[Mesa-dev] [PATCH 20/38] compiler/nir: add lowering option for 16-bit ffma

[Rhys Perry]

The lowering needs to be disabled for sufficient precision to pass
deqp-vk's 16-bit fma test on radv.

Signed-off-by: Rhys Perry <pendingchaos02 at gmail.com>
---
 src/broadcom/compiler/nir_to_vir.c    | 1 +
 src/compiler/nir/nir.h                | 1 +
 src/compiler/nir/nir_opt_algebraic.py | 4 +++-
 src/gallium/drivers/radeonsi/si_get.c | 1 +
 src/gallium/drivers/vc4/vc4_program.c | 1 +
 5 files changed, 7 insertions(+), 1 deletion(-)

diff --git a/src/broadcom/compiler/nir_to_vir.c b/src/broadcom/compiler/nir_to_vir.c
index 57be43d724..ec73ed269d 100644
--- a/src/broadcom/compiler/nir_to_vir.c
+++ b/src/broadcom/compiler/nir_to_vir.c
@@ -1975,6 +1975,7 @@ const nir_shader_compiler_options v3d_nir_options = {
         .lower_fdiv = true,
         .lower_find_lsb = true,
         .lower_ffma = true,
+        .lower_ffma16 = true,
         .lower_flrp32 = true,
         .lower_fpow = true,
         .lower_fsat = true,
diff --git a/src/compiler/nir/nir.h b/src/compiler/nir/nir.h
index e731653afc..0e40d6a97d 100644
--- a/src/compiler/nir/nir.h
+++ b/src/compiler/nir/nir.h
@@ -2072,6 +2072,7 @@ typedef struct nir_function {
 
 typedef struct nir_shader_compiler_options {
    bool lower_fdiv;
+   bool lower_ffma16;
    bool lower_ffma;
    bool fuse_ffma;
    bool lower_flrp16;
diff --git a/src/compiler/nir/nir_opt_algebraic.py b/src/compiler/nir/nir_opt_algebraic.py
index e3821320e8..c2a8b61d92 100644
--- a/src/compiler/nir/nir_opt_algebraic.py
+++ b/src/compiler/nir/nir_opt_algebraic.py
@@ -133,7 +133,9 @@ optimizations = [
    (('~fadd', a, ('fmul', ('b2f', c), ('fadd', b, ('fneg', a)))), ('bcsel', c, b, a), 'options->lower_flrp32'),
    (('~fadd at 32', a, ('fmul',         c , ('fadd', b, ('fneg', a)))), ('flrp', a, b, c), '!options->lower_flrp32'),
    (('~fadd at 64', a, ('fmul',         c , ('fadd', b, ('fneg', a)))), ('flrp', a, b, c), '!options->lower_flrp64'),
-   (('ffma', a, b, c), ('fadd', ('fmul', a, b), c), 'options->lower_ffma'),
+   (('ffma at 16', a, b, c), ('fadd', ('fmul', a, b), c), 'options->lower_ffma16'),
+   (('ffma at 32', a, b, c), ('fadd', ('fmul', a, b), c), 'options->lower_ffma'),
+   (('ffma at 64', a, b, c), ('fadd', ('fmul', a, b), c), 'options->lower_ffma'),
    (('~fadd', ('fmul', a, b), c), ('ffma', a, b, c), 'options->fuse_ffma'),
 
    (('fdot4', ('vec4', a, b,   c,   1.0), d), ('fdph',  ('vec3', a, b, c), d)),
diff --git a/src/gallium/drivers/radeonsi/si_get.c b/src/gallium/drivers/radeonsi/si_get.c
index 91f38329d5..d295821d65 100644
--- a/src/gallium/drivers/radeonsi/si_get.c
+++ b/src/gallium/drivers/radeonsi/si_get.c
@@ -497,6 +497,7 @@ static const struct nir_shader_compiler_options nir_options = {
 	.lower_fdiv = true,
 	.lower_sub = true,
 	.lower_ffma = true,
+	.lower_ffma16 = true,
 	.lower_pack_snorm_2x16 = true,
 	.lower_pack_snorm_4x8 = true,
 	.lower_pack_unorm_2x16 = true,
diff --git a/src/gallium/drivers/vc4/vc4_program.c b/src/gallium/drivers/vc4/vc4_program.c
index b98baca30c..9e8c6607cc 100644
--- a/src/gallium/drivers/vc4/vc4_program.c
+++ b/src/gallium/drivers/vc4/vc4_program.c
@@ -2234,6 +2234,7 @@ static const nir_shader_compiler_options nir_options = {
         .lower_extract_word = true,
         .lower_fdiv = true,
         .lower_ffma = true,
+        .lower_ffma16 = true,
         .lower_flrp32 = true,
         .lower_fpow = true,
         .lower_fsat = true,
-- 
2.19.2