vulkan: Make sure workgroup size is defined before being referenced

Author: remyoudompheng

ggml/src/ggml-vulkan/vulkan-shaders/copy_from_quant.comp

@@ -1,16 +1,16 @@
 #version 450
 
-#include "types.comp"
-#include "generic_unary_head.comp"
-#include "dequant_funcs.comp"
-
 #if defined(DATA_A_IQ4_NL)
 // 16 invocations needed for init_iq4nl_shmem
 layout(local_size_x = 16, local_size_y = 1, local_size_z = 1) in;
 #else
 layout(local_size_x = 1, local_size_y = 1, local_size_z = 1) in;
 #endif
 
+#include "types.comp"
+#include "generic_unary_head.comp"
+#include "dequant_funcs.comp"
+
 void main() {
 #if defined(DATA_A_IQ2_XXS) || defined(DATA_A_IQ2_XS) || defined(DATA_A_IQ2_S) || defined(DATA_A_IQ3_XXS) || defined(DATA_A_IQ3_S) || defined(DATA_A_IQ4_NL)
     init_iq_shmem();

ggml/src/ggml-vulkan/vulkan-shaders/copy_to_quant.comp

@@ -1,15 +1,15 @@
 #version 450
 
-#include "types.comp"
-#include "generic_unary_head.comp"
-
 #if defined(DATA_A_IQ4_NL)
 // 16 invocations needed for init_iq4nl_shmem
 layout(local_size_x = 16, local_size_y = 1, local_size_z = 1) in;
 #else
 layout(local_size_x = 1, local_size_y = 1, local_size_z = 1) in;
 #endif
 
+#include "types.comp"
+#include "generic_unary_head.comp"
+
 layout (binding = 0) readonly buffer S {float data_s[];};
 layout (binding = 1) writeonly buffer Q {A_TYPE data_q[];};
 

ggml/src/ggml-vulkan/vulkan-shaders/dequant_iq2_s.comp

@@ -1,9 +1,9 @@
 #version 450
 
-#include "dequant_head.comp"
-
 layout(local_size_x = 256, local_size_y = 1, local_size_z = 1) in;
 
+#include "dequant_head.comp"
+
 layout (binding = 0) readonly buffer A {block_iq2_s data_a[];};
 layout (binding = 1) writeonly buffer D {D_TYPE data_b[];};
 

ggml/src/ggml-vulkan/vulkan-shaders/dequant_iq2_xs.comp

@@ -1,9 +1,9 @@
 #version 450
 
-#include "dequant_head.comp"
-
 layout(local_size_x = 256, local_size_y = 1, local_size_z = 1) in;
 
+#include "dequant_head.comp"
+
 layout (binding = 0) readonly buffer A {block_iq2_xs data_a[];};
 layout (binding = 1) writeonly buffer D {D_TYPE data_b[];};
 

ggml/src/ggml-vulkan/vulkan-shaders/dequant_iq2_xxs.comp

@@ -1,9 +1,9 @@
 #version 450
 
-#include "dequant_head.comp"
-
 layout(local_size_x = 256, local_size_y = 1, local_size_z = 1) in;
 
+#include "dequant_head.comp"
+
 layout (binding = 0) readonly buffer A {block_iq2_xxs data_a[];};
 layout (binding = 1) writeonly buffer D {D_TYPE data_b[];};
 

ggml/src/ggml-vulkan/vulkan-shaders/dequant_iq3_s.comp

@@ -1,9 +1,9 @@
 #version 450
 
-#include "dequant_head.comp"
-
 layout(local_size_x = 256, local_size_y = 1, local_size_z = 1) in;
 
+#include "dequant_head.comp"
+
 layout (binding = 0) readonly buffer A {block_iq3_s data_a[];};
 layout (binding = 1) writeonly buffer D {D_TYPE data_b[];};
 

ggml/src/ggml-vulkan/vulkan-shaders/dequant_iq3_xxs.comp

@@ -1,9 +1,9 @@
 #version 450
 
-#include "dequant_head.comp"
-
 layout(local_size_x = 256, local_size_y = 1, local_size_z = 1) in;
 
+#include "dequant_head.comp"
+
 layout (binding = 0) readonly buffer A {block_iq3_xxs data_a[];};
 layout (binding = 1) writeonly buffer D {D_TYPE data_b[];};
 

ggml/src/ggml-vulkan/vulkan-shaders/flash_attn_cm2.comp

@@ -16,11 +16,11 @@
 #extension GL_KHR_shader_subgroup_vote : enable
 #extension GL_EXT_null_initializer : enable
 
+layout(local_size_x_id = 0, local_size_y = 1, local_size_z = 1) in;
+
 #include "types.comp"
 #include "dequant_funcs_cm2.comp"
 
-layout(local_size_x_id = 0, local_size_y = 1, local_size_z = 1) in;
-
 layout (constant_id = 1) const uint32_t Br = 32;
 layout (constant_id = 2) const uint32_t Bc = 32;
 layout (constant_id = 3) const uint32_t D = 32;

ggml/src/ggml-vulkan/vulkan-shaders/get_rows_quant.comp

@@ -1,11 +1,11 @@
 #version 450
 
+layout(local_size_x = 512, local_size_y = 1, local_size_z = 1) in;
+
 #include "types.comp"
 #include "generic_binary_head.comp"
 #include "dequant_funcs.comp"
 
-layout(local_size_x = 512, local_size_y = 1, local_size_z = 1) in;
-
 void main() {
     const uint i00 = (gl_GlobalInvocationID.x)*2;
     const uint i10 = gl_GlobalInvocationID.y;

ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec.comp

@@ -2,10 +2,10 @@
 
 #extension GL_EXT_shader_explicit_arithmetic_types_int32 : require
 
-#include "mul_mat_vec_base.comp"
-
 layout(local_size_x_id = 0, local_size_y = 1, local_size_z = 1) in;
 
+#include "mul_mat_vec_base.comp"
+
 #if !defined(DATA_A_F32) && !defined(DATA_A_F16)
 #define K_PER_ITER 8
 #else

ggml/src/ggml-vulkan/vulkan-shaders/mul_mm.comp

@@ -17,6 +17,8 @@
 #extension GL_EXT_shader_explicit_arithmetic_types_int16 : require
 #endif
 
+layout(local_size_x_id = 0, local_size_y = 1, local_size_z = 1) in;
+
 #include "types.comp"
 
 #ifndef LOAD_VEC_A
@@ -26,8 +28,6 @@
 #define LOAD_VEC_B 1
 #endif
 
-layout(local_size_x_id = 0, local_size_y = 1, local_size_z = 1) in;
-
 layout (binding = 0) readonly buffer A {A_TYPE data_a[];};
 layout (binding = 1) readonly buffer B {B_TYPE data_b[];};
 layout (binding = 2) writeonly buffer D {D_TYPE data_d[];};

ggml/src/ggml-vulkan/vulkan-shaders/mul_mm_cm2.comp

@@ -15,10 +15,10 @@
 #extension GL_KHR_shader_subgroup_ballot : enable
 #extension GL_KHR_shader_subgroup_vote : enable
 
-#include "types.comp"
-
 layout(local_size_x_id = 0, local_size_y = 1, local_size_z = 1) in;
 
+#include "types.comp"
+
 layout (constant_id = 1) const uint BM = 64;
 layout (constant_id = 2) const uint BN = 64;
 layout (constant_id = 3) const uint BK = 16;  // Assumed to be 32 if working with a quant