovups	%xmm14, 192(%rsp)
+	vmovups	%xmm15, 208(%rsp)
 
 	movq	ARG1,      OLD_M
 	movq	ARG2,      OLD_N
@@ -456,14 +456,15 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	movq	OLD_C,     C
 	movq	OLD_LDC,   LDC
 #ifdef TRMMKERNEL
-	movsd	OLD_OFFSET, %xmm12
+	vmovsd	OLD_OFFSET, %xmm12
 #endif
 	vmovaps	%xmm3, %xmm0
+	vmovsd   OLD_ALPHA_I, %xmm1
 
 #else
 	movq	STACKSIZE +  8(%rsp), LDC
 #ifdef TRMMKERNEL
-	movsd	STACKSIZE + 16(%rsp), %xmm12
+	vmovsd	STACKSIZE + 16(%rsp), %xmm12
 #endif
 
 #endif
@@ -1397,6 +1398,8 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 
 .L999:
+	vzeroupper
+
 	movq   		SP, %rsp
 	movq	   (%rsp), %rbx
 	movq	  8(%rsp), %rbp
@@ -1408,16 +1411,16 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #ifdef WINDOWS_ABI
 	movq	 48(%rsp), %rdi
 	movq	 56(%rsp), %rsi
-	movups	 64(%rsp), %xmm6
-	movups	 80(%rsp), %xmm7
-	movups	 96(%rsp), %xmm8
-	movups	112(%rsp), %xmm9
-	movups	128(%rsp), %xmm10
-	movups	144(%rsp), %xmm11
-	movups	160(%rsp), %xmm12
-	movups	176(%rsp), %xmm13
-	movups	192(%rsp), %xmm14
-	movups	208(%rsp), %xmm15
+	vmovups	 64(%rsp), %xmm6
+	vmovups	 80(%rsp), %xmm7
+	vmovups	 96(%rsp), %xmm8
+	vmovups	112(%rsp), %xmm9
+	vmovups	128(%rsp), %xmm10
+	vmovups	144(%rsp), %xmm11
+	vmovups	160(%rsp), %xmm12
+	vmovups	176(%rsp), %xmm13
+	vmovups	192(%rsp), %xmm14
+	vmovups	208(%rsp), %xmm15
 #endif
 
 	addq	$STACKSIZE, %rsp