From: Mika Lindqvist <postmaster@raasu.org>
Date: Tue, 22 Jun 2021 19:19:13 +0000 (+0300)
Subject: [PowerPC] Use templatized code for slide_hash as code for VMX and VSX is very similar
X-Git-Tag: 2.1.0-beta1~525
X-Git-Url: http://git.ipfire.org/cgi-bin/gitweb.cgi?a=commitdiff_plain;h=f456924eaa9f08a3557cced8ed40b7ef6f9d76ea;p=thirdparty%2Fzlib-ng.git

[PowerPC] Use templatized code for slide_hash as code for VMX and VSX is very similar
* Any differences can be handled using compiler options or added as macros before including template header
---

diff --git a/arch/power/slide_hash_power8.c b/arch/power/slide_hash_power8.c
index f15305727..5b078ec9f 100644
--- a/arch/power/slide_hash_power8.c
+++ b/arch/power/slide_hash_power8.c
@@ -6,49 +6,7 @@
 
 #ifdef POWER8_VSX_SLIDEHASH
 
-#include <altivec.h>
-#include "zbuild.h"
-#include "deflate.h"
-
-static inline void slide_hash_chain(Pos *table, uint32_t entries, uint16_t wsize) {
-    vector unsigned short vw, vm, *vp;
-    unsigned chunks;
-
-    table += entries;
-
-    /* Each vector register (chunk) corresponds to 128 bits == 8 Posf,
-     * so instead of processing each of the entries in the hash table
-     * individually, we can do it in chunks of 8 with vector instructions.
-     *
-     * This function is only called from slide_hash_power8(), and both calls
-     * pass entries as a power of 2 higher than 2^7, as defined by
-     * deflateInit2_(), so entries will always be a multiple of 8. */
-    chunks = entries >> 3;
-    Assert(entries % 8 == 0, "Weird hash table size!");
-
-    vw[0] = wsize;
-    vw = vec_splat(vw,0);
-
-    vp = (vector unsigned short *)table;
-
-    do {
-        /* Processing 8 elements at a time */
-        vp--;
-        vm = *vp;
-
-        /* This is equivalent to: m >= wsize ? m - wsize : 0
-         * Since we are using a saturated unsigned subtraction, any
-         * values that are <= wsize will be set to 0, while the others
-         * will be subtracted by wsize. */
-        *vp = vec_subs(vm,vw);
-    } while (--chunks);
-}
-
-void Z_INTERNAL slide_hash_power8(deflate_state *s) {
-    uint16_t wsize = s->w_size;
-
-    slide_hash_chain(s->head, HASH_SIZE, wsize);
-    slide_hash_chain(s->prev, wsize, wsize);
-}
+#define SLIDE_PPC slide_hash_power8
+#include "slide_ppc_tpl.h"
 
 #endif /* POWER8_VSX_SLIDEHASH */
diff --git a/arch/power/slide_hash_vmx.c b/arch/power/slide_hash_vmx.c
index b16df1538..cf9bd7b79 100644
--- a/arch/power/slide_hash_vmx.c
+++ b/arch/power/slide_hash_vmx.c
@@ -4,31 +4,7 @@
  */
 #ifdef PPC_VMX_SLIDEHASH
 
-#include <altivec.h>
-#include "zbuild.h"
-#include "deflate.h"
-
-static inline void slide_hash_chain(Pos *table, uint32_t entries, uint16_t wsize) {
-    const vector unsigned short vmx_wsize = vec_splats(wsize);
-    Pos *p = table;
-
-    do {
-        vector unsigned short value, result;
-
-        value = vec_ld(0, p);
-        result = vec_subs(value, vmx_wsize);
-        vec_st(result, 0, p);
-
-        p += 8;
-        entries -= 8;
-   } while (entries > 0);
-}
-
-void Z_INTERNAL slide_hash_vmx(deflate_state *s) {
-    uint16_t wsize = s->w_size;
-
-    slide_hash_chain(s->head, HASH_SIZE, wsize);
-    slide_hash_chain(s->prev, wsize, wsize);
-}
+#define SLIDE_PPC slide_hash_vmx
+#include "slide_ppc_tpl.h"
 
 #endif /* PPC_VMX_SLIDEHASH */
diff --git a/arch/power/slide_ppc_tpl.h b/arch/power/slide_ppc_tpl.h
new file mode 100644
index 000000000..5c17e38fb
--- /dev/null
+++ b/arch/power/slide_ppc_tpl.h
@@ -0,0 +1,31 @@
+/* Optimized slide_hash for PowerPC processors
+ * Copyright (C) 2017-2021 Mika T. Lindqvist <postmaster@raasu.org>
+ * For conditions of distribution and use, see copyright notice in zlib.h
+ */
+
+#include <altivec.h>
+#include "zbuild.h"
+#include "deflate.h"
+
+static inline void slide_hash_chain(Pos *table, uint32_t entries, uint16_t wsize) {
+    const vector unsigned short vmx_wsize = vec_splats(wsize);
+    Pos *p = table;
+
+    do {
+        vector unsigned short value, result;
+
+        value = vec_ld(0, p);
+        result = vec_subs(value, vmx_wsize);
+        vec_st(result, 0, p);
+
+        p += 8;
+        entries -= 8;
+   } while (entries > 0);
+}
+
+void Z_INTERNAL SLIDE_PPC(deflate_state *s) {
+    uint16_t wsize = s->w_size;
+
+    slide_hash_chain(s->head, HASH_SIZE, wsize);
+    slide_hash_chain(s->prev, wsize, wsize);
+}