nir: add i2imp and u2ump opcodes for conversions to mediump

Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Reviewed-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5002>

diff --git a/src/compiler/nir/nir_opcodes.py b/src/compiler/nir/nir_opcodes.py
index 142d7a427d553a358f4500c86974206362235c8d..6a5d025f6366bbb97b1dcad02743ac2fcec86c8f 100644 (file)
--- a/src/compiler/nir/nir_opcodes.py
+++ b/src/compiler/nir/nir_opcodes.py
@@ -100,6 +100,7 @@ tbool16 = "bool16"
 tbool32 = "bool32"
 tuint = "uint"
 tuint8 = "uint8"
+tint16 = "int16"
 tuint16 = "uint16"
 tfloat16 = "float16"
 tfloat32 = "float32"
@@ -267,11 +268,13 @@ for src_t in [tint, tuint, tfloat, tbool]:
                                                        dst_bit_size),
                                    dst_t + str(dst_bit_size), src_t, conv_expr)
 
-# Special opcode that is the same as f2f16 except that it is safe to remove it
-# if the result is immediately converted back to float32 again. This is
-# generated as part of the precision lowering pass. mp stands for medium
+# Special opcode that is the same as f2f16, i2i16, u2u16 except that it is safe
+# to remove it if the result is immediately converted back to 32 bits again.
+# This is generated as part of the precision lowering pass. mp stands for medium
 # precision.
 unop_numeric_convert("f2fmp", tfloat16, tfloat, opcodes["f2f16"].const_expr)
+unop_numeric_convert("i2imp", tint16, tint, opcodes["i2i16"].const_expr)
+unop_numeric_convert("u2ump", tuint16, tuint, opcodes["u2u16"].const_expr)
 
 # Unary floating-point rounding operations.
 

diff --git a/src/compiler/nir/nir_opt_algebraic.py b/src/compiler/nir/nir_opt_algebraic.py
index 44b0ecc2c085503c581ee14c608e1d5c61aea4a7..a822813605b95e718fea9a39702cbdf89f213c61 100644 (file)
--- a/src/compiler/nir/nir_opt_algebraic.py
+++ b/src/compiler/nir/nir_opt_algebraic.py
@@ -896,14 +896,24 @@ optimizations.extend([
    (('~f2u32', ('i2f', 'a@32')), a),
    (('~f2u32', ('u2f', 'a@32')), a),
 
-   # Conversions from float16 to float32 and back can always be removed
+   # Conversions from 16 bits to 32 bits and back can always be removed
    (('f2f16', ('f2f32', 'a@16')), a),
    (('f2fmp', ('f2f32', 'a@16')), a),
+   (('i2i16', ('i2i32', 'a@16')), a),
+   (('i2imp', ('i2i32', 'a@16')), a),
+   (('u2u16', ('u2u32', 'a@16')), a),
+   (('u2ump', ('u2u32', 'a@16')), a),
    (('f2f16', ('b2f32', 'a@1')), ('b2f16', a)),
    (('f2fmp', ('b2f32', 'a@1')), ('b2f16', a)),
-   # Conversions to float16 would be lossy so they should only be removed if
+   (('i2i16', ('b2i32', 'a@1')), ('b2i16', a)),
+   (('i2imp', ('b2i32', 'a@1')), ('b2i16', a)),
+   (('u2u16', ('b2i32', 'a@1')), ('b2i16', a)),
+   (('u2ump', ('b2i32', 'a@1')), ('b2i16', a)),
+   # Conversions to 16 bits would be lossy so they should only be removed if
    # the instruction was generated by the precision lowering pass.
    (('f2f32', ('f2fmp', 'a@32')), a),
+   (('i2i32', ('i2imp', 'a@32')), a),
+   (('u2u32', ('u2ump', 'a@32')), a),
 
    (('ffloor', 'a(is_integral)'), a),
    (('fceil', 'a(is_integral)'), a),
@@ -1926,10 +1936,12 @@ late_optimizations = [
    (('~fadd', ('ffma(is_used_once)', a, b, ('fmul', 'c(is_not_const_and_not_fsign)', 'd(is_not_const_and_not_fsign)') ), 'e(is_not_const)'),
     ('ffma', a, b, ('ffma', c, d, e)), '(info->stage != MESA_SHADER_VERTEX && info->stage != MESA_SHADER_GEOMETRY) && !options->intel_vec4'),
 
-   # Convert f2fmp instructions to concrete f2f16 instructions. At this point
+   # Convert *2*mp instructions to concrete *2*16 instructions. At this point
    # any conversions that could have been removed will have been removed in
    # nir_opt_algebraic so any remaining ones are required.
    (('f2fmp', a), ('f2f16', a)),
+   (('i2imp', a), ('i2i16', a)),
+   (('u2ump', a), ('u2u16', a)),
 
    # Section 8.8 (Integer Functions) of the GLSL 4.60 spec says:
    #