[PATCH 0/2] MR740: Implement hyperbolic trigonometry operations.

[Petrichor Park (＠petrathekat)]

I'm working on adding all the intrinsics we haven't implemented yet. Here's sinh, cosh, and tanh.

Sinh/cosh are implemented in the same commit because they forward to the same backing function (because the identities used only differ by a plus or minus).

[PATCH 1/2] vkd3d-shader/hlsl: Implement hyperbolic sin and cos.

[Petrichor Park]

From: Petrichor Park

Also tightens the TODOs on some of the tests.
---
 libs/vkd3d-shader/hlsl.y            | 55 +++++++++++++++++++++++++++++
 tests/hlsl/trigonometry.shader_test | 16 ++++-----
 2 files changed, 63 insertions(+), 8 deletions(-)

diff --git a/libs/vkd3d-shader/hlsl.y b/libs/vkd3d-shader/hlsl.y
index ec8b3d22a..e4a5bf1f3 100644
--- a/libs/vkd3d-shader/hlsl.y
+++ b/libs/vkd3d-shader/hlsl.y
@@ -3022,6 +3022,53 @@ static bool intrinsic_cos(struct hlsl_ctx *ctx,
     return !!add_unary_arithmetic_expr(ctx, params->instrs, HLSL_OP1_COS, arg, loc);
 }

+static bool write_cosh_or_sinh(struct hlsl_ctx *ctx,
+        const struct parse_initializer *params, const struct vkd3d_shader_location *loc, bool sinh_mode)
+{
+    struct hlsl_ir_function_decl *func;
+    struct hlsl_type *type;
+    const char *fn_name, *combiner;
+    char *body;
+
+    static const char template[] =
+            "%s %s(%s x)\n"
+            "{\n"
+            "    %s expPos, expNeg, combined;"
+            "    expPos = exp(x);\n"
+            "    expNeg = exp(-x);\n"
+            "    combined = %s;\n"
+            "    return combined / 2;\n"
+            "}\n";
+    static const char fn_name_sinh[] = "sinh";
+    static const char fn_name_cosh[] = "cosh";
+    static const char combiner_sinh[] = "expPos - expNeg";
+    static const char combiner_cosh[] = "expPos + expNeg";
+
+    type = params->args[0]->data_type;
+    type = hlsl_get_numeric_type(ctx, type->class, HLSL_TYPE_FLOAT, type->dimx, type->dimy);
+
+    fn_name = sinh_mode ? fn_name_sinh : fn_name_cosh;
+    combiner = sinh_mode ? combiner_sinh : combiner_cosh;
+
+    if (!(body = hlsl_sprintf_alloc(ctx, template,
+            type->name, fn_name, type->name,
+            type->name, combiner)))
+        return false;           
+
+    func = hlsl_compile_internal_function(ctx, fn_name, body);
+    vkd3d_free(body);
+    if (!func)
+        return false;
+
+    return add_user_call(ctx, func, params, loc);
+}
+
+static bool intrinsic_cosh(struct hlsl_ctx *ctx,
+        const struct parse_initializer *params, const struct vkd3d_shader_location *loc)
+{
+    return write_cosh_or_sinh(ctx, params, loc, false);
+}
+
 static bool intrinsic_cross(struct hlsl_ctx *ctx,
         const struct parse_initializer *params, const struct vkd3d_shader_location *loc)
 {
@@ -3814,6 +3861,12 @@ static bool intrinsic_sin(struct hlsl_ctx *ctx,
     return !!add_unary_arithmetic_expr(ctx, params->instrs, HLSL_OP1_SIN, arg, loc);
 }

+static bool intrinsic_sinh(struct hlsl_ctx *ctx,
+        const struct parse_initializer *params, const struct vkd3d_shader_location *loc)
+{
+    return write_cosh_or_sinh(ctx, params, loc, true);
+}
+
 /* smoothstep(a, b, x) = p^2 (3 - 2p), where p = saturate((x - a)/(b - a)) */
 static bool intrinsic_smoothstep(struct hlsl_ctx *ctx,
         const struct parse_initializer *params, const struct vkd3d_shader_location *loc)
@@ -4218,6 +4271,7 @@ intrinsic_functions[] =
     {"clamp",                               3, true,  intrinsic_clamp},
     {"clip",                                1, true,  intrinsic_clip},
     {"cos",                                 1, true,  intrinsic_cos},
+    {"cosh",                                1, true,  intrinsic_cosh},
     {"cross",                               2, true,  intrinsic_cross},
     {"ddx",                                 1, true,  intrinsic_ddx},
     {"ddx_coarse",                          1, true,  intrinsic_ddx_coarse},
@@ -4254,6 +4308,7 @@ intrinsic_functions[] =
     {"saturate",                            1, true,  intrinsic_saturate},
     {"sign",                                1, true,  intrinsic_sign},
     {"sin",                                 1, true,  intrinsic_sin},
+    {"sinh",                                1, true,  intrinsic_sinh},
     {"smoothstep",                          3, true,  intrinsic_smoothstep},
     {"sqrt",                                1, true,  intrinsic_sqrt},
     {"step",                                2, true,  intrinsic_step},
diff --git a/tests/hlsl/trigonometry.shader_test b/tests/hlsl/trigonometry.shader_test
index 3f7f11685..392728098 100644
--- a/tests/hlsl/trigonometry.shader_test
+++ b/tests/hlsl/trigonometry.shader_test
@@ -76,7 +76,7 @@ todo(sm<4 | glsl) draw quad
 probe all rgba (0, 1000, -1000.0, 0)

-[pixel shader todo]
+[pixel shader]
 uniform float4 a;

float4 main() : sv_target
@@ -86,15 +86,15 @@ float4 main() : sv_target

[test]
 uniform 0 float4 -6.28318531 -0.88137359 0.88137359 6.28318531
-todo(sm<6) draw quad
+todo(glsl) draw quad
 probe all rgba (-267.744894, -1.0, 1.0, 267.744894) 2
 uniform 0 float4 -0.0 0.0 -90.0 90.0
-todo(sm<6) draw quad
+todo(glsl) draw quad
 % mingw does not support "inf" for scanf(), but numbers beyond FLOAT_MAX consistently result in inf.
 probe all rgba (0.0, 0.0, -1.0e39, 1.0e39) 1

-[pixel shader todo]
+[pixel shader]
 uniform float4 a;

float4 main() : sv_target
@@ -104,10 +104,10 @@ float4 main() : sv_target

[test]
 uniform 0 float4 -1.76274717 -1.3169579 1.3169579 1.76274717
-todo(sm<6) draw quad
+todo(glsl) draw quad
 probe all rgba (3.0, 2.0, 2.0, 3.0) 2
 uniform 0 float4 -0.0 0.0 -90.0 90.0
-todo(sm<6) draw quad
+todo(glsl) draw quad
 probe all rgba (1.0, 1.0, 1.0e39, 1.0e39) 1

@@ -121,8 +121,8 @@ float4 main() : sv_target

[test]
 uniform 0 float4 -1.57079633 -0.54930614 0.54930614 1.57079633
-todo(sm<6) draw quad
+todo draw quad
 probe all rgba (-0.91715234, -0.5, 0.5, 0.91715234) 2
 uniform 0 float4 -10.0 -0.0 0.0 10.0
-todo(sm<6) draw quad
+todo draw quad
 probe all rgba (-1.0, 0.0, 0.0, 1.0) 1

[PATCH 2/2] vkd3d-shader/hlsl: Implement tanh.

[Petrichor Park]

From: Petrichor Park

---
 libs/vkd3d-shader/hlsl.y            | 32 +++++++++++++++++++++++++++++
 tests/hlsl/trigonometry.shader_test |  6 +++---
 2 files changed, 35 insertions(+), 3 deletions(-)

diff --git a/libs/vkd3d-shader/hlsl.y b/libs/vkd3d-shader/hlsl.y
index e4a5bf1f3..a5ea27ad3 100644
--- a/libs/vkd3d-shader/hlsl.y
+++ b/libs/vkd3d-shader/hlsl.y
@@ -3939,6 +3939,37 @@ static bool intrinsic_tan(struct hlsl_ctx *ctx,
     return !!add_binary_arithmetic_expr(ctx, params->instrs, HLSL_OP2_DIV, sin, cos, loc);
 }

+static bool intrinsic_tanh(struct hlsl_ctx *ctx,
+        const struct parse_initializer *params, const struct vkd3d_shader_location *loc)
+{
+    struct hlsl_ir_function_decl *func;
+    struct hlsl_type *type;
+    char *body;
+
+    static const char template[] =
+            "%s tanh(%s x)\n"
+            "{\n"
+            "    %s expPos, expNeg;\n"
+            "    expPos = exp(x);\n"
+            "    expNeg = exp(-x);\n"
+            "    return (expPos - expNeg) / (expPos + expNeg);\n"
+            "}\n";
+
+    type = params->args[0]->data_type;
+    type = hlsl_get_numeric_type(ctx, type->class, HLSL_TYPE_FLOAT, type->dimx, type->dimy);
+
+    if (!(body = hlsl_sprintf_alloc(ctx, template,
+            type->name, type->name, type->name)))
+        return false;           
+
+    func = hlsl_compile_internal_function(ctx, "tanh", body);
+    vkd3d_free(body);
+    if (!func)
+        return false;
+
+    return add_user_call(ctx, func, params, loc);
+}
+
 static bool intrinsic_tex(struct hlsl_ctx *ctx, const struct parse_initializer *params,
         const struct vkd3d_shader_location *loc, const char *name, enum hlsl_sampler_dim dim)
 {
@@ -4313,6 +4344,7 @@ intrinsic_functions[] =
     {"sqrt",                                1, true,  intrinsic_sqrt},
     {"step",                                2, true,  intrinsic_step},
     {"tan",                                 1, true,  intrinsic_tan},
+    {"tanh",                                1, true,  intrinsic_tanh},
     {"tex1D",                              -1, false, intrinsic_tex1D},
     {"tex2D",                              -1, false, intrinsic_tex2D},
     {"tex2Dlod",                            2, false, intrinsic_tex2Dlod},
diff --git a/tests/hlsl/trigonometry.shader_test b/tests/hlsl/trigonometry.shader_test
index 392728098..cf6165876 100644
--- a/tests/hlsl/trigonometry.shader_test
+++ b/tests/hlsl/trigonometry.shader_test
@@ -111,7 +111,7 @@ todo(glsl) draw quad
 probe all rgba (1.0, 1.0, 1.0e39, 1.0e39) 1

-[pixel shader todo]
+[pixel shader]
 uniform float4 a;

float4 main() : sv_target
@@ -121,8 +121,8 @@ float4 main() : sv_target

[test]
 uniform 0 float4 -1.57079633 -0.54930614 0.54930614 1.57079633
-todo draw quad
+todo(glsl) draw quad
 probe all rgba (-0.91715234, -0.5, 0.5, 0.91715234) 2
 uniform 0 float4 -10.0 -0.0 0.0 10.0
-todo draw quad
+todo(glsl) draw quad
 probe all rgba (-1.0, 0.0, 0.0, 1.0) 1

[Zebediah Figura (＠zfigura)]

Zebediah Figura (@zfigura) commented about libs/vkd3d-shader/hlsl.y:
> +static bool write_cosh_or_sinh(struct hlsl_ctx *ctx,
> -
>    const struct parse_initializer *params, const struct vkd3d_shader_location *loc, bool sinh_mode)
>
> +{
> - struct hlsl_ir_function_decl *func;
> - struct hlsl_type *type;
> - const char *fn_name, *combiner;
> - char *body;
> -
> - static const char template[] =
> -
>        "%s %s(%s x)\n"
> -
>        "{\n"
> -
>        "    %s expPos, expNeg, combined;"
> -
>        "    expPos = exp(x);\n"
> -
>        "    expNeg = exp(-x);\n"
> -
>        "    combined = %s;\n"
> -
>        "    return combined / 2;\n"

Let's avoid camel case please. That said, though, the function seems simple enough that I'd just write "return (exp(x) ± exp(-x)) / 2"?

[Zebediah Figura (＠zfigura)]

Zebediah Figura (@zfigura) commented about libs/vkd3d-shader/hlsl.y:
> - static const char template[] =
> -
>        "%s %s(%s x)\n"
> -
>        "{\n"
> -
>        "    %s expPos, expNeg, combined;"
> -
>        "    expPos = exp(x);\n"
> -
>        "    expNeg = exp(-x);\n"
> -
>        "    combined = %s;\n"
> -
>        "    return combined / 2;\n"
> -
>        "}\n";
> - static const char fn_name_sinh[] = "sinh";
> - static const char fn_name_cosh[] = "cosh";
> - static const char combiner_sinh[] = "expPos - expNeg";
> - static const char combiner_cosh[] = "expPos + expNeg";
> -
> - type = params->args[0]->data_type;
> - type = hlsl_get_numeric_type(ctx, type->class, HLSL_TYPE_FLOAT, type->dimx, type->dimy);

We probably want intrinsic_float_convert_arg() instead.

[Giovanni Mascellani (＠giomasce)]

Giovanni Mascellani (@giomasce) commented about tests/hlsl/trigonometry.shader_test:
> [test]
>  uniform 0 float4 -1.57079633 -0.54930614 0.54930614 1.57079633
> -todo(sm<6) draw quad
> +todo draw quad
>  probe all rgba (-0.91715234, -0.5, 0.5, 0.91715234) 2
>  uniform 0 float4 -10.0 -0.0 0.0 10.0
> -todo(sm<6) draw quad
> +todo draw quad

These two changes fail (i.e., unexpectedly succeed) on the CI and on every implementation I've got. Is it possible it's a mistake? Which hardware are you testing on?

[Giovanni Mascellani (＠giomasce)]

> Also tightens the TODOs on some of the tests.

I don't think there is need to specify this, every time you add a feature you tighten the corresponding TODOs.

[Petrichor Park (＠petrathekat)]

On Thu Mar 28 11:39:57 2024 +0000, Giovanni Mascellani wrote:
> These two changes fail (i.e., unexpectedly succeed) on the CI and on
> every implementation I've got. Is it possible it's a mistake? Which
> hardware are you testing on?

I'm testing on KDE neon 6.0 on a ThinkPad E15 Gen 4. It does pass on my machine, except for the native implementation. Francisco and I were talking about how the local native failures are probably false negatives.