optimize get_conf_stride_1()

diff --git a/src/fdr/fdr.c b/src/fdr/fdr.c
index 356cc3e6c39eea5ce299f88e493b137587b54376..715ab68469c018ec2fd216150d18aa958faa9def 100644 (file)
--- a/src/fdr/fdr.c
+++ b/src/fdr/fdr.c
@@ -147,74 +147,43 @@ void get_conf_stride_1(const u8 *itPtr, UNUSED const u8 *start_ptr,
                        const u64a *ft, u64a *conf0, u64a *conf8, m128 *s) {
     /* +1: the zones ensure that we can read the byte at z->end */
     assert(itPtr >= start_ptr && itPtr + ITER_BYTES <= end_ptr);
-
-    u64a ALIGN_ATTR(16) ptr[16];
-    ptr[0] = unaligned_load_u32(itPtr + 0);
-    ptr[1] = unaligned_load_u32(itPtr + 1);
-    ptr[2] = unaligned_load_u32(itPtr + 2);
-    ptr[3] = unaligned_load_u32(itPtr + 3);
-    ptr[4] = unaligned_load_u32(itPtr + 4);
-    ptr[5] = unaligned_load_u32(itPtr + 5);
-    ptr[6] = unaligned_load_u32(itPtr + 6);
-    ptr[7] = unaligned_load_u32(itPtr + 7);
-    ptr[8]  = unaligned_load_u32(itPtr + 8);
-    ptr[9]  = unaligned_load_u32(itPtr + 9);
-    ptr[10] = unaligned_load_u32(itPtr + 10);
-    ptr[11] = unaligned_load_u32(itPtr + 11);
-    ptr[12] = unaligned_load_u32(itPtr + 12);
-    ptr[13] = unaligned_load_u32(itPtr + 13);
-    ptr[14] = unaligned_load_u32(itPtr + 14);
-    ptr[15] = unaligned_load_u32(itPtr + 15);
-
-    u64a mask_not = ~domain_mask_flipped;
-    u64a reach0 = mask_not & ptr[0];
-    u64a reach1 = mask_not & ptr[1];
-    u64a reach2 = mask_not & ptr[2];
-    u64a reach3 = mask_not & ptr[3];
-    u64a reach4 = mask_not & ptr[4];
-    u64a reach5 = mask_not & ptr[5];
-    u64a reach6 = mask_not & ptr[6];
-    u64a reach7 = mask_not & ptr[7];
-    u64a reach8  = mask_not & ptr[8];
-    u64a reach9  = mask_not & ptr[9];
-    u64a reach10 = mask_not & ptr[10];
-    u64a reach11 = mask_not & ptr[11];
-    u64a reach12 = mask_not & ptr[12];
-    u64a reach13 = mask_not & ptr[13];
-    u64a reach14 = mask_not & ptr[14];
-    u64a reach15 = mask_not & ptr[15];
-
-    m128 st0 = load_m128_from_u64a(ft + reach0);
-    m128 st1 = load_m128_from_u64a(ft + reach1);
-    m128 st2 = load_m128_from_u64a(ft + reach2);
-    m128 st3 = load_m128_from_u64a(ft + reach3);
-    m128 st4 = load_m128_from_u64a(ft + reach4);
-    m128 st5 = load_m128_from_u64a(ft + reach5);
-    m128 st6 = load_m128_from_u64a(ft + reach6);
-    m128 st7 = load_m128_from_u64a(ft + reach7);
+    u64a domain_mask = ~domain_mask_flipped;
+
+    u64a it_hi = *(const u64a *)itPtr;
+    u64a it_lo = *(const u64a *)(itPtr + 8);
+    u64a reach0  = domain_mask & it_hi;
+    u64a reach1  = domain_mask & (it_hi >> 8);
+    u64a reach2  = domain_mask & (it_hi >> 16);
+    u64a reach3  = domain_mask & (it_hi >> 24);
+    u64a reach4  = domain_mask & (it_hi >> 32);
+    u64a reach5  = domain_mask & (it_hi >> 40);
+    u64a reach6  = domain_mask & (it_hi >> 48);
+    u64a reach7  = domain_mask & ((it_hi >> 56) | (it_lo << 8));
+    u64a reach8  = domain_mask & it_lo;
+    u64a reach9  = domain_mask & (it_lo >> 8);
+    u64a reach10 = domain_mask & (it_lo >> 16);
+    u64a reach11 = domain_mask & (it_lo >> 24);
+    u64a reach12 = domain_mask & (it_lo >> 32);
+    u64a reach13 = domain_mask & (it_lo >> 40);
+    u64a reach14 = domain_mask & (it_lo >> 48);
+    u64a reach15 = domain_mask & unaligned_load_u32(itPtr + 15);
+
+    m128 st0  = load_m128_from_u64a(ft + reach0);
+    m128 st1  = lshiftbyte_m128(load_m128_from_u64a(ft + reach1), 1);
+    m128 st2  = lshiftbyte_m128(load_m128_from_u64a(ft + reach2), 2);
+    m128 st3  = lshiftbyte_m128(load_m128_from_u64a(ft + reach3), 3);
+    m128 st4  = lshiftbyte_m128(load_m128_from_u64a(ft + reach4), 4);
+    m128 st5  = lshiftbyte_m128(load_m128_from_u64a(ft + reach5), 5);
+    m128 st6  = lshiftbyte_m128(load_m128_from_u64a(ft + reach6), 6);
+    m128 st7  = lshiftbyte_m128(load_m128_from_u64a(ft + reach7), 7);
     m128 st8  = load_m128_from_u64a(ft + reach8);
-    m128 st9  = load_m128_from_u64a(ft + reach9);
-    m128 st10 = load_m128_from_u64a(ft + reach10);
-    m128 st11 = load_m128_from_u64a(ft + reach11);
-    m128 st12 = load_m128_from_u64a(ft + reach12);
-    m128 st13 = load_m128_from_u64a(ft + reach13);
-    m128 st14 = load_m128_from_u64a(ft + reach14);
-    m128 st15 = load_m128_from_u64a(ft + reach15);
-
-    st1 = lshiftbyte_m128(st1, 1);
-    st2 = lshiftbyte_m128(st2, 2);
-    st3 = lshiftbyte_m128(st3, 3);
-    st4 = lshiftbyte_m128(st4, 4);
-    st5 = lshiftbyte_m128(st5, 5);
-    st6 = lshiftbyte_m128(st6, 6);
-    st7 = lshiftbyte_m128(st7, 7);
-    st9 = lshiftbyte_m128(st9, 1);
-    st10 = lshiftbyte_m128(st10, 2);
-    st11 = lshiftbyte_m128(st11, 3);
-    st12 = lshiftbyte_m128(st12, 4);
-    st13 = lshiftbyte_m128(st13, 5);
-    st14 = lshiftbyte_m128(st14, 6);
-    st15 = lshiftbyte_m128(st15, 7);
+    m128 st9  = lshiftbyte_m128(load_m128_from_u64a(ft + reach9), 1);
+    m128 st10 = lshiftbyte_m128(load_m128_from_u64a(ft + reach10), 2);
+    m128 st11 = lshiftbyte_m128(load_m128_from_u64a(ft + reach11), 3);
+    m128 st12 = lshiftbyte_m128(load_m128_from_u64a(ft + reach12), 4);
+    m128 st13 = lshiftbyte_m128(load_m128_from_u64a(ft + reach13), 5);
+    m128 st14 = lshiftbyte_m128(load_m128_from_u64a(ft + reach14), 6);
+    m128 st15 = lshiftbyte_m128(load_m128_from_u64a(ft + reach15), 7);
 
     st0 = or128(st0, st1);
     st2 = or128(st2, st3);