Add forgotten conditional uses of PREFETCH

This fixes a (cross-)compilation/linker error for PRESCOTT
on Yocto.

Signed-off-by: Zoltán Böszörményi <zoltan.boszormenyi@xenial.com>

kernel/x86_64/gemm_ncopy_4.S

@@ -189,12 +189,16 @@
 	movss	%xmm6,   6 * SIZE(B)
 	movss	%xmm7,   7 * SIZE(B)
 
+#ifdef PREFETCH
 	PREFETCH	RPREFETCHSIZE * SIZE(AO1)
 	PREFETCH	RPREFETCHSIZE * SIZE(AO2)
 	PREFETCH	RPREFETCHSIZE * SIZE(AO3)
 	PREFETCH	RPREFETCHSIZE * SIZE(AO4)
+#endif
 
+#ifdef PREFETCHW
 	PREFETCHW	WPREFETCHSIZE * SIZE(B)
+#endif
 
 	movss	%xmm8,   8 * SIZE(B)
 	movss	%xmm9,   9 * SIZE(B)
@@ -205,29 +209,39 @@
 	movss	%xmm14, 14 * SIZE(B)
 	movss	%xmm15, 15 * SIZE(B)
 #else
+#ifdef PREFETCH
 	PREFETCH	RPREFETCHSIZE * SIZE(AO1)
+#endif
 	movsd	0 * SIZE(AO1), %xmm0
 	movhpd	0 * SIZE(AO2), %xmm0
 	movsd	1 * SIZE(AO1), %xmm2
 	movhpd	1 * SIZE(AO2), %xmm2
+#ifdef PREFETCH
 	PREFETCH	RPREFETCHSIZE * SIZE(AO2)
+#endif
 	movsd	2 * SIZE(AO1), %xmm4
 	movhpd	2 * SIZE(AO2), %xmm4
 	movsd	3 * SIZE(AO1), %xmm6
 	movhpd	3 * SIZE(AO2), %xmm6
 
+#ifdef PREFETCH
 	PREFETCH	RPREFETCHSIZE * SIZE(AO3)
+#endif
 	movsd	0 * SIZE(AO3), %xmm1
 	movhpd	0 * SIZE(AO4), %xmm1
 	movsd	1 * SIZE(AO3), %xmm3
 	movhpd	1 * SIZE(AO4), %xmm3
+#ifdef PREFETCH
 	PREFETCH	RPREFETCHSIZE * SIZE(AO4)
+#endif
 	movsd	2 * SIZE(AO3), %xmm5
 	movhpd	2 * SIZE(AO4), %xmm5
 	movsd	3 * SIZE(AO3), %xmm7
 	movhpd	3 * SIZE(AO4), %xmm7
 
+#ifdef PREFETCHW
 	PREFETCHW	WPREFETCHSIZE * SIZE(B)
+#endif
 	movapd	%xmm0,   0 * SIZE(B)
 	movapd	%xmm1,   2 * SIZE(B)
 	movapd	%xmm2,   4 * SIZE(B)
@@ -342,10 +356,14 @@
 	movapd	%xmm3,   6 * SIZE(B)
 #endif
 
+#ifdef PREFETCH
 	PREFETCH	RPREFETCHSIZE * SIZE(AO1)
 	PREFETCH	RPREFETCHSIZE * SIZE(AO2)
+#endif
 
+#ifdef PREFETCHW
 	PREFETCHW	WPREFETCHSIZE * SIZE(B)
+#endif
 
 	addq	$4 * SIZE, AO1
 	addq	$4 * SIZE, AO2

kernel/x86_64/gemm_tcopy_4.S

@@ -219,31 +219,41 @@
 	movaps	%xmm3,   12 * SIZE(BO)
 #else
 
+#ifdef PREFETCH
 	PREFETCH	RPREFETCHSIZE * SIZE(AO1)
+#endif
 	movsd	0 * SIZE(AO1), %xmm0
 	movhpd	1 * SIZE(AO1), %xmm0
 	movsd	2 * SIZE(AO1), %xmm1
 	movhpd	3 * SIZE(AO1), %xmm1
 
+#ifdef PREFETCH
 	PREFETCH	RPREFETCHSIZE * SIZE(AO2)
+#endif
 	movsd	0 * SIZE(AO2), %xmm2
 	movhpd	1 * SIZE(AO2), %xmm2
 	movsd	2 * SIZE(AO2), %xmm3
 	movhpd	3 * SIZE(AO2), %xmm3
 
+#ifdef PREFETCH
 	PREFETCH	RPREFETCHSIZE * SIZE(AO3)
+#endif
 	movsd	0 * SIZE(AO3), %xmm4
 	movhpd	1 * SIZE(AO3), %xmm4
 	movsd	2 * SIZE(AO3), %xmm5
 	movhpd	3 * SIZE(AO3), %xmm5
 
+#ifdef PREFETCH
 	PREFETCH	RPREFETCHSIZE * SIZE(AO4)
+#endif
 	movsd	0 * SIZE(AO4), %xmm6
 	movhpd	1 * SIZE(AO4), %xmm6
 	movsd	2 * SIZE(AO4), %xmm7
 	movhpd	3 * SIZE(AO4), %xmm7
 
+#ifdef PREFETCHW
 	PREFETCHW	WPREFETCHSIZE * SIZE(BO)
+#endif
 	movapd	%xmm0,    0 * SIZE(BO)
 	movapd	%xmm1,    2 * SIZE(BO)
 	movapd	%xmm2,    4 * SIZE(BO)

kernel/x86_64/zgemm_kernel_4x2_sse.S

@@ -102,6 +102,14 @@
 #define RPREFETCHSIZE (8 *  7 + 4)
 #define WPREFETCHSIZE (8 *  8 + 4)
 
+#ifdef PREFETCH
+#define PREFETCH_KERNEL1(xx) PREFETCH	(PREFETCHSIZE	  +  0) * SIZE + 1 * (xx) * SIZE(AO, %rax, 4) ;
+#define PREFETCH_KERNEL4(xx) PREFETCH	(PREFETCHSIZE     + 16) * SIZE + 1 * (xx) * SIZE(AO, %rax, 4) ;
+#else
+#define PREFETCH_KERNEL1(xx)
+#define PREFETCH_KERNEL4(xx)
+#endif
+
 #ifndef GENERIC
 #define KERNEL1(xx) \
 	mulps	%xmm0, %xmm1 ;\
@@ -111,7 +119,7 @@
 	addps	%xmm3, %xmm9 ;\
 	movaps	-28 * SIZE + 2 * (xx) * SIZE(BO, %rax, 8), %xmm3 ;\
 	mulps	%xmm0, %xmm5 ;\
-	PREFETCH	(PREFETCHSIZE     +  0) * SIZE + 1 * (xx) * SIZE(AO, %rax, 4) ;\
+	PREFETCH_KERNEL1(xx) \
 	mulps	-20 * SIZE + 2 * (xx) * SIZE(BO, %rax, 8), %xmm0 ;\
 	addps	%xmm5, %xmm10 ;\
 	movaps	-24 * SIZE + 2 * (xx) * SIZE(BO, %rax, 8), %xmm5 ;\
@@ -157,7 +165,7 @@
 	mulps	 -4 * SIZE + 2 * (xx) * SIZE(BO, %rax, 8), %xmm6 ;\
 	addps	%xmm5, %xmm14 ;\
 	movaps	  8 * SIZE + 2 * (xx) * SIZE(BO, %rax, 8), %xmm5 ;\
- 	PREFETCH	(PREFETCHSIZE     + 16) * SIZE + 1 * (xx) * SIZE(AO, %rax, 4) ;\
+	PREFETCH_KERNEL4(xx) \
 	addps	%xmm6, %xmm15 ;\
 	movaps	 -4 * SIZE + 1 * (xx) * SIZE(AO, %rax, 4), %xmm6
 
@@ -1026,7 +1034,9 @@
 
 .L22:
 	mulps	%xmm0, %xmm1
+#ifdef PREFETCH
 	PREFETCH	(PREFETCHSIZE +  0) * SIZE(AO)
+#endif
 	addps	%xmm1, %xmm8
 	movaps	-28 * SIZE(BO), %xmm1
 	mulps	%xmm0, %xmm1
@@ -1079,7 +1089,9 @@
 	movaps	  0 * SIZE(AO), %xmm0
 
 	mulps	%xmm2, %xmm1
+#ifdef PREFETCH
 	PREFETCH	(PREFETCHSIZE + 16) * SIZE(AO)
+#endif
 	addps	%xmm1, %xmm8
 	movaps	 36 * SIZE(BO), %xmm1
 	mulps	%xmm2, %xmm1
@@ -1285,7 +1297,9 @@
 
 .L32:
 	mulps	%xmm0, %xmm1
+#ifdef PREFETCH
 	PREFETCH	(PREFETCHSIZE +  0) * SIZE(AO)
+#endif
 	addps	%xmm1, %xmm8
 	movaps	-28 * SIZE(BO), %xmm1
 	mulps	%xmm0, %xmm1
@@ -1679,7 +1693,9 @@
 
 .L52:
 	mulps	%xmm0, %xmm1
+#ifdef PREFETCH
 	PREFETCH	(PREFETCHSIZE +  0) * SIZE(AO)
+#endif
 	mulps	-28 * SIZE(BO), %xmm0
 	addps	%xmm1, %xmm8
 	movaps	-32 * SIZE(BO), %xmm1
@@ -1705,7 +1721,9 @@
 	addps	%xmm0, %xmm13
 	movaps	 32 * SIZE(AO), %xmm0
 
+#ifdef PREFETCH
 	PREFETCH	(PREFETCHSIZE + 16) * SIZE(AO)
+#endif
 
 	mulps	%xmm2, %xmm3
 	mulps	-12 * SIZE(BO), %xmm2
@@ -1733,7 +1751,9 @@
 	addps	%xmm2, %xmm13
 	movaps	 48 * SIZE(AO), %xmm2
 
+#ifdef PREFETCH
 	PREFETCH	(PREFETCHSIZE + 32) * SIZE(AO)
+#endif
 
 	mulps	%xmm4, %xmm5
 	mulps	  4 * SIZE(BO), %xmm4
@@ -1761,7 +1781,9 @@
 	addps	%xmm4, %xmm13
 	movaps	 64 * SIZE(AO), %xmm4
 
+#ifdef PREFETCH
 	PREFETCH	(PREFETCHSIZE + 48) * SIZE(AO)
+#endif
 
 	mulps	%xmm6, %xmm7
 	mulps	 20 * SIZE(BO), %xmm6
@@ -1942,7 +1964,9 @@
 
 .L62:
 	mulps	%xmm0, %xmm1
+#ifdef PREFETCH
 	PREFETCH	(PREFETCHSIZE +  0) * SIZE(AO)
+#endif
 	mulps	-28 * SIZE(BO), %xmm0
 	addps	%xmm1, %xmm8
 	movaps	-24 * SIZE(BO), %xmm1
@@ -1968,7 +1992,9 @@
 	addps	%xmm0, %xmm11
 	movaps    0 * SIZE(AO), %xmm0
 
+#ifdef PREFETCH
 	PREFETCH	(PREFETCHSIZE + 16) * SIZE(AO)
+#endif
 
 	mulps	%xmm2, %xmm5
 	mulps	  4 * SIZE(BO), %xmm2
@@ -2130,7 +2156,9 @@
 
 .L72:
 	mulps	%xmm0, %xmm1
+#ifdef PREFETCH
 	PREFETCH	(PREFETCHSIZE +  0) * SIZE(AO)
+#endif
 	addps	%xmm1, %xmm8
 	movaps	-28 * SIZE(BO), %xmm1
 	mulps	%xmm0, %xmm1

kernel/x86_64/zsymv_L_sse2.S

@@ -484,7 +484,9 @@
 	addpd	  a1,     yy1
 	MOVDDUP(1 * SIZE, A1, a1)
 
+#ifdef PREFETCH
 	PREFETCH	PREFETCHSIZE(A1)
+#endif
 
 	movapd	  xtemp3, xt1
 	mulpd	  a2,     xt1
@@ -507,7 +509,9 @@
 	addpd	  a2,     yy2
 	MOVDDUP(0 * SIZE, A2, a2)
 
+#ifdef PREFETCH
 	PREFETCH	PREFETCHSIZE(XX)
+#endif
 
 	movapd	  xtemp3, xt1
 	movapd	12 * SIZE(XX), xtemp3
@@ -546,7 +550,9 @@
 	addpd	  a2,     yy1
 	MOVDDUP(6 * SIZE, A2, a2)
 
+#ifdef PREFETCH
 	PREFETCH	PREFETCHSIZE(A2)
+#endif
 
 	movlpd	 yy1, 0 * SIZE(YY)
 	movhpd	 yy1, 1 * SIZE(YY)
@@ -574,7 +580,9 @@
 	addpd	  a1,     yy1
 	MOVDDUP(6 * SIZE, A1, a1)
 
+#ifdef PREFETCHW
 	PREFETCHW	PREFETCHSIZE(YY)
+#endif
 
 	movapd	  xtemp4, xt1
 	mulpd	  a2,     xt1

kernel/x86_64/zsymv_U_sse2.S

@@ -442,7 +442,9 @@
 	addpd	  a1,     yy1
 	MOVDDUP(3 * SIZE, A2, a1)
 
+#ifdef PREFETCH
 	PREFETCH	PREFETCHSIZE(A1)
+#endif
 
 	movapd	  xtemp3, xt1
 	mulpd	  a2,     xt1
@@ -465,7 +467,9 @@
 	addpd	  a1,     yy2
 	MOVDDUP(3 * SIZE, A1, a1)
 
+#ifdef PREFETCH
 	PREFETCH	PREFETCHSIZE(XX)
+#endif
 
 	movapd	  xtemp3, xt1
 	movapd	12 * SIZE(XX), xtemp3
@@ -504,7 +508,9 @@
 	addpd	  a2,     yy1
 	MOVDDUP(5 * SIZE, A1, a2)
 
+#ifdef PREFETCH
 	PREFETCH	PREFETCHSIZE(A2)
+#endif
 
 	movlpd	 yy1, 0 * SIZE(YY)
 	movhpd	 yy1, 1 * SIZE(YY)
@@ -532,7 +538,9 @@
 	addpd	  a2,     yy1
 	MOVDDUP(4 * SIZE, A2, a2)
 
+#ifdef PREFETCH
 	PREFETCHW	PREFETCHSIZE(YY)
+#endif
 
 	movapd	  xtemp4, xt1
 	mulpd	  a3,     xt1

kernel/x86_64/ztrsm_kernel_LN_4x2_sse.S

@@ -109,12 +109,20 @@
 #define PREFETCHSIZE (8 * 6 + 4)
 #endif
 
+#ifdef PREFETCH
+#define PREFETCH_KERNEL1(xx) PREFETCH	(PREFETCHSIZE     +  0) * SIZE + 1 * (xx) * SIZE(AO) ;
+#define PREFETCH_KERNEL5(xx) PREFETCH	(PREFETCHSIZE     +  8) * SIZE + 1 * (xx) * SIZE(AO) ;
+#else
+#define PREFETCH_KERNEL1(xx)
+#define PREFETCH_KERNEL5(xx)
+#endif
+
 #define KERNEL1(xx) \
 	mulps	%xmm8, %xmm9 ;\
 	addps	%xmm9, %xmm0 ;\
 	movaps	 0 * SIZE + 2 * (xx) * SIZE(BO), %xmm9 ;\
 	mulps	%xmm8, %xmm11 ;\
-	PREFETCH	(PREFETCHSIZE     +  0) * SIZE + 1 * (xx) * SIZE(AO) ;\
+	PREFETCH_KERNEL1(xx) \
 	addps	%xmm11, %xmm1 ;\
 	movaps	 4 * SIZE + 2 * (xx) * SIZE(BO), %xmm11 ;\
 	mulps	%xmm8, %xmm13 ;\
@@ -171,7 +179,7 @@
 	addps	%xmm9, %xmm0 ;\
 	movaps	32 * SIZE + 2 * (xx) * SIZE(BO), %xmm9 ;\
 	mulps	%xmm8, %xmm11 ;\
-	PREFETCH	(PREFETCHSIZE     +  8) * SIZE + 1 * (xx) * SIZE(AO) ;\
+	PREFETCH_KERNEL5(xx) \
 	addps	%xmm11, %xmm1 ;\
 	movaps	36 * SIZE + 2 * (xx) * SIZE(BO), %xmm11 ;\
 	mulps	%xmm8, %xmm13 ;\

kernel/x86_64/ztrsm_kernel_LT_4x2_sse.S

@@ -109,12 +109,20 @@
 #define PREFETCHSIZE (8 * 6 + 4)
 #endif
 
+#ifdef PREFETCH
+#define PREFETCH_KERNEL1(xx) PREFETCH	(PREFETCHSIZE     +  0) * SIZE + 1 * (xx) * SIZE(AO) ;
+#define PREFETCH_KERNEL5(xx) PREFETCH	(PREFETCHSIZE     +  8) * SIZE + 1 * (xx) * SIZE(AO) ;
+#else
+#define PREFETCH_KERNEL1(xx)
+#define PREFETCH_KERNEL5(xx)
+#endif
+
 #define KERNEL1(xx) \
 	mulps	%xmm8, %xmm9 ;\
 	addps	%xmm9, %xmm0 ;\
 	movaps	 0 * SIZE + 2 * (xx) * SIZE(BO), %xmm9 ;\
 	mulps	%xmm8, %xmm11 ;\
-	PREFETCH	(PREFETCHSIZE     +  0) * SIZE + 1 * (xx) * SIZE(AO) ;\
+	PREFETCH_KERNEL1(xx) \
 	addps	%xmm11, %xmm1 ;\
 	movaps	 4 * SIZE + 2 * (xx) * SIZE(BO), %xmm11 ;\
 	mulps	%xmm8, %xmm13 ;\
@@ -171,7 +179,7 @@
 	addps	%xmm9, %xmm0 ;\
 	movaps	32 * SIZE + 2 * (xx) * SIZE(BO), %xmm9 ;\
 	mulps	%xmm8, %xmm11 ;\
-	PREFETCH	(PREFETCHSIZE     +  8) * SIZE + 1 * (xx) * SIZE(AO) ;\
+	PREFETCH_KERNEL5(xx) \
 	addps	%xmm11, %xmm1 ;\
 	movaps	36 * SIZE + 2 * (xx) * SIZE(BO), %xmm11 ;\
 	mulps	%xmm8, %xmm13 ;\

kernel/x86_64/ztrsm_kernel_RT_4x2_sse.S

@@ -109,12 +109,20 @@
 #define PREFETCHSIZE (8 * 6 + 4)
 #endif
 
+#ifdef PREFETCH
+#define PREFETCH_KERNEL1(xx) PREFETCH	(PREFETCHSIZE     +  0) * SIZE + 1 * (xx) * SIZE(AO) ;
+#define PREFETCH_KERNEL5(xx) PREFETCH	(PREFETCHSIZE     +  8) * SIZE + 1 * (xx) * SIZE(AO) ;
+#else
+#define PREFETCH_KERNEL1(xx)
+#define PREFETCH_KERNEL5(xx)
+#endif
+
 #define KERNEL1(xx) \
 	mulps	%xmm8, %xmm9 ;\
 	addps	%xmm9, %xmm0 ;\
 	movaps	 0 * SIZE + 2 * (xx) * SIZE(BO), %xmm9 ;\
 	mulps	%xmm8, %xmm11 ;\
-	PREFETCH	(PREFETCHSIZE     +  0) * SIZE + 1 * (xx) * SIZE(AO) ;\
+	PREFETCH_KERNEL1(xx) \
 	addps	%xmm11, %xmm1 ;\
 	movaps	 4 * SIZE + 2 * (xx) * SIZE(BO), %xmm11 ;\
 	mulps	%xmm8, %xmm13 ;\
@@ -171,7 +179,7 @@
 	addps	%xmm9, %xmm0 ;\
 	movaps	32 * SIZE + 2 * (xx) * SIZE(BO), %xmm9 ;\
 	mulps	%xmm8, %xmm11 ;\
-	PREFETCH	(PREFETCHSIZE     +  8) * SIZE + 1 * (xx) * SIZE(AO) ;\
+	PREFETCH_KERNEL5(xx) \
 	addps	%xmm11, %xmm1 ;\
 	movaps	36 * SIZE + 2 * (xx) * SIZE(BO), %xmm11 ;\
 	mulps	%xmm8, %xmm13 ;\