a few more fixes/tests

Author: kilograham

src/rp2_common/pico_double/double_aeabi_rp2040.S

@@ -650,17 +650,11 @@ regular_func double2fix_z
   // adjust
   adds r3, r2
   ble 2f // adjusted input is zero or dedornmal or < 1
-  lsrs r2, r3, #11
+  lsrs r3, r3, #11
   bne 3f // adjusted input is > infinite
 
-  // put updated exponent back in double and convert to int
-  movs r2, #1
-  lsls r2, #11
-  subs r2, #1
-  lsls r2, #20
-  bics r1, r2
-  lsls r3, #20
-  orrs r1, r3
+  lsls r2, r2, #20 // align exponent adjustment offset
+  adds r1, r1, r2  // we know adjustment is safe
   b double2int_z
 2:
   // result is zero

src/rp2_common/pico_double/double_conv_m33.S

@@ -259,11 +259,11 @@ regular_func double2int64
   beq 2f // input zero or denormal, means answer remains zero
   sub r3, #0x3ff
   cmp r3, #0
-  blt 1f // modified input zero or denormal, or less than 1.0
+  blt 1f // input is less than 1.0
   cmp r3, #52
   bge double2int64_z // modified input must be an integer or infinite
   adds r3, #12
-  lsls r2, r1, r3    // ip has remaining fractional mantissa bits of r1
+  lsls r2, r1, r3    // r2 has remaining fractional mantissa bits of r1
   bne 1f             // not integer as non zero fractional bits remain
   subs r3, #32
   bics r3, r3, r3, asr #31 // map negative shift to zero

src/rp2_common/pico_float/float_aeabi_dcp.S

@@ -160,19 +160,21 @@ regular_func float2fix_z
   ubfx r2, r0, #23, #8
   cbz r2, 2f // input is zero or denormal
   cmp r2, #0xff
-  beq 1f // infinte just pass thru
+  beq 3f // input infinite or nan
   adds r2, r1
-  asrs r3, r2, #8
-  beq 1f // exponent in range
-  ite pl
-  movpl r2, #0xff
-  movmi r2, #0
+  ble 2f // modified input is denormal so zero
+  cmp r2, #0xff
+  beq 3f // modified input is infinite
 1:
   bfi r0, r2, #23, #8
   b float2int_z_entry
 2:
   movs r0, #0
   bx lr
+3:
+  mvn r1, #0x80000000
+  add r0, r1, r0, lsr#31 @ so -Inf → 0x80000000, +Inf → 0x7fffffff
+  bx lr
 
 float_wrapper_section __aeabi_f2iz
 saving_func wrapper __aeabi_f2iz float2int_z
@@ -187,19 +189,20 @@ regular_func float2ufix_z
   ubfx r2, r0, #23, #8
   cbz r2, 2f // input is zero or denormal
   cmp r2, #0xff
-  beq 1f // infinte just pass thru
+  beq 3f // input infinite or nan
   adds r2, r1
-  asrs r3, r2, #8
-  beq 1f // exponent in range
-  ite pl
-  movpl r2, #0xff
-  movmi r2, #0
+  ble 2f // modified input is denormal so zero
+  cmp r2, #0xff
+  beq 3f // modified input is infinite
 1:
   bfi r0, r2, #23, #8
   b float2uint_z_entry
 2:
   movs r0, #0
   bx lr
+3:
+  mvn r0, r0, asr #31
+  bx lr
 
 float_wrapper_section __aeabi_f2uiz
 saving_func wrapper __aeabi_f2uiz float2uint_z float2uint
@@ -213,19 +216,21 @@ saving_func regular float2fix
   ubfx r2, r0, #23, #8
   cbz r2, 2f // input is zero or denormal
   cmp r2, #0xff
-  beq 1f // infinte just pass thru
+  beq 3f // input infinite or nan
   adds r2, r1
-  asrs r3, r2, #8
-  beq 1f // exponent in range
-  ite pl
-  movpl r2, #0xff
-  movmi r2, #0
+  ble 2f // modified input is denormal so zero
+  cmp r2, #0xff
+  beq 3f // modified input is infinite
 1:
   bfi r0, r2, #23, #8
   b float2int_entry
 2:
   movs r0, #0
   bx lr
+3:
+  mvn r1, #0x80000000
+  add r0, r1, r0, lsr#31 @ so -Inf → 0x80000000, +Inf → 0x7fffffff
+  bx lr
 
 float_section float2int
 // (not a real thing - kept because we use wrapper in saving_func)

test/pico_float_test/custom_float_funcs_test.c

@@ -110,6 +110,11 @@ int test() {
     // todo test correct rounding around maximum precision
     test_checkf(uint642float(UINT64_MAX), 18446744073709551615.0f, "uint642float7");
 
+    union {
+        uint32_t u;
+        float f;
+    } u32f;
+
 #if !(LIB_PICO_FLOAT_COMPILER || defined(__riscv))
     printf("fix2float\n");
     // todo test correct rounding around maximum precision
@@ -154,6 +159,10 @@ int test() {
     test_checki(float2fix(-3.25f, 2), -13, "float2fix9");
     test_checki(float2fix(-0.75f, 1), -2, "float2fix10");
     test_checki(float2fix(-3.0f, -1), -2, "float2fix11"); // not very useful
+    u32f.u = 0x7f012345;
+    test_checki(float2fix(u32f.f, 1), INT32_MAX, "float2fix12");
+    u32f.u = 0xff012345;
+    test_checki(float2fix(u32f.f, 1), INT32_MIN, "float2fix13");
 
     printf("float2ufix\n");
     test_checku(float2ufix(3.5f, 8), 0x380, "float2ufix1");
@@ -203,6 +212,10 @@ int test() {
     test_checki(float2fix_z(-3.25f, 2), -13, "float2fix_z9");
     test_checki(float2fix_z(-0.75f, 1), -1, "float2fix_z10");
     test_checki(float2fix_z(-3.0f, -1), -1, "float2fix_z11"); // not very useful
+    u32f.u = 0x7f012345;
+    test_checki(float2fix_z(u32f.f, 1), INT32_MAX, "float2fix_z12");
+    u32f.u = 0xff012345;
+    test_checki(float2fix_z(u32f.f, 1), INT32_MIN, "float2fix_z13");
 
     printf("float2ufix_z\n");
     test_checku(float2ufix_z(3.5f, 8), 0x380, "float2ufix_z1");
@@ -213,6 +226,10 @@ int test() {
     test_checku(float2ufix_z(3.24999f, 2), 12, "float2ufix_z6");
     test_checku(float2ufix_z(3.25f, 2), 13, "float2ufix_z7");
     test_checku(float2ufix_z(3.0f, -1), 1, "float2ufix_z8"); // not very useful
+    u32f.u = 0x7f012345;
+    test_checku(float2ufix_z(u32f.f, 1), UINT32_MAX, "float2fix_z9");
+    u32f.u = 0xff012345;
+    test_checku(float2ufix_z(u32f.f, 1), 0, "float2fix_z10");
 
     printf("float2fix64_z\n");
     test_checki64(float2fix64_z(3.5f, 8), 0x380, "float2fix64_z1");