[Mesa-dev] [PATCH 1/2] nir: aplit nir_shader_compiler_options::lower_ffma into 16/32/64 bits

[Samuel Pitoiset]

RADV doesn't have to lower 16-bits FMA.
Original patch from Rhys Perry.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset at gmail.com>
---
 src/amd/vulkan/radv_shader.c                           | 4 +++-
 src/broadcom/compiler/nir_to_vir.c                     | 4 +++-
 src/compiler/nir/nir.h                                 | 4 +++-
 src/compiler/nir/nir_opt_algebraic.py                  | 4 +++-
 src/gallium/drivers/nouveau/nvc0/nvc0_screen.c         | 4 +++-
 src/gallium/drivers/panfrost/midgard/midgard_compile.h | 3 ++-
 src/gallium/drivers/radeonsi/si_get.c                  | 4 +++-
 src/gallium/drivers/vc4/vc4_program.c                  | 4 +++-
 8 files changed, 23 insertions(+), 8 deletions(-)

diff --git a/src/amd/vulkan/radv_shader.c b/src/amd/vulkan/radv_shader.c
index 7c5447c5b56..63d4147460c 100644
--- a/src/amd/vulkan/radv_shader.c
+++ b/src/amd/vulkan/radv_shader.c
@@ -70,7 +70,9 @@ static const struct nir_shader_compiler_options nir_options = {
 	.lower_unpack_unorm_4x8 = true,
 	.lower_extract_byte = true,
 	.lower_extract_word = true,
-	.lower_ffma = true,
+	.lower_ffma16 = true,
+	.lower_ffma32 = true,
+	.lower_ffma64 = true,
 	.lower_fpow = true,
 	.lower_mul_2x32_64 = true,
 	.max_unroll_iterations = 32
diff --git a/src/broadcom/compiler/nir_to_vir.c b/src/broadcom/compiler/nir_to_vir.c
index 2c411b86ed1..c516bd4eb74 100644
--- a/src/broadcom/compiler/nir_to_vir.c
+++ b/src/broadcom/compiler/nir_to_vir.c
@@ -2330,7 +2330,9 @@ const nir_shader_compiler_options v3d_nir_options = {
         .lower_unpack_half_2x16 = true,
         .lower_fdiv = true,
         .lower_find_lsb = true,
-        .lower_ffma = true,
+        .lower_ffma16 = true,
+        .lower_ffma32 = true,
+        .lower_ffma64 = 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 ca9d5e6ffc8..84f7f0ec23b 100644
--- a/src/compiler/nir/nir.h
+++ b/src/compiler/nir/nir.h
@@ -2161,7 +2161,9 @@ typedef enum {
 
 typedef struct nir_shader_compiler_options {
    bool lower_fdiv;
-   bool lower_ffma;
+   bool lower_ffma16;
+   bool lower_ffma32;
+   bool lower_ffma64;
    bool fuse_ffma;
    bool lower_flrp16;
    bool lower_flrp32;
diff --git a/src/compiler/nir/nir_opt_algebraic.py b/src/compiler/nir/nir_opt_algebraic.py
index 2e5e43d801e..b1559133281 100644
--- a/src/compiler/nir/nir_opt_algebraic.py
+++ b/src/compiler/nir/nir_opt_algebraic.py
@@ -141,7 +141,9 @@ optimizations = [
    (('~fadd', a, ('fmul', ('b2f', 'c at 1'), ('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_ffma32'),
+   (('ffma at 64', a, b, c), ('fadd', ('fmul', a, b), c), 'options->lower_ffma64'),
    (('~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/nouveau/nvc0/nvc0_screen.c b/src/gallium/drivers/nouveau/nvc0/nvc0_screen.c
index afad48b5920..957ed5904cd 100644
--- a/src/gallium/drivers/nouveau/nvc0/nvc0_screen.c
+++ b/src/gallium/drivers/nouveau/nvc0/nvc0_screen.c
@@ -896,7 +896,9 @@ nvc0_screen_bind_cb_3d(struct nvc0_screen *screen, bool *can_serialize,
 
 static const nir_shader_compiler_options nir_options = {
    .lower_fdiv = false,
-   .lower_ffma = false,
+   .lower_ffma16 = false,
+   .lower_ffma32 = false,
+   .lower_ffma64 = false,
    .fuse_ffma = false, /* nir doesn't track mad vs fma */
    .lower_flrp32 = true,
    .lower_flrp64 = true,
diff --git a/src/gallium/drivers/panfrost/midgard/midgard_compile.h b/src/gallium/drivers/panfrost/midgard/midgard_compile.h
index b3308c866f6..d609cbf0ae7 100644
--- a/src/gallium/drivers/panfrost/midgard/midgard_compile.h
+++ b/src/gallium/drivers/panfrost/midgard/midgard_compile.h
@@ -62,8 +62,9 @@ midgard_compile_shader_nir(nir_shader *nir, midgard_program *program, bool is_bl
  * solution. */
 
 static const nir_shader_compiler_options midgard_nir_options = {
-        .lower_ffma = true,
         .lower_ffma16 = true,
+        .lower_ffma32 = true,
+        .lower_ffma64 = true,
         .lower_sub = true,
         .lower_scmp = true,
         .lower_flrp32 = true,
diff --git a/src/gallium/drivers/radeonsi/si_get.c b/src/gallium/drivers/radeonsi/si_get.c
index 6fa67087c7d..de0213995cb 100644
--- a/src/gallium/drivers/radeonsi/si_get.c
+++ b/src/gallium/drivers/radeonsi/si_get.c
@@ -490,7 +490,9 @@ static const struct nir_shader_compiler_options nir_options = {
 	.lower_fsat = true,
 	.lower_fdiv = true,
 	.lower_sub = true,
-	.lower_ffma = true,
+	.lower_ffma16 = true,
+	.lower_ffma32 = true,
+	.lower_ffma64 = 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 eb8b3a2c377..e1f166abe82 100644
--- a/src/gallium/drivers/vc4/vc4_program.c
+++ b/src/gallium/drivers/vc4/vc4_program.c
@@ -2233,7 +2233,9 @@ static const nir_shader_compiler_options nir_options = {
         .lower_extract_byte = true,
         .lower_extract_word = true,
         .lower_fdiv = true,
-        .lower_ffma = true,
+        .lower_ffma16 = true,
+        .lower_ffma32 = true,
+        .lower_ffma64 = true,
         .lower_flrp32 = true,
         .lower_fpow = true,
         .lower_fsat = true,
-- 
2.21.0