From 00ec13cb6e6b0cb9cba2a0d8b29f5d21e40d6c02 Mon Sep 17 00:00:00 2001
From: Theodore Dubois <tblodt@icloud.com>
Date: Tue, 30 Oct 2018 14:41:39 -0700
Subject: [PATCH] Implement log base 2 (fyl2x)

---
 emu/decode.h     |  1 +
 emu/float80.c    | 36 ++++++++++++++++++++++++++++++++++++
 emu/float80.h    |  2 ++
 emu/fpu.c        |  5 +++++
 emu/fpu.h        |  1 +
 emu/interp/fpu.h |  1 +
 jit/gen.c        |  1 +
 7 files changed, 47 insertions(+)

diff --git a/emu/decode.h b/emu/decode.h
index e9593f8f4d..cbfa084d91 100644
--- a/emu/decode.h
+++ b/emu/decode.h
@@ -608,6 +608,7 @@ __no_instrument DECODER_RET glue(DECODER_NAME, OP_SIZE)(DECODER_ARGS) {
                     case 0xd950: TRACE("fld1"); FLDC(one); break;
                     case 0xd955: TRACE("fldln2"); FLDC(ln2); break;
                     case 0xd956: TRACE("fldz"); FLDC(zero); break;
+                    case 0xd961: TRACE("fyl2x"); FYL2X(); break;
                     case 0xd970: TRACE("fprem"); FPREM(); break;
                     case 0xd974: TRACE("frndint"); FRNDINT(); break;
                     case 0xdf40: TRACE("fnstsw ax"); FSTSW(reg_a); break;
diff --git a/emu/float80.c b/emu/float80.c
index fea6668b90..587ccc0380 100644
--- a/emu/float80.c
+++ b/emu/float80.c
@@ -417,3 +417,39 @@ bool f80_eq(float80 a, float80 b) {
     if (f80_iszero(a)) b.sign = 0;
     return a.sign == b.sign && a.exp == b.exp && a.signif == b.signif;
 }
+
+bool f80_lte(float80 a, float80 b) {
+    return f80_lt(a, b) || f80_eq(a, b);
+}
+bool f80_gt(float80 a, float80 b) {
+    return !f80_lte(a, b);
+}
+
+float80 f80_log2(float80 x) {
+    float80 zero = f80_from_int(0);
+    float80 one = f80_from_int(1);
+    float80 two = f80_from_int(2);
+
+    int ipart = 0;
+    while (f80_lt(x, one)) {
+        ipart--;
+        x = f80_mul(x, two);
+    }
+    while (f80_gt(x, two)) {
+        ipart++;
+        x = f80_div(x, two);
+    }
+    float80 res = f80_from_int(ipart);
+
+    float80 bit = one;
+    while (f80_gt(bit, zero)) {
+        while (f80_lte(x, two) && f80_gt(bit, zero)) {
+            x = f80_mul(x, x);
+            bit = f80_div(bit, two);
+        }
+        res = f80_add(res, bit);
+        x = f80_div(x, two);
+    }
+    return res;
+}
+
diff --git a/emu/float80.h b/emu/float80.h
index 9c209b9ab8..c9152e3e2d 100644
--- a/emu/float80.h
+++ b/emu/float80.h
@@ -46,4 +46,6 @@ extern __thread enum f80_rounding_mode f80_rounding_mode;
 #define F80_NAN ((float80) {.signif = 0xc000000000000000, .exp = 0x7fff, .sign = 0})
 #define F80_INF ((float80) {.signif = 0x8000000000000000, .exp = 0x7fff, .sign = 0})
 
+float80 f80_log2(float80 x);
+
 #endif
diff --git a/emu/fpu.c b/emu/fpu.c
index b51f4c53e0..7124569ec4 100644
--- a/emu/fpu.c
+++ b/emu/fpu.c
@@ -86,6 +86,11 @@ void fpu_rndint(struct cpu_state *cpu) {
     ST(0) = f80_from_int(f80_to_int(ST(0)));
 }
 
+void fpu_yl2x(struct cpu_state *cpu) {
+    ST(1) = f80_mul(ST(1), f80_log2(ST(0)));
+    fpu_pop(cpu);
+}
+
 void fpu_ucom(struct cpu_state *cpu, int i) {
     cpu->c1 = 0;
     cpu->c0 = f80_lt(ST(0), ST(i));
diff --git a/emu/fpu.h b/emu/fpu.h
index 26de0126a3..6d6cd05c0c 100644
--- a/emu/fpu.h
+++ b/emu/fpu.h
@@ -38,6 +38,7 @@ void fpu_ldm80(struct cpu_state *cpu, float80 *f);
 
 void fpu_prem(struct cpu_state *cpu);
 void fpu_rndint(struct cpu_state *cpu);
+void fpu_yl2x(struct cpu_state *cpu);
 void fpu_ucom(struct cpu_state *cpu, int i);
 void fpu_abs(struct cpu_state *cpu);
 void fpu_chs(struct cpu_state *cpu);
diff --git a/emu/interp/fpu.h b/emu/interp/fpu.h
index 51de9bcb34..530e6b0129 100644
--- a/emu/interp/fpu.h
+++ b/emu/interp/fpu.h
@@ -78,6 +78,7 @@
     ST(0) = f80_mod(ST(0), ST(1))
 
 #define FRNDINT() UNDEFINED
+#define FYL2X() UNDEFINED
 
 #define FUCOMI() \
     cpu->zf = f80_eq(ST(0), ST_i); \
diff --git a/jit/gen.c b/jit/gen.c
index 9437db128e..70473f1c5b 100644
--- a/jit/gen.c
+++ b/jit/gen.c
@@ -358,6 +358,7 @@ static inline bool gen_op(struct gen_state *state, gadget_t *gadgets, enum arg a
 #define FLDC(what) hh(fpu_ldc, fconst_##what)
 #define FPREM() h(fpu_prem)
 #define FRNDINT() h(fpu_rndint)
+#define FYL2X() h(fpu_yl2x)
 #define FSTSW(dst) if (arg_##dst == arg_reg_a) g(fstsw_ax); else UNDEFINED
 #define FSTCW(dst) if (arg_##dst == arg_reg_a) UNDEFINED; else h_write(fpu_stcw, 16)
 #define FLDCW(dst) if (arg_##dst == arg_reg_a) UNDEFINED; else h_read(fpu_ldcw, 16)